fix(neuron): as_deref for Arc'd prefix_cache in cuda-only worker path (#98)

candle.rs:2441 sits behind cfg(cuda) — the CPU build can't see it, and
the CUDA type-check caught the Option<&Arc<Mutex<…>>> vs
Option<&Mutex<…>> mismatch introduced by the LoadedModel Arc change.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TczcGF7JSjJs8r15RSSGpx

.gitea/workflows/build-prerelease.yml

@@ -56,6 +56,7 @@ env:
 jobs:
   prepare:
     name: Resolve version stamps + change detection
+    timeout-minutes: 10
     runs-on: rust
     outputs:
       version: ${{ steps.info.outputs.version }}
@@ -65,6 +66,7 @@ jobs:
       build_cortex: ${{ steps.changes.outputs.build_cortex }}
       build_neuron: ${{ steps.changes.outputs.build_neuron }}
       build_bench: ${{ steps.changes.outputs.build_bench }}
+      build_upstream: ${{ steps.changes.outputs.build_upstream }}
       check_rust: ${{ steps.changes.outputs.check_rust }}
     steps:
       - uses: actions/checkout@v4
@@ -85,7 +87,14 @@ jobs:
           # rpmvercmp ranks digit-prefixed segments above alpha ones.
           # The SHA stays only as a debug identifier; sort order is
           # decided entirely by the timestamp.
-          COMMIT_TIMESTAMP=$(git log -1 --format=%cd --date=format:%Y%m%d%H%M%S HEAD)
+          # format-local + TZ=UTC0 renders the stamp in UTC regardless
+          # of the committer's recorded timezone. Plain `format:` uses
+          # each commit's own TZ offset — local commits (UTC+3) and
+          # Gitea server-side merge commits (UTC) interleaved
+          # non-monotonically, letting an older build out-rank newer
+          # ones in RPM EVR comparison (the 2026-07-01 "235959" stamp
+          # that froze the fleet on a stale build).
+          COMMIT_TIMESTAMP=$(TZ=UTC0 git log -1 --format=%cd --date=format-local:%Y%m%d%H%M%S HEAD)
           RELEASE="0.1.${COMMIT_TIMESTAMP}.git${SHORT_SHA}"
           echo "version=${VERSION}" >> "$GITHUB_OUTPUT"
           echo "release=${RELEASE}" >> "$GITHUB_OUTPUT"
@@ -103,6 +112,7 @@ jobs:
           BUILD_CORTEX=true
           BUILD_NEURON=true
           BUILD_BENCH=true
+          BUILD_UPSTREAM=true
           CHECK_RUST=true
 
           if [ "${GITHUB_EVENT_NAME}" = "push" ]; then
@@ -148,6 +158,7 @@ jobs:
               NEURON_RE='^crates/neuron/|^crates/cortex-core/|^Cargo\.toml$|^Cargo\.lock$|^rpm/helexa-neuron-prerelease\.spec$|^data/neuron|^neuron\.example\.toml$|^\.gitea/workflows/build-prerelease\.yml$'
               CORTEX_RE='^crates/cortex-gateway/|^crates/cortex-cli/|^crates/cortex-core/|^Cargo\.toml$|^Cargo\.lock$|^rpm/cortex-prerelease\.spec$|^data/cortex|^cortex\.example\.toml$|^models\.example\.toml$|^\.gitea/workflows/build-prerelease\.yml$'
               BENCH_RE='^crates/helexa-bench/|^crates/cortex-core/|^Cargo\.toml$|^Cargo\.lock$|^rpm/helexa-bench-prerelease\.spec$|^data/helexa-bench|^helexa-bench\.example\.toml$|^\.gitea/workflows/build-prerelease\.yml$'
+              UPSTREAM_RE='^crates/helexa-upstream/|^crates/cortex-core/|^Cargo\.toml$|^Cargo\.lock$|^rpm/helexa-upstream-prerelease\.spec$|^data/helexa-upstream|^helexa-upstream\.example\.toml$|^\.gitea/workflows/build-prerelease\.yml$'
               # Any Rust change (incl. crates not packaged here, e.g.
               # helexa-acp) still needs lint+test on main.
               RUST_RE='\.rs$|^crates/|Cargo\.toml$|^Cargo\.lock$'
@@ -155,10 +166,12 @@ jobs:
               CORTEX_BASE=$(base_for cortex)
               NEURON_BASE=$(base_for helexa-neuron-blackwell)
               BENCH_BASE=$(base_for helexa-bench)
+              UPSTREAM_BASE=$(base_for helexa-upstream)
               BUILD_CORTEX=$(decide "$CORTEX_BASE" "$CORTEX_RE")
               BUILD_NEURON=$(decide "$NEURON_BASE" "$NEURON_RE")
               BUILD_BENCH=$(decide "$BENCH_BASE" "$BENCH_RE")
-              if [ "$BUILD_CORTEX" = "true" ] || [ "$BUILD_NEURON" = "true" ] || [ "$BUILD_BENCH" = "true" ]; then
+              BUILD_UPSTREAM=$(decide "$UPSTREAM_BASE" "$UPSTREAM_RE")
+              if [ "$BUILD_CORTEX" = "true" ] || [ "$BUILD_NEURON" = "true" ] || [ "$BUILD_BENCH" = "true" ] || [ "$BUILD_UPSTREAM" = "true" ]; then
                 CHECK_RUST=true
               else
                 CHECK_RUST=$(decide "$CORTEX_BASE" "$RUST_RE")
@@ -169,8 +182,9 @@ jobs:
           echo "build_cortex=${BUILD_CORTEX}" >> "$GITHUB_OUTPUT"
           echo "build_neuron=${BUILD_NEURON}" >> "$GITHUB_OUTPUT"
           echo "build_bench=${BUILD_BENCH}" >> "$GITHUB_OUTPUT"
+          echo "build_upstream=${BUILD_UPSTREAM}" >> "$GITHUB_OUTPUT"
           echo "check_rust=${CHECK_RUST}" >> "$GITHUB_OUTPUT"
-          echo "### change detection: build_cortex=${BUILD_CORTEX} build_neuron=${BUILD_NEURON} build_bench=${BUILD_BENCH} check_rust=${CHECK_RUST}"
+          echo "### change detection: build_cortex=${BUILD_CORTEX} build_neuron=${BUILD_NEURON} build_bench=${BUILD_BENCH} build_upstream=${BUILD_UPSTREAM} check_rust=${CHECK_RUST}"
 
   # fmt + clippy + test moved here from ci.yml for main pushes so the
   # two workflows stop queueing against each other (ci.yml's checks
@@ -180,6 +194,7 @@ jobs:
   # fleet, but it also doesn't serialize the pipeline.
   lint:
     name: Lint (fmt + clippy)
+    timeout-minutes: 25
     needs: prepare
     if: needs.prepare.outputs.check_rust == 'true'
     runs-on: rust
@@ -196,38 +211,16 @@ jobs:
         with:
           ref: ${{ inputs.ref }}
       - run: cargo fmt --check --all
-      # sccache failures come in two modes: transient races (a plain
-      # retry clears them) and a wedged/dead server, where every
-      # same-VM retry fails identically (sccache fatal error, ENOENT
-      # on its own tmp files). Escalate accordingly: retry → restart
-      # the server → final attempt uncached. A sick cache costs build
-      # time, never the run.
-      - name: Clippy (with sccache escalation)
-        run: |
-          for attempt in 1 2 3; do
-            echo "::group::clippy attempt ${attempt}"
-            if [ "${attempt}" -eq 3 ]; then
-              echo "final attempt: building without sccache"
-              export RUSTC_WRAPPER=""
-            fi
-            if cargo clippy --workspace -- -D warnings; then
-              echo "::endgroup::"
-              exit 0
-            fi
-            echo "::endgroup::"
-            echo "clippy failed on attempt ${attempt}"
-            if [ "${attempt}" -eq 1 ]; then
-              sccache --stop-server || true
-              sccache --start-server || true
-            fi
-            sleep 5
-          done
-          echo "clippy failed after 3 attempts"
-          exit 1
-      - run: sccache --show-stats || true
+      # Failure-aware sccache escalation lives in the shared script: a
+      # signal death (rustc SIGSEGV / OOM-kill) keeps the cache and fails
+      # fast instead of triggering a slower uncached rebuild; only a real
+      # sccache fault drops the cache. See script/ci-cargo-escalate.sh.
+      - name: Clippy (sccache escalation)
+        run: script/ci-cargo-escalate.sh cargo clippy --workspace -- -D warnings
 
   test:
     name: Test
+    timeout-minutes: 25
     needs: prepare
     if: needs.prepare.outputs.check_rust == 'true'
     runs-on: rust
@@ -243,33 +236,13 @@ jobs:
       - uses: actions/checkout@v4
         with:
           ref: ${{ inputs.ref }}
-      # See the lint job for the escalation rationale.
-      - name: Test (with sccache escalation)
-        run: |
-          for attempt in 1 2 3; do
-            echo "::group::test attempt ${attempt}"
-            if [ "${attempt}" -eq 3 ]; then
-              echo "final attempt: building without sccache"
-              export RUSTC_WRAPPER=""
-            fi
-            if cargo test --workspace; then
-              echo "::endgroup::"
-              exit 0
-            fi
-            echo "::endgroup::"
-            echo "test failed on attempt ${attempt}"
-            if [ "${attempt}" -eq 1 ]; then
-              sccache --stop-server || true
-              sccache --start-server || true
-            fi
-            sleep 5
-          done
-          echo "test failed after 3 attempts"
-          exit 1
-      - run: sccache --show-stats || true
+      # See script/ci-cargo-escalate.sh for the escalation rationale.
+      - name: Test (sccache escalation)
+        run: script/ci-cargo-escalate.sh cargo test --workspace
 
   build-cortex:
     name: Build cortex binary
+    timeout-minutes: 25
     needs: prepare
     if: needs.prepare.outputs.build_cortex == 'true'
     # runner-rust image already provides rust/cargo/clippy/rustfmt via
@@ -288,32 +261,9 @@ jobs:
         with:
           ref: ${{ inputs.ref }}
 
-      # Escalation mirrors the lint/test jobs: retry → restart the
-      # sccache server → final attempt uncached. A sick cache costs
-      # build time, never the run.
-      - name: Build cortex (release, with sccache escalation)
-        run: |
-          for attempt in 1 2 3; do
-            echo "::group::build attempt ${attempt}"
-            if [ "${attempt}" -eq 3 ]; then
-              echo "final attempt: building without sccache"
-              export RUSTC_WRAPPER=""
-            fi
-            if cargo build --release -p cortex-cli; then
-              echo "::endgroup::"
-              sccache --show-stats || true
-              exit 0
-            fi
-            echo "::endgroup::"
-            echo "build failed on attempt ${attempt}"
-            if [ "${attempt}" -eq 1 ]; then
-              sccache --stop-server || true
-              sccache --start-server || true
-            fi
-            sleep 5
-          done
-          echo "build failed after 3 attempts"
-          exit 1
+      # See script/ci-cargo-escalate.sh for the escalation rationale.
+      - name: Build cortex (release, sccache escalation)
+        run: script/ci-cargo-escalate.sh cargo build --release -p cortex-cli
 
       - name: Stage binary
         run: |
@@ -329,6 +279,7 @@ jobs:
 
   build-bench:
     name: Build helexa-bench binary
+    timeout-minutes: 25
     needs: prepare
     if: needs.prepare.outputs.build_bench == 'true'
     # Pure-Rust, non-CUDA binary — same runner as cortex.
@@ -346,32 +297,12 @@ jobs:
         with:
           ref: ${{ inputs.ref }}
 
-      - name: Build helexa-bench (release, with sccache escalation)
+      - name: Build helexa-bench (release, sccache escalation)
         run: |
           # Stamp the SHA helexa-bench records as bench_sha against every
           # run (option_env! in sweep.rs reads it at compile time).
           export HELEXA_BUILD_SHA="$(git rev-parse HEAD)"
-          for attempt in 1 2 3; do
-            echo "::group::build attempt ${attempt}"
-            if [ "${attempt}" -eq 3 ]; then
-              echo "final attempt: building without sccache"
-              export RUSTC_WRAPPER=""
-            fi
-            if cargo build --release -p helexa-bench; then
-              echo "::endgroup::"
-              sccache --show-stats || true
-              exit 0
-            fi
-            echo "::endgroup::"
-            echo "build failed on attempt ${attempt}"
-            if [ "${attempt}" -eq 1 ]; then
-              sccache --stop-server || true
-              sccache --start-server || true
-            fi
-            sleep 5
-          done
-          echo "build failed after 3 attempts"
-          exit 1
+          script/ci-cargo-escalate.sh cargo build --release -p helexa-bench
 
       - name: Stage binary
         run: |
@@ -385,8 +316,48 @@ jobs:
           path: artifacts/helexa-bench
           retention-days: 1
 
+  build-upstream:
+    name: Build helexa-upstream binary
+    timeout-minutes: 25
+    needs: prepare
+    if: needs.prepare.outputs.build_upstream == 'true'
+    # Pure-Rust, non-CUDA binary — same runner as cortex/bench.
+    runs-on: rust
+    env:
+      RUSTC_WRAPPER: sccache
+      SCCACHE_BUCKET: sccache
+      SCCACHE_ENDPOINT: http://caveman.kosherinata.internal:9000
+      SCCACHE_REGION: auto
+      SCCACHE_S3_USE_SSL: "false"
+      AWS_ACCESS_KEY_ID: ${{ secrets.SCCACHE_S3_ACCESS_KEY }}
+      AWS_SECRET_ACCESS_KEY: ${{ secrets.SCCACHE_S3_SECRET_KEY }}
+      # helexa-upstream uses the sqlx runtime query API (no compile-time
+      # query macros), so it builds without a database or a .sqlx cache.
+      # Set OFFLINE defensively so a stray macro can never reach for a DB.
+      SQLX_OFFLINE: "true"
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref }}
+
+      - name: Build helexa-upstream (release, sccache escalation)
+        run: script/ci-cargo-escalate.sh cargo build --release -p helexa-upstream
+
+      - name: Stage binary
+        run: |
+          mkdir --parents artifacts
+          cp target/release/helexa-upstream artifacts/helexa-upstream
+          ./artifacts/helexa-upstream --version || true
+
+      - uses: actions/upload-artifact@v3
+        with:
+          name: upstream-fc43
+          path: artifacts/helexa-upstream
+          retention-days: 1
+
   build-neuron:
     name: Build neuron-${{ matrix.flavour }}
+    timeout-minutes: 35
     needs: prepare
     if: needs.prepare.outputs.build_neuron == 'true'
     strategy:
@@ -413,7 +384,10 @@ jobs:
             cuda_home: /usr/local/cuda-13.0
             build_jobs: 8
             nvcc_threads: 4
-            cargo_features: "cuda cudnn"
+            # flash-attn on blackwell first (#95): beast carries the
+            # agentic prefill pain; ada/ampere follow once the win is
+            # measured. NEURON_FLASH_ATTN=0 is the runtime rollback.
+            cargo_features: "cuda cudnn flash-attn"
     runs-on: ${{ matrix.runner }}
     env:
       SCCACHE_BUCKET: sccache
@@ -427,28 +401,16 @@ jobs:
         with:
           ref: ${{ inputs.ref }}
 
-      # Escalation mirrors the lint/test jobs: retry → restart the
-      # sccache server → final attempt uncached.
-      #
-      # The CUDA image may or may not ship sccache — probe inside this
-      # step (NOT via GITHUB_ENV from a prior step, which this runner
-      # does not propagate; observed: probe step said "enabled", build
-      # ran unwrapped, server stats showed 4 compile requests). A
-      # missing binary degrades to an uncached build rather than
-      # failing cargo at `sccache rustc -vV`. The cache covers the
-      # ~600-crate host-side dep tree (the bulk of the 10-14 min
-      # build); rustc compilations are shared across all three
-      # flavours, so even one run seeds the next.
+      # sccache handling + failure classification lives in
+      # script/ci-cargo-escalate.sh: it probes for sccache (the CUDA
+      # image may not ship it — a missing binary degrades to an uncached
+      # build rather than failing at `sccache rustc -vV`), and a rustc
+      # SIGSEGV / OOM-kill keeps the cache and fails fast instead of
+      # escalating to a slower uncached rebuild. The cache covers the
+      # ~600-crate host-side dep tree (the bulk of the 10-14 min build),
+      # shared across all three flavours, so even one run seeds the next.
       - name: Build neuron with CUDA (${{ matrix.flavour }})
         run: |
-          set -ux
-          if command -v sccache >/dev/null 2>&1; then
-            export RUSTC_WRAPPER=sccache
-            sccache --start-server 2>/dev/null || true
-            echo "sccache enabled"
-          else
-            echo "sccache not on PATH — building uncached"
-          fi
           export PATH="${{ matrix.cuda_home }}/bin:${PATH}"
           export LD_LIBRARY_PATH="${{ matrix.cuda_home }}/targets/x86_64-linux/lib:${{ matrix.cuda_home }}/lib64:${LD_LIBRARY_PATH:-}"
           export LIBRARY_PATH="${{ matrix.cuda_home }}/targets/x86_64-linux/lib:${{ matrix.cuda_home }}/lib64:${LIBRARY_PATH:-}"
@@ -457,27 +419,7 @@ jobs:
           # injecting the exact checked-out commit is unambiguous under
           # shallow/detached states and makes the artifact self-describing.
           export HELEXA_BUILD_SHA="$(git rev-parse HEAD)"
-          for attempt in 1 2 3; do
-            echo "::group::build attempt ${attempt}"
-            if [ "${attempt}" -eq 3 ]; then
-              echo "final attempt: building without sccache"
-              export RUSTC_WRAPPER=""
-            fi
-            if cargo build --release -p neuron --features "${{ matrix.cargo_features }}"; then
-              echo "::endgroup::"
-              command -v sccache >/dev/null 2>&1 && sccache --show-stats || true
-              exit 0
-            fi
-            echo "::endgroup::"
-            echo "build failed on attempt ${attempt}"
-            if [ "${attempt}" -eq 1 ] && command -v sccache >/dev/null 2>&1; then
-              sccache --stop-server || true
-              sccache --start-server || true
-            fi
-            sleep 5
-          done
-          echo "build failed after 3 attempts"
-          exit 1
+          script/ci-cargo-escalate.sh cargo build --release -p neuron --features "${{ matrix.cargo_features }}"
         env:
           CUDA_COMPUTE_CAP: ${{ matrix.compute_cap }}
           CARGO_BUILD_JOBS: ${{ matrix.build_jobs }}
@@ -497,6 +439,7 @@ jobs:
 
   package-cortex:
     name: Package cortex RPM
+    timeout-minutes: 20
     needs: [prepare, build-cortex]
     runs-on: rpm
     steps:
@@ -535,6 +478,7 @@ jobs:
 
   package-bench:
     name: Package helexa-bench RPM
+    timeout-minutes: 20
     needs: [prepare, build-bench]
     runs-on: rpm
     steps:
@@ -555,6 +499,7 @@ jobs:
           cp artifacts/helexa-bench ~/rpmbuild/SOURCES/
           cp data/helexa-bench.service ~/rpmbuild/SOURCES/
           cp data/helexa-bench-sysusers.conf ~/rpmbuild/SOURCES/
+          cp data/helexa-bench-firewalld.xml ~/rpmbuild/SOURCES/
           cp helexa-bench.example.toml ~/rpmbuild/SOURCES/
           cp LICENSE ~/rpmbuild/SOURCES/
           rpmbuild -bb rpm/helexa-bench-prerelease.spec \
@@ -569,8 +514,47 @@ jobs:
           path: ~/rpmbuild/RPMS/x86_64/*.rpm
           retention-days: 7
 
+  package-upstream:
+    name: Package helexa-upstream RPM
+    timeout-minutes: 20
+    needs: [prepare, build-upstream]
+    runs-on: rpm
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref }}
+
+      - uses: actions/download-artifact@v3
+        with:
+          name: upstream-fc43
+          path: artifacts/
+
+      - name: Build RPM
+        run: |
+          set -eux
+          rm -f ~/.rpmmacros
+          rpmdev-setuptree
+          cp artifacts/helexa-upstream ~/rpmbuild/SOURCES/
+          cp data/helexa-upstream.service ~/rpmbuild/SOURCES/
+          cp data/helexa-upstream-sysusers.conf ~/rpmbuild/SOURCES/
+          cp data/helexa-upstream-firewalld.xml ~/rpmbuild/SOURCES/
+          cp helexa-upstream.example.toml ~/rpmbuild/SOURCES/
+          cp LICENSE ~/rpmbuild/SOURCES/
+          rpmbuild -bb rpm/helexa-upstream-prerelease.spec \
+            --define "upstream_version ${{ needs.prepare.outputs.version }}" \
+            --define "upstream_prerelease ${{ needs.prepare.outputs.release }}" \
+            --undefine dist \
+            --define "dist .fc43"
+
+      - uses: actions/upload-artifact@v3
+        with:
+          name: rpm-upstream-fc43
+          path: ~/rpmbuild/RPMS/x86_64/*.rpm
+          retention-days: 7
+
   package-neuron:
     name: Package helexa-neuron-${{ matrix.flavour }} RPM
+    timeout-minutes: 20
     needs: [prepare, build-neuron]
     runs-on: rpm
     strategy:
@@ -616,7 +600,8 @@ jobs:
 
   publish:
     name: Publish to rpm.lair.cafe (unstable)
-    needs: [lint, test, package-cortex, package-neuron, package-bench]
+    timeout-minutes: 25
+    needs: [lint, test, package-cortex, package-neuron, package-bench, package-upstream]
     # Runs when at least one package was built and nothing failed.
     # lint/test may be skipped (docs-only refs never get here because
     # no packages build), but a real failure in any blocks the
@@ -626,10 +611,11 @@ jobs:
         !cancelled()
         && (needs.lint.result == 'success' || needs.lint.result == 'skipped')
         && (needs.test.result == 'success' || needs.test.result == 'skipped')
-        && (needs.package-cortex.result == 'success' || needs.package-neuron.result == 'success' || needs.package-bench.result == 'success')
+        && (needs.package-cortex.result == 'success' || needs.package-neuron.result == 'success' || needs.package-bench.result == 'success' || needs.package-upstream.result == 'success')
         && needs.package-cortex.result != 'failure'
         && needs.package-neuron.result != 'failure'
         && needs.package-bench.result != 'failure'
+        && needs.package-upstream.result != 'failure'
       }}
     runs-on: rpm
     concurrency:

.gitea/workflows/ci.yml

@@ -41,6 +41,7 @@ env:
 jobs:
   fmt:
     name: Format
+    timeout-minutes: 15
     runs-on: rust
     steps:
       - uses: actions/checkout@v4
@@ -48,67 +49,26 @@ jobs:
 
   clippy:
     name: Clippy
+    timeout-minutes: 25
     runs-on: rust
     steps:
       - uses: actions/checkout@v4
-      # sccache failures come in two modes: transient races (a plain
-      # retry clears them) and a wedged/dead server, where every
-      # same-VM retry fails identically. Escalate: retry → restart the
-      # server → final attempt uncached. A sick cache costs build
-      # time, never the run. Keep in sync with build-prerelease.yml.
-      - name: Clippy (with sccache escalation)
-        run: |
-          for attempt in 1 2 3; do
-            echo "::group::clippy attempt ${attempt}"
-            if [ "${attempt}" -eq 3 ]; then
-              echo "final attempt: building without sccache"
-              export RUSTC_WRAPPER=""
-            fi
-            if cargo clippy --workspace -- -D warnings; then
-              echo "::endgroup::"
-              exit 0
-            fi
-            echo "::endgroup::"
-            echo "clippy failed on attempt ${attempt}"
-            if [ "${attempt}" -eq 1 ]; then
-              sccache --stop-server || true
-              sccache --start-server || true
-            fi
-            sleep 5
-          done
-          echo "clippy failed after 3 attempts"
-          exit 1
-      - run: sccache --show-stats || true
+      # Failure-aware sccache escalation lives in the shared script (kept
+      # in sync with build-prerelease.yml): a signal death (rustc SIGSEGV
+      # / OOM-kill) keeps the cache and fails fast instead of an uncached
+      # rebuild; only a real sccache fault drops the cache.
+      - name: Clippy (sccache escalation)
+        run: script/ci-cargo-escalate.sh cargo clippy --workspace -- -D warnings
 
   test:
     name: Test
+    timeout-minutes: 25
     runs-on: rust
     steps:
       - uses: actions/checkout@v4
-      # See the clippy job for the escalation rationale.
-      - name: Test (with sccache escalation)
-        run: |
-          for attempt in 1 2 3; do
-            echo "::group::test attempt ${attempt}"
-            if [ "${attempt}" -eq 3 ]; then
-              echo "final attempt: building without sccache"
-              export RUSTC_WRAPPER=""
-            fi
-            if cargo test --workspace; then
-              echo "::endgroup::"
-              exit 0
-            fi
-            echo "::endgroup::"
-            echo "test failed on attempt ${attempt}"
-            if [ "${attempt}" -eq 1 ]; then
-              sccache --stop-server || true
-              sccache --start-server || true
-            fi
-            sleep 5
-          done
-          echo "test failed after 3 attempts"
-          exit 1
-      - run: sccache --show-stats || true
+      # See script/ci-cargo-escalate.sh for the escalation rationale.
+      - name: Test (sccache escalation)
+        run: script/ci-cargo-escalate.sh cargo test --workspace
 
   # Type-check the CUDA-only code path. Borrow-check-only — we
   # never run the tests here (the runner has no GPU). This catches
@@ -122,6 +82,9 @@ jobs:
   # see commit history).
   cuda-check:
     name: CUDA type-check
+    # flash-attn kernel compilation dominates the first uncached run;
+    # sccache + the cargo target cache absorb it afterwards.
+    timeout-minutes: 70
     runs-on: cuda-13.0
     # The workflow-level env sets `RUSTC_WRAPPER: sccache`
     # unconditionally, which hard-fails cargo if the CUDA image
@@ -139,52 +102,26 @@ jobs:
       CUDA_COMPUTE_CAP: "86"
     steps:
       - uses: actions/checkout@v4
-      # sccache is probed inside this step (NOT via GITHUB_ENV from a
-      # prior step — this runner doesn't propagate it; see
-      # build-prerelease.yml for the observed failure).
-      - name: cargo check --features cuda (with sccache escalation)
+      # sccache probing + failure classification lives in the shared
+      # script (see build-prerelease.yml's neuron build for the same
+      # pattern). It probes for sccache and, on a rustc SIGSEGV / OOM,
+      # keeps the cache and fails fast rather than rebuilding uncached.
+      - name: cargo check --features cuda (sccache escalation)
         run: |
-          if command -v sccache >/dev/null 2>&1; then
-            export RUSTC_WRAPPER=sccache
-            sccache --start-server 2>/dev/null || true
-            echo "sccache enabled"
-          else
-            echo "sccache not on PATH — building uncached"
-          fi
           # act launches the step shell without /etc/profile, so the
           # gitea_runner user's inherited PATH lacks /usr/local/cuda-13.0/bin.
-          # cudarc's build.rs:157 shells out to `nvcc --version` (because
-          # the neuron crate enables cuda-version-from-build-system) and
-          # panics with ENOENT if nvcc isn't resolvable. build-prerelease.yml
-          # does the same export — keep them in sync.
+          # cudarc's build.rs shells out to `nvcc --version` (the neuron
+          # crate enables cuda-version-from-build-system) and panics with
+          # ENOENT if nvcc isn't resolvable — keep this export in sync
+          # with build-prerelease.yml.
           export PATH="/usr/local/cuda-13.0/bin:${PATH}"
           export LD_LIBRARY_PATH="/usr/local/cuda-13.0/targets/x86_64-linux/lib:/usr/local/cuda-13.0/lib64:${LD_LIBRARY_PATH:-}"
           export LIBRARY_PATH="/usr/local/cuda-13.0/targets/x86_64-linux/lib:/usr/local/cuda-13.0/lib64:${LIBRARY_PATH:-}"
-          # Escalation mirrors the lint/test jobs: plain retry →
-          # sccache server restart → final attempt uncached.
-          for attempt in 1 2 3; do
-            echo "::group::cuda-check attempt ${attempt}"
-            if [ "${attempt}" -eq 3 ]; then
-              echo "final attempt: building without sccache"
-              export RUSTC_WRAPPER=""
-            fi
-            if cargo check -p neuron --features cuda --all-targets; then
-              echo "::endgroup::"
-              exit 0
-            fi
-            echo "::endgroup::"
-            echo "cuda-check failed on attempt ${attempt}"
-            if [ "${attempt}" -eq 1 ] && command -v sccache >/dev/null 2>&1; then
-              sccache --stop-server || true
-              sccache --start-server || true
-            fi
-            sleep 5
-          done
-          echo "cuda-check failed after 3 attempts"
-          exit 1
+          script/ci-cargo-escalate.sh cargo check -p neuron --features cuda,flash-attn --all-targets
 
   srpm-cortex:
     name: Build cortex SRPM
+    timeout-minutes: 25
     runs-on: rpm
     needs: [fmt, clippy, test, cuda-check]
     if: startsWith(github.ref, 'refs/tags/v')
@@ -245,6 +182,7 @@ jobs:
 
   srpm-neuron:
     name: Build neuron SRPM
+    timeout-minutes: 25
     runs-on: rpm
     needs: [fmt, clippy, test, cuda-check]
     if: startsWith(github.ref, 'refs/tags/v')
@@ -305,6 +243,7 @@ jobs:
 
   copr-cortex:
     name: Publish cortex to COPR
+    timeout-minutes: 60
     runs-on: fedora-43
     needs: srpm-cortex
     steps:
@@ -322,6 +261,7 @@ jobs:
 
   copr-neuron:
     name: Publish neuron to COPR
+    timeout-minutes: 60
     runs-on: fedora-43
     needs: srpm-neuron
     steps:
@@ -339,6 +279,7 @@ jobs:
 
   bump-version:
     name: Bump version in source
+    timeout-minutes: 15
     runs-on: rust
     needs: [copr-cortex, copr-neuron]
     steps:

.gitea/workflows/deploy-dev.yml

@@ -123,7 +123,9 @@ jobs:
             # Exact command form required by the sudoers rule in
             # asset/sudoers.d/neuron-host.conf — change both together.
             sudo /usr/bin/install -o root -g root -m 0755 /var/lib/gitea_ci/neuron-dev /usr/bin/neuron
-            sudo /usr/bin/systemctl start neuron.service
+            # enable --now so a dev deploy also leaves the unit enabled
+            # for boot, consistent with deploy.yml.
+            sudo /usr/bin/systemctl enable --now neuron.service
             rm -f /var/lib/gitea_ci/neuron-dev'
 
       - name: Capture neuron.service startup journal

.gitea/workflows/deploy.yml

@@ -1,7 +1,8 @@
 name: deploy
 
 # Roll the freshly-published unstable RPMs onto the helexa fleet:
-# cortex on the gateway, helexa-neuron-<flavour> on each neuron host.
+# cortex on the gateway, helexa-neuron-<flavour> on each neuron host,
+# and helexa-bench on bob (the bench host).
 #
 # Triggered automatically after `build-prerelease` succeeds (by which
 # point the new RPMs are live on rpm.lair.cafe/unstable), and also
@@ -88,7 +89,9 @@ jobs:
             sudo /usr/bin/dnf install --refresh --allowerasing -y cortex
           fi
           sudo /usr/bin/systemctl daemon-reload
-          sudo /usr/bin/systemctl start cortex.service
+          # enable --now: start the service AND enable it for boot so the
+          # fleet self-heals after a host reboot.
+          sudo /usr/bin/systemctl enable --now cortex.service
           DEPLOY
 
       # Wait for the service to either come up or wedge, then capture
@@ -115,15 +118,27 @@ jobs:
           # loading after a restart. beast cold-loads Qwen3.6-27B Q6K
           # TP=2 (~5-6 min typical, see #1); benjy/quadbrat load small
           # single-GPU models in well under a minute.
+          #
+          # max_prompt_tokens: per-model context cap, written to the
+          # neuron.service.d/model.conf drop-in (NEURON_MAX_PROMPT_TOKENS).
+          # A change here restarts the neuron even with no new RPM. Values
+          # are VRAM-safe ceilings derived per model — see
+          # doc/context-limits.md. beast (Qwen3.6-27B, hybrid linear, 2x
+          # 32GB) has ample KV headroom; benjy (Qwen3-8B dense, ~6GB free)
+          # is VRAM-bound and stays at the default; quadbrat (Qwen3-1.7B)
+          # likewise conservative.
           - host: beast.hanzalova.internal
             flavour: blackwell
             load_timeout: 900
+            max_prompt_tokens: 131072
           - host: benjy.hanzalova.internal
             flavour: ada
             load_timeout: 300
+            max_prompt_tokens: 16384
           - host: quadbrat.hanzalova.internal
             flavour: ampere
             load_timeout: 300
+            max_prompt_tokens: 16384
     steps:
       - name: SSH init
         run: |
@@ -140,6 +155,26 @@ jobs:
           ssh gitea_ci@${{ matrix.host }} 'bash -s' <<'DEPLOY'
           set -eu
           pkg=helexa-neuron-${{ matrix.flavour }}
+          max_prompt_tokens="${{ matrix.max_prompt_tokens }}"
+
+          # ── Desired per-model systemd drop-in ─────────────────────────
+          # model.conf carries NEURON_MAX_PROMPT_TOKENS so the context cap
+          # is deterministic per host and rolled out (with a restart) by
+          # this workflow, not hand-edited. It sorts after local.conf, so a
+          # deploy-managed value wins over any manual local override of the
+          # same variable. See doc/context-limits.md.
+          conf=/etc/systemd/system/neuron.service.d/model.conf
+          config_changed=0
+          if [ -n "${max_prompt_tokens}" ]; then
+            desired=$(printf '%s\n%s\n%s\n%s' \
+              "# Managed by .gitea/workflows/deploy.yml - do not edit by hand." \
+              "# Per-model context cap; see doc/context-limits.md." \
+              "[Service]" \
+              "Environment=NEURON_MAX_PROMPT_TOKENS=${max_prompt_tokens}")
+            [ "${desired}" = "$(cat "${conf}" 2>/dev/null || true)" ] || config_changed=1
+          fi
+
+          # ── Package version gate (manifest rationale: see deploy-cortex) ──
           installed=$(rpm -q --qf '%{VERSION}-%{RELEASE}' "${pkg}" 2>/dev/null || echo "not-installed")
           latest=$(curl -fsS --max-time 15 "https://rpm.lair.cafe/fedora/43/x86_64/unstable/packages.json" 2>/dev/null \
             | python3 -c '
@@ -150,21 +185,42 @@ jobs:
               p = max(cands, key=lambda p: p.get("buildTime", 0))
               print(p["version"] + "-" + p["release"])
           ' "${pkg}" 2>/dev/null || true)
+          pkg_changed=1
           if [ -n "${latest}" ] && [ "${latest}" = "${installed}" ]; then
-            echo "${pkg}-${installed} already current — leaving service untouched"
+            pkg_changed=0
+          fi
+
+          # Skip only when BOTH the package and the drop-in are unchanged —
+          # a context-cap change must restart the neuron even with no new RPM.
+          if [ "${pkg_changed}" -eq 0 ] && [ "${config_changed}" -eq 0 ]; then
+            echo "${pkg}-${installed} current; NEURON_MAX_PROMPT_TOKENS=${max_prompt_tokens:-<unset>} unchanged — leaving service untouched"
             exit 0
           fi
-          echo "installed=${installed} published=${latest:-unknown} — deploying"
+          echo "installed=${installed} published=${latest:-unknown} pkg_changed=${pkg_changed} config_changed=${config_changed} — deploying"
+
+          # Write the drop-in (staged in gitea_ci's dir, installed root-owned).
+          if [ "${config_changed}" -eq 1 ]; then
+            printf '%s\n' "${desired}" > /var/lib/gitea_ci/model.conf
+            sudo /usr/bin/install -o root -g root -m 0644 -D /var/lib/gitea_ci/model.conf "${conf}"
+            rm -f /var/lib/gitea_ci/model.conf
+            echo "applied ${conf}: NEURON_MAX_PROMPT_TOKENS=${max_prompt_tokens}"
+          fi
+
           if systemctl is-active --quiet neuron.service; then
             sudo /usr/bin/systemctl stop neuron.service
           fi
+          if [ "${pkg_changed}" -eq 1 ]; then
             if rpm -q "${pkg}" >/dev/null 2>&1; then
               sudo /usr/bin/dnf upgrade --refresh --allowerasing -y "${pkg}"
             else
               sudo /usr/bin/dnf install --refresh --allowerasing -y "${pkg}"
             fi
+          fi
+          # daemon-reload picks up both a new unit (dnf) and the drop-in.
           sudo /usr/bin/systemctl daemon-reload
-          sudo /usr/bin/systemctl start neuron.service
+          # enable --now: start the service AND enable it for boot so the
+          # fleet self-heals after a host reboot.
+          sudo /usr/bin/systemctl enable --now neuron.service
 
           # ── Post-deploy validation ────────────────────────────────
           # A deploy only goes green if the neuron (a) finishes loading
@@ -218,19 +274,52 @@ jobs:
 
           echo "LLM probe against ${model}"
           probe_body=$(printf '{"model":"%s","messages":[{"role":"user","content":"Reply with exactly one word: pineapple"}],"max_tokens":512,"temperature":0}' "${model}")
-          resp=$(curl -fsS --max-time 180 -H "content-type: application/json" \
-            -d "${probe_body}" http://localhost:13131/v1/chat/completions) || {
-            echo "FAIL: probe request errored"
-            exit 1
-          }
+          # The probe races real traffic: a deploy publishes a new build
+          # SHA, which is exactly what triggers helexa-bench to re-sweep
+          # every scenario against this neuron — long capability-probe
+          # generations can hold the batch-1 model for minutes. Admission
+          # control answers concurrent requests with 429/503 +
+          # Retry-After (#53/#63); treat those as "busy, try again", not
+          # deploy failure. Overall budget ~6 min.
+          probe_deadline=$(( $(date +%s) + 360 ))
+          attempt=0
+          while :; do
+            attempt=$(( attempt + 1 ))
+            hdrs=$(mktemp)
+            resp=$(curl -sS --max-time 180 -D "${hdrs}" \
+              -H "content-type: application/json" \
+              -d "${probe_body}" http://localhost:13131/v1/chat/completions) || resp=""
+            status=$(awk 'toupper($1) ~ /^HTTP/ {code=$2} END {print code}' "${hdrs}")
+            retry_after=$(awk 'tolower($1) == "retry-after:" {print $2+0; exit}' "${hdrs}")
+            rm -f "${hdrs}"
+            if [ "${status}" = "200" ]; then
               if printf %s "${resp}" | grep -qi pineapple; then
-            echo "LLM probe passed"
-          else
+                echo "LLM probe passed (attempt ${attempt})"
+                break
+              fi
               echo "FAIL: probe response missing expected token"
               printf %s "${resp}" | head -c 2000
               echo
               exit 1
             fi
+            case "${status}" in
+              429|503) ;;  # busy/queue-full — retryable
+              *)
+                echo "FAIL: probe request errored (HTTP ${status:-none})"
+                printf %s "${resp}" | head -c 2000
+                echo
+                exit 1
+                ;;
+            esac
+            if [ "$(date +%s)" -ge "${probe_deadline}" ]; then
+              echo "FAIL: model still busy (HTTP ${status}) after probe budget"
+              exit 1
+            fi
+            wait_s=${retry_after:-15}
+            [ "${wait_s}" -ge 5 ] 2>/dev/null || wait_s=15
+            echo "model busy (HTTP ${status}), retrying in ${wait_s}s (attempt ${attempt})"
+            sleep "${wait_s}"
+          done
           DEPLOY
 
       - name: Ensure firewalld allows helexa-neuron
@@ -250,3 +339,143 @@ jobs:
           sleep 10
           ssh gitea_ci@${{ matrix.host }} \
               'journalctl --unit neuron.service -I --no-pager'
+
+  # helexa-bench is a separate package on a separate host (bob), and it
+  # only consumes the fleet's HTTP APIs — it has no deploy-ordering
+  # dependency on cortex or the neurons (the sweep loop is version-aware
+  # and picks up whatever each neuron reports whenever). So it runs
+  # alongside the cortex→neurons chain rather than after it.
+  deploy-bench:
+    runs-on: fedora-43
+    if: >-
+      ${{
+        github.event_name == 'workflow_dispatch'
+        || github.event.workflow_run.conclusion == 'success'
+      }}
+    steps:
+      - name: SSH init
+        run: |
+          mkdir -p ~/.ssh
+          echo "${DEPLOY_KEY}" > ~/.ssh/id_ed25519
+          chmod 600 ~/.ssh/id_ed25519
+          ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=accept-new \
+              gitea_ci@bob.hanzalova.internal 'hostname -f'
+
+      # See deploy-cortex for why gating uses the publish manifest and
+      # not unprivileged `dnf check-update`.
+      - name: Deploy helexa-bench (skips when already current)
+        run: |
+          ssh gitea_ci@bob.hanzalova.internal 'bash -s' <<'DEPLOY'
+          set -eu
+          pkg=helexa-bench
+          installed=$(rpm -q --qf '%{VERSION}-%{RELEASE}' "${pkg}" 2>/dev/null || echo "not-installed")
+          latest=$(curl -fsS --max-time 15 "https://rpm.lair.cafe/fedora/43/x86_64/unstable/packages.json" 2>/dev/null \
+            | python3 -c '
+          import json, sys
+          name = sys.argv[1]
+          cands = [p for p in json.load(sys.stdin)["packages"] if p.get("name") == name]
+          if cands:
+              p = max(cands, key=lambda p: p.get("buildTime", 0))
+              print(p["version"] + "-" + p["release"])
+          ' "${pkg}" 2>/dev/null || true)
+          if [ -n "${latest}" ] && [ "${latest}" = "${installed}" ]; then
+            echo "${pkg}-${installed} already current — leaving service untouched"
+            exit 0
+          fi
+          echo "installed=${installed} published=${latest:-unknown} — deploying"
+          if systemctl is-active --quiet helexa-bench.service; then
+            sudo /usr/bin/systemctl stop helexa-bench.service
+          fi
+          if rpm -q "${pkg}" >/dev/null 2>&1; then
+            sudo /usr/bin/dnf upgrade --refresh --allowerasing -y helexa-bench
+          else
+            sudo /usr/bin/dnf install --refresh --allowerasing -y helexa-bench
+          fi
+          sudo /usr/bin/systemctl daemon-reload
+          # enable --now: start the service AND enable it for boot so the
+          # bench resumes collecting after a host reboot.
+          sudo /usr/bin/systemctl enable --now helexa-bench.service
+
+          # ── Post-deploy validation ────────────────────────────────
+          # The bench serves a read-only API on :13132 alongside the
+          # outbound sweep loop. Probe the API over localhost (bypasses
+          # firewalld) — catches a crash-on-start or a bad bind. Bail
+          # early if the unit drops out of active (Restart backoff).
+          echo "waiting for bench API on :13132"
+          deadline=$(( $(date +%s) + 30 ))
+          while :; do
+            if curl -fsS --max-time 5 http://localhost:13132/api/health >/dev/null 2>&1; then
+              echo "bench API healthy"
+              break
+            fi
+            if ! systemctl is-active --quiet helexa-bench.service; then
+              echo "FAIL: helexa-bench.service is not active"
+              systemctl --no-pager status helexa-bench.service | head -20 || true
+              exit 1
+            fi
+            if [ "$(date +%s)" -ge "${deadline}" ]; then
+              echo "FAIL: bench API not healthy within 30s"
+              exit 1
+            fi
+            sleep 3
+          done
+          DEPLOY
+
+      - name: Ensure firewalld allows helexa-bench
+        run: |
+          ssh gitea_ci@bob.hanzalova.internal '
+            if ! sudo /usr/bin/firewall-cmd --query-service=helexa-bench --quiet 2>/dev/null; then
+              sudo /usr/bin/firewall-cmd --add-service=helexa-bench --permanent
+              sudo /usr/bin/firewall-cmd --reload
+            fi'
+
+      # Wait for the service to either come up or wedge, then capture
+      # the latest-invocation journal. Runs even on prior failure so a
+      # failed start step still leaves a usable record in the deploy log.
+      - name: Capture helexa-bench.service startup journal
+        if: always()
+        run: |
+          sleep 10
+          ssh gitea_ci@bob.hanzalova.internal \
+              'journalctl --unit helexa-bench.service -I --no-pager'
+
+  # Build the bench UI and publish it to the public nginx vhost on the
+  # gateway (https://bench.helexa.ai). The vhost + Let's Encrypt cert are
+  # one-time host setup (script/infra-setup.sh); this job just refreshes
+  # the static assets. nginx reverse-proxies /api to the bob API, so the
+  # SPA is built same-origin (no VITE_API_BASE). Independent of the other
+  # deploy jobs.
+  deploy-bench-ui:
+    runs-on: fedora-43
+    if: >-
+      ${{
+        github.event_name == 'workflow_dispatch'
+        || github.event.workflow_run.conclusion == 'success'
+      }}
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: "20"
+
+      - name: Build UI
+        run: |
+          cd bench
+          npm ci
+          npm run build
+
+      - name: SSH init
+        run: |
+          mkdir -p ~/.ssh
+          echo "${DEPLOY_KEY}" > ~/.ssh/id_ed25519
+          chmod 600 ~/.ssh/id_ed25519
+          ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=accept-new \
+              gitea_ci@hanzalova.internal 'hostname -f'
+
+      - name: Rsync built UI to gateway webroot
+        run: |
+          rsync --archive --compress --delete \
+            --rsync-path 'sudo rsync' \
+            bench/dist/ \
+            gitea_ci@hanzalova.internal:/var/www/bench.helexa.ai/

.gitignore

@@ -1,4 +1,9 @@
 /target
+/bench/node_modules
+/bench/dist
+/helexa.ai/node_modules
+/helexa.ai/dist
+helexa.ai/.env.local
 *.swp
 *.swo
 .idea/

AGENTS.md

@@ -0,0 +1,268 @@
+# AGENTS.md — helexa/cortex
+
+## Project Overview
+
+helexa is a self-hosted LLM serving stack for multi-node GPU inference clusters. It has two components:
+
+- **cortex** — the per-operator control plane and LLM proxy. A Rust reverse-proxy that sits in front of the fleet and presents a unified OpenAI + Anthropic compatible API surface. It handles model routing, lifecycle management (load/unload/evict), request translation, and metrics collection.
+- **neuron** — the per-host LLM harness. One instance runs on every GPU host, serving candle-based in-process inference and managing local hardware discovery and model lifecycle.
+
+## Repository Layout
+
+```
+cortex/
+├── Cargo.toml              # workspace root (Rust 2024 edition, GPL-3.0)
+├── cortex.example.toml     # example gateway config
+├── models.example.toml     # example model catalogue
+├── neuron.example.toml     # example neuron config
+├── README.md               # public-facing documentation
+├── CLAUDE.md               # detailed design rationale and implementation history
+├── AGENTS.md               # ← you are here
+├── cortex.spec             # RPM spec for cortex
+├── helexa-neuron.spec      # RPM spec for neuron (renamed to avoid Fedora collision)
+├── rpm/                    # prerelease RPM specs
+│   ├── cortex-prerelease.spec
+│   ├── helexa-neuron-prerelease.spec
+│   └── helexa-bench-prerelease.spec
+├── data/                   # systemd units and example configs for packaging
+│   ├── cortex.service
+│   ├── neuron.service
+│   ├── cortex.example.toml
+│   ├── neuron.example.toml
+│   └── models.example.toml
+└── crates/
+    ├── cortex-core/            # shared types, config, envelopes
+    │   └── src/
+    │       ├── lib.rs
+    │       ├── build_info.rs   # BuildInfo type for /version endpoint
+    │       ├── config.rs       # figment-based config structs
+    │       ├── catalogue.rs    # ModelProfile, placement matching
+    │       ├── discovery.rs    # DeviceInfo, DiscoveryResponse
+    │       ├── harness.rs      # Harness trait, HarnessConfig, HarnessHealth
+    │       ├── node.rs         # NodeState, ModelStatus
+    │       ├── openai.rs       # OpenAI request/response types
+    │       ├── anthropic.rs    # Anthropic request/response types
+    │       ├── translate.rs    # OpenAI <-> Anthropic translation
+    │       └── metrics.rs      # RequestMetrics, histogram helpers
+    ├── cortex-gateway/         # the HTTP proxy server
+    │   └── src/
+    │       ├── lib.rs
+    │       ├── state.rs        # CortexState: Arc<RwLock<...>>
+    │       ├── router.rs       # model -> node routing logic
+    │       ├── proxy.rs        # streaming HTTP proxy to backends
+    │       ├── evictor.rs      # LRU/priority eviction logic
+    │       ├── poller.rs       # background task polling neuron status
+    │       ├── handlers.rs     # axum handlers (chat, completions, models, etc.)
+    │       └── metrics.rs      # prometheus exporter endpoint
+    ├── cortex-cli/             # CLI entrypoint
+    │   └── src/main.rs         # binary: `cortex`
+    ├── neuron/                 # per-host LLM daemon (replaces cortex-agent)
+    │   ├── Cargo.toml          # features: cuda, cudnn, flash-attn, cuda-integration
+    │   ├── build.rs            # compiles CUDA kernels, emits build metadata
+    │   └── src/
+    │       ├── main.rs         # binary: `neuron`
+    │       ├── discovery.rs    # nvidia-smi parsing, device enumeration
+    │       ├── health.rs       # runtime GPU polling
+    │       ├── api.rs          # HTTP handlers for /discovery, /models, etc.
+    │       ├── version.rs      # GET /version endpoint with BuildInfo
+    │       ├── models.rs       # local model lifecycle orchestration
+    │       └── harness/        # in-process candle inference
+    │           ├── device_worker/  # per-device CUDA worker threads
+    │           │   ├── mod.rs      # canonical narrative for worker architecture
+    │           │   ├── jobs.rs     # Job enum, dispatch handlers
+    │ │           └── dispatch.rs   # DeviceWorkerState struct
+    │           ├── candle.rs       # candle model implementation
+    │           └── tp/             # tensor parallelism
+    │               └── worker.rs   # TP worker subprocesses
+    ├── helexa-acp/             # Agent Client Protocol bridge (Apache-2.0)
+    │   └── src/main.rs         # binary: `helexa-acp`, self-contained (no workspace deps)
+    └── helexa-bench/           # benchmark harness
+        └── src/main.rs         # binary: `helexa-bench`, SQLite-backed, version-aware
+```
+
+## Key Design Decisions
+
+### Architecture
+- **cortex** is the control plane. It exposes the unified API, routes requests, manages model lifecycle across the fleet, and collects metrics.
+- **neuron** is the node plane. One instance runs on every GPU host. It discovers local hardware, manages in-process candle inference, handles NCCL tensor parallelism, and reports runtime state.
+- cortex never shells out to `nvidia-smi`, never touches systemd units, and never talks directly to a harness. It talks only to neurons via HTTP API on port 13131.
+
+### Per-device worker thread (neuron)
+Every CUDA device gets one dedicated OS thread that owns its `CudaContext` for the daemon's lifetime. All CUDA operations route through this thread via a `std::sync::mpsc` job channel. Tensors never escape the worker thread alive. Inference replies carry `Vec<f32>` CPU-side logits; sampled tokens come back as `u32`. The opaque `ArchHandle(u64)` and `TpHandle(u64)` are indices into the worker's state slab, not pointers.
+
+CPU loads (`Device::Cpu` fallback) keep the legacy `tokio::task::spawn_blocking + Arc<Mutex<ModelArch>>` path — there's no context to own and the channel hop would only add latency. Four `spawn_blocking` references in `harness/candle.rs` are deliberate CPU fallback.
+
+### candle-native (not mistral.rs)
+neuron builds directly on [candle](https://github.com/huggingface/candle). Every model architecture it serves is implemented in this repository, ported against the HuggingFace reference. No external inference server to babysit. The Harness trait remains as an internal seam for adding future engines (vision/audio/diffusion) but its only implementation is in-process candle.
+
+### Streaming proxy
+Chat completions are proxied as SSE streams. The gateway must:
+1. Parse the inbound request to extract the model name
+2. Route to the correct backend neuron
+3. Stream the response back, capturing token timing for metrics
+4. NOT buffer the full response — true streaming passthrough
+
+### Anthropic translation
+When a request arrives at `/v1/messages` (Anthropic format), the gateway translates it to OpenAI format before proxying to neuron, then translates the response back. This is stateless envelope transformation. Non-streaming round-trip is implemented; streaming SSE translation deferred.
+
+### Eviction
+The evictor runs as a background task. Before loading a model on a node where VRAM is tight:
+1. Check if the model is already loaded elsewhere → route there instead
+2. Find the LRU model on the target node (excluding pinned models)
+3. Call `POST {neuron}/models/unload` on that model
+4. The incoming request's lazy-load triggers the new model load
+
+### Metrics
+Per-request: model, node, prompt_tokens, completion_tokens, total_tokens, tok_per_sec, time_to_first_token_ms, total_latency_ms. Exposed as Prometheus histograms/counters on a separate port (31314).
+
+## Tech Stack
+
+- **Rust 2024 edition** — workspace with 6 crates
+- **Axum 0.8** — HTTP framework
+- **reqwest** — HTTP client for proxying to backends
+- **figment** — config loading (TOML + env vars)
+- **tokio** — async runtime
+- **metrics + metrics-exporter-prometheus** — observability
+- **tracing** — structured logging
+- **candle** — in-process inference engine (neuron only, with CUDA support)
+- **cudarc** — patched for neuron's needs (see workspace `[patch]`)
+- **clap** — CLI parsing
+- **rusqlite** (bundled) — helexa-bench SQLite system-of-record
+
+## Build Commands
+
+```sh
+cargo build --release           # build all crates
+cargo run -p cortex-cli -- serve    # run the gateway
+cargo test                      # run all tests
+cargo clippy --workspace        # lint
+```
+
+### neuron Features
+- `cuda`: Enables CUDA acceleration in candle and cudarc/nccl bindings. Without it, falls back to CPU.
+- `cudnn`: Use cuDNN for convolution/attention kernels (requires `cuda`).
+- `flash-attn`: FlashAttention kernels (requires `cuda`).
+- `cuda-integration`: Reserved for GPU-only integration tests (requires multiple CUDA devices + libnccl).
+
+### Build Scripts
+- `neuron/build.rs`: Compiles CUDA kernels (`src/cuda/*.cu`) using `cudaforge::KernelBuilder` when `cuda` feature is enabled. Handles compute capability checks (sm_<80 disables bf16 intrinsics). Also captures build metadata: git SHA, dirty flag, timestamp, rustc version, profile, features, candle-core version.
+
+## CI
+
+Gitea Actions runs on every push to any branch. All three checks must pass before merging:
+
+```sh
+cargo fmt --check --all                    # formatting
+cargo clippy --workspace -- -D warnings   # lint (warnings are errors)
+cargo test --workspace                     # tests
+```
+
+Run these locally before pushing. `cargo fmt --all` fixes formatting automatically. Clippy warnings must be resolved, not suppressed with `#[allow(...)]` unless there is a clear rationale.
+
+Tagged releases (`v*`) build SRPMs for `cortex`, `helexa-neuron`, and `helexa-bench` and publish to COPR (`helexa/helexa`). Build metadata SHA injection: CI sets `HELEXA_BUILD_SHA=$(git rev-parse HEAD)`.
+
+## Environment
+
+- Targets Fedora 43 (systemd, SELinux enforcing)
+- Nodes communicate over a private network (e.g. WireGuard mesh)
+- cortex listens on port 31313 (API) and 31314 (metrics)
+- neuron listens on port 13131 on each GPU host
+- TLS terminated at gateway or via nginx; internal traffic is plaintext over WireGuard
+
+## Conventions
+
+- Error handling: `anyhow` for binaries, `thiserror` for library crates
+- No `unwrap()` in library code; `expect()` only with clear rationale
+- All public types derive `Debug, Clone, Serialize, Deserialize` where sensible
+- Config structs use `figment` with TOML as primary source, env vars as override
+- Prefer `Arc<RwLock<...>>` for shared fleet state; minimize lock duration
+- SSE streaming uses `tokio_stream` + `eventsource-stream` for parsing
+- Log at `info` for request routing, `debug` for proxy details, `warn` for eviction and node health, `error` for proxy failures
+
+## Testing
+
+### Gateway tests
+Use mock neurons spawned via axum in `crates/cortex-gateway/tests/common/mod.rs`. Helpers: `spawn_mock_backend()`, `spawn_gateway()`.
+
+### neuron integration tests
+- Numerical reference tests (`numerical_reference.rs`) require `NEURON_REF_MODEL_PATH` env var pointing to a HF snapshot directory. Fixtures are f32-based for precision validation against HuggingFace transformers.
+- CUDA integration tests (`tp_worker_lifecycle_cuda.rs`) gated behind `cuda-integration` feature; requires 2+ CUDA devices (e.g., 2x RTX 5090).
+
+### Metrics testing
+Use `install_test_recorder()` in test code to capture metrics without the HTTP listener.
+
+## helexa-bench
+
+A continuous, version-aware benchmark harness. Hits each neuron directly on `:13131`, exercises each warm model with a Scenario suite (chat-latency family), and records results into SQLite stamped with the neuron's full `BuildInfo`. The loop is version-aware: skips any (target, build SHA, model, scenario) cell already at `samples_per_version`.
+
+Packaged as `helexa-bench` RPM (prebuilt-binary spec). One systemd unit, typically on the metrics host.
+
+## helexa-acp
+
+Agent Client Protocol bridge — connects ACP editors (Zed, etc.) to any OpenAI-compatible endpoint, cortex by default. Intentionally self-contained: no workspace crate dependencies. Uses `agent-client-protocol` with `unstable_session_model` feature for Zed model picker support. Licensed Apache-2.0 (workspace is GPL-3.0).
+
+## RPM Packaging
+
+- `cortex.spec` — installs the `cortex` binary
+- `helexa-neuron.spec` — installs the `neuron` binary under package name `helexa-neuron` (renamed to avoid Fedora's NEURON neural-simulation package collision)
+- Systemd units in `data/cortex.service`, `data/neuron.service`
+- Example configs: `cortex.example.toml`, `neuron.example.toml`, `models.example.toml`
+
+Install:
+```sh
+dnf copr enable helexa/helexa
+dnf install cortex                # gateway host
+dnf install helexa-neuron         # GPU nodes
+```
+
+## Configuration Files
+
+### cortex.toml (gateway)
+```toml
+[gateway]
+listen = "0.0.0.0:31313"
+metrics_listen = "0.0.0.0:31314"
+
+[eviction]
+strategy = "lru"          # lru | priority
+defrag_after_cycles = 50
+
+[[neurons]]
+name = "beast"
+endpoint = "http://beast.internal:13131"
+```
+
+### models.toml (catalogue)
+```toml
+[[models]]
+id = "Qwen/Qwen3-Coder-30B-A3B-Instruct"
+harness = "candle"
+quant = "Q4_K_M"
+vram_mb = 19000
+min_devices = 2
+min_device_vram_mb = 10000
+pinned_on = ["beast"]       # optional: never evict from these neurons
+```
+
+### neuron.toml (per-host)
+Configured via figment + env override. See `neuron.example.toml` for reference.
+
+## neuron API Endpoints
+
+```
+GET  /discovery        → hardware discovery (hostname, OS, CUDA, devices, harnesses)
+GET  /health           → runtime GPU stats (VRAM, utilization, temperature)
+GET  /models           → loaded/unloaded models with VRAM usage
+POST /models/load      → load a model with spec (quant, TP, devices)
+POST /models/unload    → unload a model, freeing device memory
+GET  /models/{id}/endpoint → inference URL for a model
+GET  /version          → build metadata (SHA, features, candle version, etc.)
+```
+
+## Sources of Truth
+
+When prose documentation conflicts with code, trust:
+1. Executable configuration (`*.toml`, `Cargo.toml` features)
+2. Type definitions in `cortex-core/`
+3. Test files in `crates/*/tests/` and `*/src/**/*_test.rs`
+4. `CLAUDE.md` for historical design rationale

CLAUDE.md

@@ -185,6 +185,32 @@ Run these locally before pushing. `cargo fmt --all` fixes formatting
 automatically. Clippy warnings must be resolved, not suppressed with
 `#[allow(...)]` unless there is a clear rationale.
 
+## Development workflow
+
+Work each change on its own branch; `main` stays releasable.
+
+1. Implement on a feature branch (`fix/<issue>-…`, `feat/<issue>-…`).
+2. Run the CI triad locally (`cargo fmt --check --all`,
+   `cargo clippy --workspace -- -D warnings`, `cargo test --workspace`).
+   Local builds are **CPU-only** — the `#[cfg(feature = "cuda")]` neuron/TP
+   paths do NOT compile locally. The branch CI's **CUDA type-check** job is
+   the only thing that validates them, so for any neuron change the push to
+   Gitea is the real gate, not a rubber stamp.
+3. Push the branch on local-green (no need to ask first), and background-watch
+   its CI run via the gitea-mcp `actions_run_read` tools. Start the next piece
+   of work meanwhile.
+4. Merge to `main` when the four **validation** jobs are green — Format,
+   Clippy, Test, CUDA type-check. The SRPM / COPR / version-bump jobs are the
+   deploy pipeline (they run on `main`), not validation — don't wait on them.
+5. Merging/pushing to `main` triggers the auto-deploy pipeline.
+
+Docs-only changes (no `#[cfg(feature = "cuda")]` impact) can go straight to
+`main` — there's nothing for the CUDA type-check to prove.
+
+SSH note: the gitea remote host offers multiple agent keys and cuts the
+connection before reaching the right one. This repo pins the working key via
+`git config core.sshCommand "ssh -i ~/.ssh/id_grenade -o IdentitiesOnly=yes"`.
+
 ## Environment
 
 - Targets Fedora 43 (systemd, SELinux enforcing)

Cargo.toml

@@ -7,6 +7,9 @@ members = [
     "crates/neuron",
     "crates/helexa-acp",
     "crates/helexa-bench",
+    "crates/helexa-router",
+    "crates/helexa-stream",
+    "crates/helexa-upstream",
 ]
 
 [workspace.package]

asset/helexa-bench/bob.toml

@@ -0,0 +1,67 @@
+# helexa-bench config for bob.hanzalova.internal.
+#
+# Synced to /etc/helexa-bench/helexa-bench.toml by script/infra-setup.sh
+# (the helexa-bench RPM ships helexa-bench.example.toml as a
+# %config(noreplace) default; this per-host file overrides it).
+#
+# bob is a client host (it also runs Agent Zero); helexa-bench here hits
+# every neuron on the fleet directly and records build-stamped results
+# into the local SQLite store.
+
+[bench]
+sweep_interval_secs = 1800
+samples_per_version = 5
+iteration_pause_secs = 2
+request_timeout_secs = 600
+db_path = "/var/lib/helexa-bench/bench.sqlite"
+
+[scenarios]
+prompt_sizes = [128, 4096]
+max_tokens = 256
+# Concurrency / agentic-load scenarios (#89), enabled so the 27B baseline
+# carries p95-under-concurrency data before the F3 A/B gate (#94) needs
+# it for comparison. Levels mirror the real a0/hermes/opencode fan-out.
+concurrency_levels = [2, 4, 8]
+concurrency_prompt_tokens = 512
+
+# Capability probes (#91) — the reasoning/planning axis the speed
+# scenarios miss; scored manually via `helexa-bench score` (O7). Enabled
+# for the same reason: the F3 gate compares 80B-A3B variants against the
+# 27B on planning quality, so the 27B needs scored artifacts first.
+[[scenarios.capability_probes]]
+name = "rust-plan"
+max_tokens = 4096
+prompt = """
+Write an implementation plan for adding rate limiting to an Axum service.
+Honor existing conventions, call out trade-offs, and sequence the work.
+"""
+
+[[scenarios.capability_probes]]
+name = "debug-reason"
+max_tokens = 4096
+prompt = """
+A Rust axum server streams SSE responses through a reverse proxy. Clients
+report that streams stall for exactly 60 seconds and then resume, but only
+when response chunks are small and infrequent. Curling the backend directly
+never stalls. List the most likely causes in order of probability, explain
+the mechanism behind each, and describe the smallest experiment that would
+confirm or eliminate each cause.
+"""
+
+# Read-only JSON API consumed by the bench UI (hosted separately) and for
+# programmatic access. Served alongside the sweep loop.
+[api]
+enabled = true
+listen = "0.0.0.0:13132"
+
+[[targets]]
+name = "beast"
+endpoint = "http://beast.hanzalova.internal:13131"
+
+[[targets]]
+name = "benjy"
+endpoint = "http://benjy.hanzalova.internal:13131"
+
+[[targets]]
+name = "quadbrat"
+endpoint = "http://quadbrat.hanzalova.internal:13131"

asset/nginx/bench.helexa.ai.bootstrap.conf

@@ -0,0 +1,15 @@
+# Bootstrap vhost for bench.helexa.ai — http-only, used ONLY to obtain
+# the initial Let's Encrypt cert via the webroot challenge (the full TLS
+# vhost can't load before the cert file exists). script/infra-setup.sh
+# installs this, runs certbot, then swaps in bench.helexa.ai.conf.
+server {
+    listen 80;
+    server_name bench.helexa.ai;
+
+    location /.well-known/acme-challenge/ {
+        root /var/www/bench.helexa.ai;
+    }
+    location / {
+        try_files $uri $uri/ =404;
+    }
+}

asset/nginx/bench.helexa.ai.conf

@@ -0,0 +1,56 @@
+# Public, auth-less bench UI at https://bench.helexa.ai.
+#
+# Serves the static SPA from /var/www/bench.helexa.ai (rsynced by
+# .gitea/workflows/deploy.yml's deploy-bench-ui job) and reverse-proxies
+# /api to the helexa-bench read API on bob over the WireGuard mesh — so
+# the browser stays same-origin (no CORS) and the internal API never
+# needs to be exposed publicly.
+#
+# TLS via Let's Encrypt; the cert is obtained/renewed by certbot
+# (bootstrapped one-time in script/infra-setup.sh). Mirrors the
+# dev.swym.hanzalova.internal vhost convention on this host.
+
+server {
+    listen 80;
+    server_name bench.helexa.ai;
+
+    # Keep serving the ACME webroot so certbot can renew.
+    location /.well-known/acme-challenge/ {
+        root /var/www/bench.helexa.ai;
+    }
+    location / {
+        return 301 https://$host$request_uri;
+    }
+}
+
+server {
+    listen 443 ssl;
+    http2 on;
+    server_name bench.helexa.ai;
+
+    ssl_certificate     /etc/letsencrypt/live/bench.helexa.ai/fullchain.pem;
+    ssl_certificate_key /etc/letsencrypt/live/bench.helexa.ai/privkey.pem;
+    ssl_protocols TLSv1.2 TLSv1.3;
+    ssl_ciphers HIGH:!aNULL:!MD5;
+    ssl_prefer_server_ciphers on;
+    ssl_session_cache shared:SSL:10m;
+
+    root /var/www/bench.helexa.ai;
+    index index.html;
+
+    # Bench read API on bob (internal WireGuard); browser stays same-origin.
+    location /api/ {
+        proxy_pass http://bob.hanzalova.internal:13132;
+        proxy_http_version 1.1;
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+        proxy_set_header X-Forwarded-Proto $scheme;
+        proxy_read_timeout 60s;
+    }
+
+    # SPA fallback — client-side routes (/trends, /runs) resolve to index.html.
+    location / {
+        try_files $uri $uri/ /index.html;
+    }
+}

asset/nginx/bench.internal.conf

@@ -0,0 +1,34 @@
+# Internal bench UI vhost — https://bench.internal, reachable from inside
+# the WireGuard mesh (the public bench.helexa.ai dead-ends at the OPNsense
+# LAN interface, which only port-forwards :443 from the WAN). Same SPA +
+# /api→bob proxy as bench.helexa.ai, but with an internal-CA cert
+# (smallstep "lair", renewed by step@bench.timer). Mirrors the
+# *.internal vhost convention on oolon.kosherinata.internal.
+server {
+    server_name bench.internal;
+    listen 443 ssl;
+    http2 on;
+
+    ssl_certificate /etc/nginx/tls/cert/bench.internal.pem;
+    ssl_certificate_key /etc/nginx/tls/key/bench.internal.pem;
+    ssl_trusted_certificate /etc/pki/ca-trust/source/anchors/root-internal.pem;
+    ssl_protocols TLSv1.3;
+
+    # Shared webroot with the public vhost — same built SPA.
+    root /var/www/bench.helexa.ai;
+    index index.html;
+
+    location /api/ {
+        proxy_pass http://bob.hanzalova.internal:13132;
+        proxy_http_version 1.1;
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+        proxy_set_header X-Forwarded-Proto $scheme;
+        proxy_read_timeout 60s;
+    }
+
+    location / {
+        try_files $uri $uri/ /index.html;
+    }
+}

asset/sudoers.d/bench-host.conf

@@ -0,0 +1,25 @@
+# Install on the bench host (bob) as /etc/sudoers.d/helexa_gitea_ci
+# (owner root:root, mode 0440). Required by .gitea/workflows/deploy.yml,
+# which SSHes as gitea_ci@bob to roll out helexa-bench package upgrades
+# and config changes.
+#
+# Filename convention `helexa_gitea_ci` (vs bare `gitea_ci`) so other
+# helexa-org apps can drop their own sudoers files on the same host
+# without overwriting this one.
+#
+# helexa-bench polls the neuron fleet (outbound) and serves a read-only
+# JSON API on tcp/13132 for the bench UI — hence the firewall-cmd grants.
+
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/rsync * /etc/helexa-bench/helexa-bench.toml
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl start helexa-bench.service
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl stop helexa-bench.service
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl enable --now helexa-bench.service
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl daemon-reload
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/dnf install --refresh --allowerasing -y helexa-bench
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/dnf upgrade --refresh --allowerasing -y helexa-bench
+# sudoers reserves `:` and `=` and requires `\` escaping inside command
+# arguments — without it visudo errors at the first `:` in `https://`.
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/dnf config-manager addrepo --from-repofile\=https\://rpm.lair.cafe/lair-cafe-unstable.repo
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/dnf config-manager setopt lair-cafe-unstable.enabled\=1
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/firewall-cmd --add-service=helexa-bench --permanent
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/firewall-cmd --reload

asset/sudoers.d/cortex-host.conf

@@ -9,8 +9,11 @@
 
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/rsync * /etc/cortex/cortex.toml
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/rsync * /etc/cortex/models.toml
+# deploy-bench-ui rsyncs the built bench SPA into the nginx webroot.
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/rsync * /var/www/bench.helexa.ai/
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl start cortex.service
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl stop cortex.service
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl enable --now cortex.service
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl daemon-reload
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/dnf install --refresh --allowerasing -y cortex
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/dnf upgrade --refresh --allowerasing -y cortex

asset/sudoers.d/neuron-host.conf

@@ -14,8 +14,13 @@
 # flavour installed" — vandalism, not privilege escalation.
 
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/rsync * /etc/neuron/neuron.toml
+# deploy.yml writes the per-model systemd drop-in carrying
+# NEURON_MAX_PROMPT_TOKENS: gitea_ci stages it in its own dir, then
+# installs it root-owned. Exact source/dest paths; see doc/context-limits.md.
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/install -o root -g root -m 0644 -D /var/lib/gitea_ci/model.conf /etc/systemd/system/neuron.service.d/model.conf
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl start neuron.service
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl stop neuron.service
+gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl enable --now neuron.service
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/systemctl daemon-reload
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/dnf install --refresh --allowerasing -y helexa-neuron-ampere
 gitea_ci ALL=(root) NOPASSWD: /usr/bin/dnf upgrade --refresh --allowerasing -y helexa-neuron-ampere

asset/systemd/step@.service

@@ -0,0 +1,20 @@
+# Internal-CA cert renewal for %i.internal, driven by step@%i.timer.
+# Replicated from oolon.kosherinata.internal (the kosherinata DC proxy).
+# Renews an EXISTING cert via mTLS (step ca renew) — the initial cert
+# must be issued once with a provisioner (see script/infra-setup.sh).
+# Installed to /etc/systemd/system/step@.service.
+[Unit]
+Description=step cert renew for %i.internal
+Documentation=https://smallstep.com/docs/step-ca/renewal
+
+[Service]
+Type=oneshot
+ExecCondition=/usr/bin/step certificate needs-renewal \
+    /etc/nginx/tls/cert/%i.internal.pem
+ExecStart=/usr/bin/step ca renew \
+    --force \
+    --ca-url https://ca.internal \
+    --root /etc/pki/ca-trust/source/anchors/root-internal.pem \
+    /etc/nginx/tls/cert/%i.internal.pem \
+    /etc/nginx/tls/key/%i.internal.pem
+ExecStartPost=/usr/bin/systemctl reload nginx.service

asset/systemd/step@.timer

@@ -0,0 +1,15 @@
+# Periodic internal-cert renewal for %i.internal (every 15 min, jittered).
+# Replicated from oolon.kosherinata.internal. Installed to
+# /etc/systemd/system/step@.timer; enable per-cert with
+# `systemctl enable --now step@bench.timer`.
+[Unit]
+Description=step cert renew timer for %i.internal
+
+[Timer]
+Persistent=true
+OnCalendar=*:1/15
+AccuracySec=1us
+RandomizedDelaySec=5m
+
+[Install]
+WantedBy=timers.target

bench/.gitignore

@@ -0,0 +1,3 @@
+node_modules
+dist
+*.local

bench/README.md

@@ -0,0 +1,45 @@
+# helexa bench UI
+
+A Vite + React (SWC, TypeScript) app that visualises the fleet benchmark
+data collected by `helexa-bench`. It reads the read-only JSON API the
+bench daemon serves (`crates/helexa-bench/src/api.rs`, default
+`:13132` on bob).
+
+Stack: React Router, react-bootstrap, Recharts.
+
+## Pages
+
+- **Overview** — latest median results per (host, model, scenario) cell.
+- **Trends** — decode-tok/s and TTFT plotted across neuron build SHAs as
+  releases roll out (the headline view). Pick host / model / scenario.
+- **Runs** — filterable raw-run explorer.
+
+## Develop
+
+```sh
+cd bench
+npm install
+npm run dev      # http://localhost:5173
+```
+
+`vite.config.ts` proxies `/api` → `http://bob.hanzalova.internal:13132`,
+so the dev server talks to the live bench API with no CORS fuss. Point
+the proxy elsewhere (or run a local `helexa-bench serve`) to develop
+against other data.
+
+## Production hosting
+
+Public at **https://bench.helexa.ai** — nginx on the gateway
+(`hanzalova.internal`) serves the static `dist/` and reverse-proxies
+`/api` to the bench API on bob over WireGuard, so the SPA is same-origin
+(no CORS) and the internal API stays off the public internet.
+
+- `npm run build` is run with **no** `VITE_API_BASE` (the app calls
+  `/api/...` on its own origin; nginx proxies it to bob).
+- `.gitea/workflows/deploy.yml` (`deploy-bench-ui`) builds and rsyncs
+  `dist/` to `/var/www/bench.helexa.ai` on every deploy.
+- The nginx vhost (`asset/nginx/bench.helexa.ai.conf`) and the
+  Let's Encrypt cert are one-time host setup in `script/infra-setup.sh`.
+
+To host elsewhere instead, build with
+`VITE_API_BASE=<bob-api-origin>` and serve the static `dist/`.

bench/index.html

@@ -0,0 +1,12 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="UTF-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>helexa bench</title>
+  </head>
+  <body>
+    <div id="root"></div>
+    <script type="module" src="/src/main.tsx"></script>
+  </body>
+</html>

bench/package.json

@@ -0,0 +1,28 @@
+{
+  "name": "helexa-bench-ui",
+  "private": true,
+  "version": "0.1.0",
+  "type": "module",
+  "description": "Visualisation app for helexa-bench fleet benchmark data.",
+  "scripts": {
+    "dev": "vite",
+    "build": "tsc && vite build",
+    "preview": "vite preview"
+  },
+  "dependencies": {
+    "bootstrap": "^5.3.3",
+    "react": "^18.3.1",
+    "react-bootstrap": "^2.10.5",
+    "react-dom": "^18.3.1",
+    "react-router-dom": "^6.26.2",
+    "recharts": "^2.12.7"
+  },
+  "devDependencies": {
+    "@types/node": "^20.14.0",
+    "@types/react": "^18.3.5",
+    "@types/react-dom": "^18.3.0",
+    "@vitejs/plugin-react-swc": "^3.7.0",
+    "typescript": "^5.5.4",
+    "vite": "^5.4.0"
+  }
+}

bench/src/App.tsx

@@ -0,0 +1,30 @@
+import { Container, Nav, Navbar } from "react-bootstrap";
+import { NavLink, Outlet } from "react-router-dom";
+
+export default function App() {
+  return (
+    <>
+      <Navbar bg="dark" variant="dark" expand="md">
+        <Container>
+          <Navbar.Brand as={NavLink} to="/">
+            helexa&nbsp;bench
+          </Navbar.Brand>
+          <Nav className="me-auto">
+            <Nav.Link as={NavLink} to="/" end>
+              Overview
+            </Nav.Link>
+            <Nav.Link as={NavLink} to="/trends">
+              Trends
+            </Nav.Link>
+            <Nav.Link as={NavLink} to="/runs">
+              Runs
+            </Nav.Link>
+          </Nav>
+        </Container>
+      </Navbar>
+      <Container className="py-4">
+        <Outlet />
+      </Container>
+    </>
+  );
+}

bench/src/api.ts

@@ -0,0 +1,45 @@
+import type { Dimensions, ReportRow, RunRow, SeriesPoint } from "./types";
+
+// Empty default → `fetch('/api/...')` hits the dev proxy (vite.config.ts)
+// or the same origin. For a separately-hosted build, set VITE_API_BASE to
+// the bob API origin (e.g. http://bob.hanzalova.internal:13132).
+const BASE = import.meta.env.VITE_API_BASE ?? "";
+
+async function getJson<T>(path: string): Promise<T> {
+  const res = await fetch(`${BASE}${path}`);
+  if (!res.ok) {
+    throw new Error(`${res.status} ${res.statusText}: ${await res.text()}`);
+  }
+  return res.json() as Promise<T>;
+}
+
+export const getDimensions = () => getJson<Dimensions>("/api/dimensions");
+export const getSummary = () => getJson<ReportRow[]>("/api/summary");
+
+// host is resolved server-side (each model maps to one host today), so the
+// public UI selects by model + scenario alone.
+export const getSeries = (model: string, scenario: string) =>
+  getJson<SeriesPoint[]>(
+    `/api/series?model=${encodeURIComponent(model)}&scenario=${encodeURIComponent(scenario)}`,
+  );
+
+export interface RunsParams {
+  host?: string;
+  model?: string;
+  scenario?: string;
+  sha?: string;
+  ok?: boolean;
+  limit?: number;
+}
+
+export const getRuns = (p: RunsParams = {}) => {
+  const q = new URLSearchParams();
+  if (p.host) q.set("host", p.host);
+  if (p.model) q.set("model", p.model);
+  if (p.scenario) q.set("scenario", p.scenario);
+  if (p.sha) q.set("sha", p.sha);
+  if (p.ok !== undefined) q.set("ok", String(p.ok));
+  if (p.limit) q.set("limit", String(p.limit));
+  const qs = q.toString();
+  return getJson<RunRow[]>(`/api/runs${qs ? `?${qs}` : ""}`);
+};

bench/src/baseline.ts

@@ -0,0 +1,52 @@
+// Pre-helexa-bench baseline, transcribed verbatim from doc/benchmarks.md.
+//
+// IMPORTANT — different measurement regime. These were measured by
+// script/bench.py *through the cortex gateway* (so TTFT/total include a
+// proxy hop), reported as medians only, before helexa-bench existed.
+// helexa-bench measures each neuron *directly*. So these points are an
+// honest historical anchor, NOT apples-to-apples with the live series —
+// the Trends view renders them dashed + labelled, never merged into the
+// live line.
+//
+// Host is inferred from the model via the doc's Fleet table
+// (beast=27B, benjy=8B, quadbrat=1.7B). Timestamps are the two 2026-06-12
+// snapshots in the doc, ordered (08:00 = pre-#11, 16:00 = post-#11) so
+// they sort before the bench era on the shared time axis.
+
+export interface BaselinePoint {
+  host: string;
+  model: string;
+  scenario: string;
+  git_sha: string;
+  build_timestamp: string;
+  ttft_s: number;
+  decode_tps: number;
+  total_s: number;
+}
+
+/** Source: bench.py via cortex gateway — see doc/benchmarks.md. */
+export const BASELINE_SOURCE = "bench.py · via cortex gateway";
+
+export const BASELINE: BaselinePoint[] = [
+  // ── 8f6f1d3 — baseline (2026-06-12) ────────────────────────────────
+  { host: "beast", model: "Qwen/Qwen3.6-27B", scenario: "chat:128", git_sha: "8f6f1d3", build_timestamp: "2026-06-12T08:00:00Z", ttft_s: 1.658, decode_tps: 35.0, total_s: 8.981 },
+  { host: "beast", model: "Qwen/Qwen3.6-27B", scenario: "chat:4096", git_sha: "8f6f1d3", build_timestamp: "2026-06-12T08:00:00Z", ttft_s: 7.067, decode_tps: 33.7, total_s: 14.63 },
+  { host: "benjy", model: "Qwen/Qwen3-8B", scenario: "chat:128", git_sha: "8f6f1d3", build_timestamp: "2026-06-12T08:00:00Z", ttft_s: 0.884, decode_tps: 62.4, total_s: 4.938 },
+  { host: "benjy", model: "Qwen/Qwen3-8B", scenario: "chat:4096", git_sha: "8f6f1d3", build_timestamp: "2026-06-12T08:00:00Z", ttft_s: 1.818, decode_tps: 46.5, total_s: 7.27 },
+  { host: "quadbrat", model: "Qwen/Qwen3-1.7B", scenario: "chat:128", git_sha: "8f6f1d3", build_timestamp: "2026-06-12T08:00:00Z", ttft_s: 0.685, decode_tps: 81.3, total_s: 3.741 },
+  { host: "quadbrat", model: "Qwen/Qwen3-1.7B", scenario: "chat:4096", git_sha: "8f6f1d3", build_timestamp: "2026-06-12T08:00:00Z", ttft_s: 2.743, decode_tps: 35.4, total_s: 9.884 },
+  // ── a1952a4 — post prefix-KV-cache (#11, 2026-06-12) ───────────────
+  { host: "beast", model: "Qwen/Qwen3.6-27B", scenario: "chat:128", git_sha: "a1952a4", build_timestamp: "2026-06-12T16:00:00Z", ttft_s: 1.355, decode_tps: 45.8, total_s: 4.147 },
+  { host: "beast", model: "Qwen/Qwen3.6-27B", scenario: "chat:4096", git_sha: "a1952a4", build_timestamp: "2026-06-12T16:00:00Z", ttft_s: 1.431, decode_tps: 43.3, total_s: 4.387 },
+  { host: "benjy", model: "Qwen/Qwen3-8B", scenario: "chat:128", git_sha: "a1952a4", build_timestamp: "2026-06-12T16:00:00Z", ttft_s: 0.886, decode_tps: 78.6, total_s: 2.478 },
+  { host: "benjy", model: "Qwen/Qwen3-8B", scenario: "chat:4096", git_sha: "a1952a4", build_timestamp: "2026-06-12T16:00:00Z", ttft_s: 1.824, decode_tps: 58.3, total_s: 3.969 },
+  { host: "quadbrat", model: "Qwen/Qwen3-1.7B", scenario: "chat:128", git_sha: "a1952a4", build_timestamp: "2026-06-12T16:00:00Z", ttft_s: 0.702, decode_tps: 104.8, total_s: 1.895 },
+  { host: "quadbrat", model: "Qwen/Qwen3-1.7B", scenario: "chat:4096", git_sha: "a1952a4", build_timestamp: "2026-06-12T16:00:00Z", ttft_s: 2.749, decode_tps: 44.9, total_s: 5.534 },
+];
+
+/** Baseline points for one (model, scenario) cell, oldest first. */
+export function baselineFor(model: string, scenario: string): BaselinePoint[] {
+  return BASELINE.filter(
+    (b) => b.model === model && b.scenario === scenario,
+  ).sort((a, b) => a.build_timestamp.localeCompare(b.build_timestamp));
+}

bench/src/main.tsx

@@ -0,0 +1,22 @@
+import React from "react";
+import ReactDOM from "react-dom/client";
+import { BrowserRouter, Route, Routes } from "react-router-dom";
+import "bootstrap/dist/css/bootstrap.min.css";
+import App from "./App";
+import Overview from "./pages/Overview";
+import Trends from "./pages/Trends";
+import Runs from "./pages/Runs";
+
+ReactDOM.createRoot(document.getElementById("root")!).render(
+  <React.StrictMode>
+    <BrowserRouter>
+      <Routes>
+        <Route path="/" element={<App />}>
+          <Route index element={<Overview />} />
+          <Route path="trends" element={<Trends />} />
+          <Route path="runs" element={<Runs />} />
+        </Route>
+      </Routes>
+    </BrowserRouter>
+  </React.StrictMode>,
+);

bench/src/pages/Overview.tsx

@@ -0,0 +1,64 @@
+import { useEffect, useState } from "react";
+import { Alert, Spinner, Table } from "react-bootstrap";
+import { getSummary } from "../api";
+import type { ReportRow } from "../types";
+
+const f = (n: number | null, p = 2) => (n == null ? "—" : n.toFixed(p));
+
+export default function Overview() {
+  const [rows, setRows] = useState<ReportRow[]>([]);
+  const [err, setErr] = useState<string | null>(null);
+  const [loading, setLoading] = useState(true);
+
+  useEffect(() => {
+    getSummary()
+      .then(setRows)
+      .catch((e) => setErr(String(e)))
+      .finally(() => setLoading(false));
+  }, []);
+
+  if (loading) return <Spinner animation="border" />;
+  if (err) return <Alert variant="danger">{err}</Alert>;
+
+  return (
+    <>
+      <h3 className="mb-3">Latest results per cell</h3>
+      <p className="text-muted">
+        Median of each cell's samples on the most recent build seen for that
+        (host, model, scenario).
+      </p>
+      <Table striped bordered hover responsive size="sm">
+        <thead>
+          <tr>
+            <th>GPU</th>
+            <th>model</th>
+            <th className="text-end">prompt tok</th>
+            <th className="text-end">TTFT (s)</th>
+            <th className="text-end">decode tok/s</th>
+            <th className="text-end">total (s)</th>
+            <th>build</th>
+            <th className="text-end">n</th>
+          </tr>
+        </thead>
+        <tbody>
+          {rows.map((r, i) => (
+            <tr key={i}>
+              <td>{r.gpu ?? r.target_name}</td>
+              <td>{r.model_id}</td>
+              <td className="text-end">
+                {r.prompt_tokens ?? `~${r.prompt_size_approx}`}
+              </td>
+              <td className="text-end">{f(r.ttft_s_median, 3)}</td>
+              <td className="text-end">{f(r.decode_tps_median, 1)}</td>
+              <td className="text-end">{f(r.total_s_median, 3)}</td>
+              <td>
+                <code>{r.git_sha}</code>
+              </td>
+              <td className="text-end">{r.samples}</td>
+            </tr>
+          ))}
+        </tbody>
+      </Table>
+    </>
+  );
+}

bench/src/pages/Runs.tsx

@@ -0,0 +1,141 @@
+import { useEffect, useState } from "react";
+import { Alert, Badge, Col, Form, Row, Spinner, Table } from "react-bootstrap";
+import { getDimensions, getRuns } from "../api";
+import type { Dimensions, RunRow } from "../types";
+
+const f = (n: number | null, p = 2) => (n == null ? "—" : n.toFixed(p));
+
+function Picker({
+  label,
+  value,
+  set,
+  options,
+}: {
+  label: string;
+  value: string;
+  set: (v: string) => void;
+  options: string[];
+}) {
+  return (
+    <Form.Group as={Col}>
+      <Form.Label>{label}</Form.Label>
+      <Form.Select value={value} onChange={(e) => set(e.target.value)}>
+        <option value="">(all)</option>
+        {options.map((o) => (
+          <option key={o} value={o}>
+            {o}
+          </option>
+        ))}
+      </Form.Select>
+    </Form.Group>
+  );
+}
+
+export default function Runs() {
+  const [dims, setDims] = useState<Dimensions | null>(null);
+  const [host, setHost] = useState("");
+  const [model, setModel] = useState("");
+  const [scenario, setScenario] = useState("");
+  const [rows, setRows] = useState<RunRow[]>([]);
+  const [err, setErr] = useState<string | null>(null);
+  const [loading, setLoading] = useState(false);
+
+  useEffect(() => {
+    getDimensions()
+      .then(setDims)
+      .catch((e) => setErr(String(e)));
+  }, []);
+
+  useEffect(() => {
+    setLoading(true);
+    getRuns({
+      host: host || undefined,
+      model: model || undefined,
+      scenario: scenario || undefined,
+      limit: 200,
+    })
+      .then(setRows)
+      .catch((e) => setErr(String(e)))
+      .finally(() => setLoading(false));
+  }, [host, model, scenario]);
+
+  if (err) return <Alert variant="danger">{err}</Alert>;
+
+  return (
+    <>
+      <h3 className="mb-3">Runs</h3>
+      {dims && (
+        <Row className="g-3 mb-3">
+          {/* GPU filter — labelled by GPU, but filters by the underlying host. */}
+          <Form.Group as={Col}>
+            <Form.Label>GPU</Form.Label>
+            <Form.Select value={host} onChange={(e) => setHost(e.target.value)}>
+              <option value="">(all)</option>
+              {dims.hosts.map((h) => (
+                <option key={h} value={h}>
+                  {dims.host_gpus[h] ?? h}
+                </option>
+              ))}
+            </Form.Select>
+          </Form.Group>
+          <Picker
+            label="Model"
+            value={model}
+            set={setModel}
+            options={dims.models}
+          />
+          <Picker
+            label="Scenario"
+            value={scenario}
+            set={setScenario}
+            options={dims.scenarios}
+          />
+        </Row>
+      )}
+      {loading ? (
+        <Spinner animation="border" />
+      ) : (
+        <Table striped bordered hover responsive size="sm">
+          <thead>
+            <tr>
+              <th>ts</th>
+              <th>GPU</th>
+              <th>model</th>
+              <th>scenario</th>
+              <th>build</th>
+              <th className="text-end">TTFT</th>
+              <th className="text-end">tok/s</th>
+              <th className="text-end">total</th>
+              <th>ok</th>
+            </tr>
+          </thead>
+          <tbody>
+            {rows.map((r) => (
+              <tr key={r.id}>
+                <td>{r.ts}</td>
+                <td>{r.gpu ?? r.host}</td>
+                <td>{r.model_id}</td>
+                <td>{r.scenario_id}</td>
+                <td>
+                  <code>{r.git_sha}</code>
+                </td>
+                <td className="text-end">{f(r.ttft_s, 3)}</td>
+                <td className="text-end">{f(r.decode_tps, 1)}</td>
+                <td className="text-end">{f(r.total_s, 3)}</td>
+                <td>
+                  {r.ok ? (
+                    <Badge bg="success">ok</Badge>
+                  ) : (
+                    <Badge bg="danger" title={r.error ?? ""}>
+                      fail
+                    </Badge>
+                  )}
+                </td>
+              </tr>
+            ))}
+          </tbody>
+        </Table>
+      )}
+    </>
+  );
+}

bench/src/pages/Trends.tsx

@@ -0,0 +1,221 @@
+import { useEffect, useMemo, useState } from "react";
+import { Alert, Col, Form, Row, Spinner } from "react-bootstrap";
+import {
+  CartesianGrid,
+  Legend,
+  Line,
+  LineChart,
+  ReferenceLine,
+  ResponsiveContainer,
+  Tooltip,
+  XAxis,
+  YAxis,
+} from "recharts";
+import { getDimensions, getSeries } from "../api";
+import type { Dimensions, SeriesPoint } from "../types";
+import { BASELINE_SOURCE, baselineFor } from "../baseline";
+
+function Picker({
+  label,
+  value,
+  set,
+  options,
+}: {
+  label: string;
+  value: string;
+  set: (v: string) => void;
+  options: string[];
+}) {
+  return (
+    <Form.Group as={Col}>
+      <Form.Label>{label}</Form.Label>
+      <Form.Select value={value} onChange={(e) => set(e.target.value)}>
+        {options.map((o) => (
+          <option key={o} value={o}>
+            {o}
+          </option>
+        ))}
+      </Form.Select>
+    </Form.Group>
+  );
+}
+
+export default function Trends() {
+  const [dims, setDims] = useState<Dimensions | null>(null);
+  const [model, setModel] = useState("");
+  const [scenario, setScenario] = useState("");
+  const [series, setSeries] = useState<SeriesPoint[]>([]);
+  const [err, setErr] = useState<string | null>(null);
+
+  useEffect(() => {
+    getDimensions()
+      .then((d) => {
+        setDims(d);
+        if (d.models[0]) setModel(d.models[0]);
+        if (d.scenarios[0]) setScenario(d.scenarios[0]);
+      })
+      .catch((e) => setErr(String(e)));
+  }, []);
+
+  useEffect(() => {
+    if (model && scenario) {
+      getSeries(model, scenario)
+        .then(setSeries)
+        .catch((e) => setErr(String(e)));
+    }
+  }, [model, scenario]);
+
+  // Prepend the pre-helexa-bench baseline (dashed, separate keys) so it
+  // anchors the timeline without being merged into the live line. Different
+  // measurement regime — see baseline.ts / doc/benchmarks.md.
+  const base = useMemo(
+    () => baselineFor(model, scenario),
+    [model, scenario],
+  );
+  const data = useMemo(
+    () => [
+      ...base.map((p) => ({
+        label: p.git_sha,
+        baseTtft: p.ttft_s,
+        baseDecode: p.decode_tps,
+        baseTotal: p.total_s,
+      })),
+      ...series.map((p) => ({
+        label: p.git_sha,
+        ttft: p.ttft_s_median,
+        decode: p.decode_tps_median,
+        total: p.total_s_median,
+      })),
+    ],
+    [series, base],
+  );
+
+  // Divider marking the boundary between the two regimes (drawn at the
+  // first live build, with baseline points to its left).
+  const firstLive = series[0]?.git_sha;
+  const showDivider = base.length > 0 && series.length > 0;
+
+  if (err) return <Alert variant="danger">{err}</Alert>;
+  if (!dims) return <Spinner animation="border" />;
+
+  return (
+    <>
+      <h3 className="mb-3">Trends over builds</h3>
+      <Row className="g-3 mb-4">
+        <Picker
+          label="Model"
+          value={model}
+          set={setModel}
+          options={dims.models}
+        />
+        <Picker
+          label="Scenario"
+          value={scenario}
+          set={setScenario}
+          options={dims.scenarios}
+        />
+      </Row>
+
+      {dims.model_gpus[model] && (
+        <p className="text-muted mb-3">
+          Measured on <strong>{dims.model_gpus[model]}</strong>.
+        </p>
+      )}
+
+      {data.length === 0 ? (
+        <Alert variant="info">No data for this selection yet.</Alert>
+      ) : (
+        <>
+          {base.length > 0 && (
+            <p className="text-muted small mb-3">
+              Dashed = pre-helexa-bench baseline ({BASELINE_SOURCE}); solid =
+              helexa-bench (direct to neuron). Different measurement regimes —
+              see <code>doc/benchmarks.md</code>.
+            </p>
+          )}
+          <h5 className="mt-3">decode tok/s (higher is better)</h5>
+          <ResponsiveContainer width="100%" height={280}>
+            <LineChart data={data} margin={{ top: 8, right: 24, bottom: 8, left: 0 }}>
+              <CartesianGrid strokeDasharray="3 3" />
+              <XAxis dataKey="label" />
+              <YAxis />
+              <Tooltip />
+              <Legend />
+              {showDivider && firstLive && (
+                <ReferenceLine
+                  x={firstLive}
+                  stroke="#bbb"
+                  strokeDasharray="3 3"
+                  label={{
+                    value: "bench.py → helexa-bench",
+                    position: "top",
+                    fill: "#999",
+                    fontSize: 11,
+                  }}
+                />
+              )}
+              <Line
+                type="monotone"
+                dataKey="decode"
+                name="decode tok/s"
+                stroke="#0d6efd"
+                connectNulls
+              />
+              {base.length > 0 && (
+                <Line
+                  type="monotone"
+                  dataKey="baseDecode"
+                  name="baseline (bench.py · gateway)"
+                  stroke="#888"
+                  strokeDasharray="5 5"
+                  connectNulls
+                />
+              )}
+            </LineChart>
+          </ResponsiveContainer>
+
+          <h5 className="mt-4">TTFT seconds (lower is better)</h5>
+          <ResponsiveContainer width="100%" height={280}>
+            <LineChart data={data} margin={{ top: 8, right: 24, bottom: 8, left: 0 }}>
+              <CartesianGrid strokeDasharray="3 3" />
+              <XAxis dataKey="label" />
+              <YAxis />
+              <Tooltip />
+              <Legend />
+              {showDivider && firstLive && (
+                <ReferenceLine
+                  x={firstLive}
+                  stroke="#bbb"
+                  strokeDasharray="3 3"
+                  label={{
+                    value: "bench.py → helexa-bench",
+                    position: "top",
+                    fill: "#999",
+                    fontSize: 11,
+                  }}
+                />
+              )}
+              <Line
+                type="monotone"
+                dataKey="ttft"
+                name="TTFT (s)"
+                stroke="#dc3545"
+                connectNulls
+              />
+              {base.length > 0 && (
+                <Line
+                  type="monotone"
+                  dataKey="baseTtft"
+                  name="baseline (bench.py · gateway)"
+                  stroke="#888"
+                  strokeDasharray="5 5"
+                  connectNulls
+                />
+              )}
+            </LineChart>
+          </ResponsiveContainer>
+        </>
+      )}
+    </>
+  );
+}

bench/src/types.ts

@@ -0,0 +1,69 @@
+// Mirrors the JSON served by helexa-bench's read API (crates/helexa-bench/src/api.rs).
+
+export interface BuildRef {
+  git_sha: string;
+  build_timestamp: string | null;
+  package_version: string | null;
+}
+
+export interface Dimensions {
+  hosts: string[];
+  models: string[];
+  scenarios: string[];
+  builds: BuildRef[];
+  /** host → GPU label, e.g. "2× RTX 5090". */
+  host_gpus: Record<string, string>;
+  /** model → GPU label (model maps to one host today). */
+  model_gpus: Record<string, string>;
+}
+
+/** Latest-SHA-per-cell medians (the report table). */
+export interface ReportRow {
+  target_name: string;
+  model_id: string;
+  scenario_id: string;
+  prompt_size_approx: number;
+  git_sha: string;
+  prompt_tokens: number | null;
+  ttft_s_median: number | null;
+  decode_tps_median: number | null;
+  total_s_median: number | null;
+  samples: number;
+  /** Public-facing resource name (the host's GPU(s)). */
+  gpu: string | null;
+}
+
+/** One point in a per-build time-series for a (host, model, scenario) cell. */
+export interface SeriesPoint {
+  git_sha: string;
+  build_timestamp: string | null;
+  package_version: string | null;
+  ttft_s_median: number | null;
+  decode_tps_median: number | null;
+  total_s_median: number | null;
+  samples: number;
+}
+
+export interface RunRow {
+  id: number;
+  ts: string;
+  host: string;
+  /** Public-facing resource name (the host's GPU(s)). */
+  gpu: string | null;
+  hostname: string | null;
+  git_sha: string;
+  build_timestamp: string | null;
+  package_version: string;
+  model_id: string;
+  harness: string;
+  scenario_id: string;
+  prompt_size_approx: number;
+  prompt_tokens_actual: number | null;
+  max_tokens: number;
+  ttft_s: number | null;
+  decode_tps: number | null;
+  total_s: number | null;
+  completion_tokens: number | null;
+  ok: boolean;
+  error: string | null;
+}

bench/src/vite-env.d.ts

@@ -0,0 +1,9 @@
+/// <reference types="vite/client" />
+
+interface ImportMetaEnv {
+  /** Base origin of the bench API. Empty → use the dev proxy / same origin. */
+  readonly VITE_API_BASE?: string;
+}
+interface ImportMeta {
+  readonly env: ImportMetaEnv;
+}

bench/tsconfig.json

@@ -0,0 +1,22 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "useDefineForClassFields": true,
+    "lib": ["ES2022", "DOM", "DOM.Iterable"],
+    "module": "ESNext",
+    "skipLibCheck": true,
+    "moduleResolution": "bundler",
+    "allowImportingTsExtensions": true,
+    "resolveJsonModule": true,
+    "isolatedModules": true,
+    "moduleDetection": "force",
+    "noEmit": true,
+    "jsx": "react-jsx",
+    "strict": true,
+    "noUnusedLocals": true,
+    "noUnusedParameters": true,
+    "noFallthroughCasesInSwitch": true,
+    "types": ["node", "vite/client"]
+  },
+  "include": ["src", "vite.config.ts"]
+}

bench/vite.config.ts

@@ -0,0 +1,18 @@
+import { defineConfig } from "vite";
+import react from "@vitejs/plugin-react-swc";
+
+// Dev server proxies /api to the bench API on bob so `fetch('/api/...')`
+// works without CORS/mixed-origin fuss during local development.
+// For a production build hosted elsewhere, set VITE_API_BASE to the bob
+// API origin (e.g. http://bob.hanzalova.internal:13132) instead.
+export default defineConfig({
+  plugins: [react()],
+  server: {
+    proxy: {
+      "/api": {
+        target: "http://bob.hanzalova.internal:13132",
+        changeOrigin: true,
+      },
+    },
+  },
+});

cortex.example.toml

@@ -5,6 +5,11 @@
 # Environment variable overrides use CORTEX_ prefix with __ separators:
 #   CORTEX_GATEWAY__LISTEN=0.0.0.0:31313
 
+# Path to the model catalogue (limits, cost, pinning, aliases, feasibility).
+# Defaults to the packaged location below; uncomment to override for a
+# non-packaged / local run.
+# models_config = "/etc/cortex/models.toml"
+
 [gateway]
 listen = "0.0.0.0:31313"
 metrics_listen = "0.0.0.0:31314"
@@ -43,3 +48,62 @@ vram_mb = 12288           # e.g. RTX 3060 (12 GB)
 pinned = [
     "your-org/embedding-model",
 ]
+
+# -- Entitlements (multi-tenant governance, #47) -------------------------
+# Identity + per-key token budgets. Omit this section entirely for the
+# legacy single-operator behaviour: requests are anonymous and uncapped.
+#
+# The local/static provider below is the source of truth for accounts,
+# keys, and hard caps until the upstream clearing house exists. Identity
+# rides standard bearer auth only — clients send
+#   Authorization: Bearer <key>
+# no custom headers or body fields.
+
+[entitlements]
+# Reject unauthenticated requests with 401 invalid_api_key. Leave false
+# (allow-anonymous) during rollout; flip to true once keys are issued.
+require_auth = false
+
+# One entry per API key.
+[[entitlements.keys]]
+key = "sk-example-rolling"        # the bearer token the client sends
+account_id = "team-research"      # billable account (keys may share one)
+key_id = "research-ci"            # stable label for ledger/metrics (optional)
+hard_cap = 5_000_000              # hard token cap over the window
+# Rolling window that resets — over-cap requests get 429 rate_limit_exceeded
+# + Retry-After, so well-behaved clients (opencode/AI SDK) back off and retry.
+window = { kind = "rolling", seconds = 3600 }
+
+[[entitlements.keys]]
+key = "sk-example-balance"
+account_id = "team-research"
+key_id = "research-prepaid"
+hard_cap = 20_000_000
+# Hard balance, no reset — exhaustion returns 429 insufficient_quota
+# (the client surfaces and stops). This is the default when `window` is
+# omitted. Never 402.
+window = { kind = "balance" }
+
+[[entitlements.keys]]
+key = "sk-example-infra"
+account_id = "operator"
+key_id = "infra"
+# No hard_cap → uncapped operator infra key (own fleet, own use). Still
+# metered for visibility.
+
+# -- Upstream (helexa mesh) entitlements client (#57) --------------------
+# When enabled, a bearer key NOT found in [[entitlements.keys]] above is
+# validated against the helexa-upstream authority (mesh accounts), and its
+# budget is reserved/settled there. Operator-local keys (incl. the infra
+# key) never leave this process. Fail-closed: if upstream is unreachable a
+# request is refused (503 + Retry-After), never served un-authorized.
+# Disabled by default — a standalone operator runs purely local.
+[upstream]
+enabled = false
+# url = "https://upstream.helexa.ai"
+# Shared client bearer this cortex presents (maps to an operator_id
+# upstream). Override via CORTEX_UPSTREAM__BEARER in prod.
+# bearer = "replace-with-operator-client-secret"
+# timeout_secs = 5
+# How often to flush served-usage counters to upstream for reconciliation (#58).
+# served_usage_report_interval_secs = 60

crates/cortex-core/src/catalogue.rs

@@ -1,6 +1,7 @@
 //! Model catalogue — profiles describing how to serve each model.
 
 use crate::discovery::DeviceInfo;
+use crate::harness::{ModelCost, ModelLimit};
 use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
 use std::path::Path;
@@ -35,6 +36,21 @@ pub struct ModelProfile {
     /// on this being explicit per model rather than implicit.
     #[serde(default)]
     pub source: Option<String>,
+
+    // ── Enrichment (issue #62) ────────────────────────────────
+    /// Per-model token budget. When present, advertised in `/v1/models`
+    /// so clients can size and compact their context automatically.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub limit: Option<ModelLimit>,
+    /// Operator-set pricing (USD per 1M tokens). `0.0` for self-hosted.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub cost: Option<ModelCost>,
+    /// Static capability flags the operator wants to advertise even
+    /// before the model is loaded on any neuron (e.g. `"reasoning"`,
+    /// `"tool_call"`). Runtime-detected capabilities from the harness
+    /// are unioned with this set in the gateway's `/v1/models` response.
+    #[serde(default)]
+    pub capabilities: Vec<String>,
 }
 
 fn default_min_devices() -> u32 {
@@ -152,6 +168,9 @@ mod tests {
             min_device_vram_mb: Some(24_000),
             pinned_on: vec![],
             source: None,
+            limit: None,
+            cost: None,
+            capabilities: vec![],
         }
     }
 

crates/cortex-core/src/config.rs

@@ -1,3 +1,4 @@
+use crate::entitlements::CapWindow;
 use figment::{
     Figment,
     providers::{Env, Format, Toml},
@@ -11,13 +12,98 @@ pub struct GatewayConfig {
     pub eviction: EvictionSettings,
     /// Neuron endpoints (replaces old NodeConfig with static vram_mb/pinned).
     pub neurons: Vec<NeuronEndpoint>,
-    /// Path to the model catalogue file (default: "models.toml").
+    /// Path to the model catalogue file. Defaults to the packaged
+    /// location (`/etc/cortex/models.toml`); set explicitly for
+    /// non-packaged / local runs.
     #[serde(default = "default_models_path")]
     pub models_config: String,
+    /// Multi-tenant governance: auth + per-key token budgets (#47). Empty
+    /// by default — anonymous, uncapped — so existing single-operator
+    /// setups keep working until keys are configured.
+    #[serde(default)]
+    pub entitlements: EntitlementsConfig,
+    /// helexa-upstream client (#57). When enabled, keys not found in the
+    /// local `[entitlements]` config are validated against the mesh
+    /// authority, and budget is reserved/settled there. Disabled by default
+    /// — a single operator runs purely local.
+    #[serde(default)]
+    pub upstream: UpstreamClientConfig,
+}
+
+/// `[upstream]` — the helexa-upstream authority client (#57). Locally
+/// unrecognised bearer keys are resolved against `url`'s `/authz/v1` surface
+/// (mesh accounts); local keys (operator + infra) never leave the process.
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
+pub struct UpstreamClientConfig {
+    /// Enable the upstream fallthrough. Off → purely local entitlements.
+    #[serde(default)]
+    pub enabled: bool,
+    /// Base URL of helexa-upstream (e.g. "https://upstream.helexa.ai").
+    #[serde(default)]
+    pub url: String,
+    /// Shared client bearer this cortex presents to `/authz/v1` (maps to an
+    /// operator_id upstream). Sent as `Authorization: Bearer <bearer>`.
+    #[serde(default)]
+    pub bearer: String,
+    /// Per-call timeout (seconds) to upstream.
+    #[serde(default = "default_upstream_timeout")]
+    pub timeout_secs: u64,
+    /// How often (seconds) to flush served-usage counters to upstream for
+    /// reconciliation (#58).
+    #[serde(default = "default_served_usage_interval")]
+    pub served_usage_report_interval_secs: u64,
+}
+
+fn default_upstream_timeout() -> u64 {
+    5
+}
+fn default_served_usage_interval() -> u64 {
+    60
+}
+
+/// `[entitlements]` — the local/static [`crate::entitlements::EntitlementProvider`]
+/// source of truth (#50). Accounts, keys, and hard caps live here; the
+/// future upstream client (#57) ignores this section.
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
+pub struct EntitlementsConfig {
+    /// Reject unauthenticated requests with `401 invalid_api_key` when
+    /// true. Default `false` (allow-anonymous) for dev / single-operator
+    /// continuity.
+    #[serde(default)]
+    pub require_auth: bool,
+    /// Static API keys and their budgets, consumed by the local provider.
+    #[serde(default)]
+    pub keys: Vec<ApiKeyConfig>,
+}
+
+/// One configured API key: the bearer token, the account it bills to, and
+/// its hard cap. `[[entitlements.keys]]` in TOML.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ApiKeyConfig {
+    /// The bearer token clients send in `Authorization: Bearer <key>`.
+    pub key: String,
+    /// Billable account. Multiple keys may share one account.
+    pub account_id: String,
+    /// Stable per-key identifier for ledger/metrics labels. Defaults to
+    /// `account_id` when omitted, so the secret is never used as a label.
+    #[serde(default)]
+    pub key_id: Option<String>,
+    /// Hard token cap. `None`/omitted = uncapped (e.g. operator infra key).
+    #[serde(default)]
+    pub hard_cap: Option<u64>,
+    /// Cap-window semantics. Default: a non-resetting [`CapWindow::Balance`].
+    #[serde(default)]
+    pub window: CapWindow,
 }
 
 fn default_models_path() -> String {
-    "models.toml".into()
+    // Absolute, so the systemd-launched binary finds the catalogue
+    // regardless of its working directory. The RPM installs the catalogue
+    // here (`cortex.spec`); a relative "models.toml" silently resolved to
+    // the service cwd and left the catalogue empty in production
+    // (pinning / aliases / limits all no-ops). Override via `models_config`
+    // in cortex.toml for local runs.
+    "/etc/cortex/models.toml".into()
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -79,6 +165,8 @@ impl Default for GatewayConfig {
             },
             neurons: vec![],
             models_config: default_models_path(),
+            entitlements: EntitlementsConfig::default(),
+            upstream: UpstreamClientConfig::default(),
         }
     }
 }

crates/cortex-core/src/discovery.rs

@@ -33,6 +33,12 @@ pub struct DiscoveryResponse {
     /// failure.
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub cuda_unavailable_reason: Option<String>,
+    /// The neuron's effective maximum prompt size in tokens
+    /// (`NEURON_MAX_PROMPT_TOKENS`) — the enforced prompt cap on this
+    /// host. `#[serde(default)]` (→ 0) for forward-compat with neurons
+    /// that predate this field; cortex treats 0 as "unknown".
+    #[serde(default)]
+    pub max_prompt_tokens: u64,
 }
 
 /// Runtime health metrics for a single GPU device.
@@ -62,6 +68,57 @@ pub struct HealthResponse {
     pub devices: Vec<DeviceHealth>,
     #[serde(default)]
     pub activation: ActivationStatus,
+    /// Per-model admission load (#53): how many requests are running vs.
+    /// queued on each loaded model right now. Cortex's load-aware router
+    /// (#55) reads this to spread traffic across replicas and to propagate
+    /// honest backpressure. `#[serde(default)]` keeps older gateways/neurons
+    /// interoperable (absent → empty → treated as no load info).
+    #[serde(default)]
+    pub models: Vec<ModelLoad>,
+}
+
+/// Live admission load for one loaded model (#53).
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ModelLoad {
+    pub id: String,
+    /// Requests currently running (batch-1 → 0 or 1).
+    pub in_flight: usize,
+    /// Requests waiting in the bounded admission queue.
+    pub queue_depth: usize,
+}
+
+#[cfg(test)]
+mod health_load_tests {
+    use super::*;
+
+    #[test]
+    fn health_response_without_models_field_still_deserializes() {
+        // A pre-#53 neuron's /health payload omits `models`; the gateway
+        // must still parse it (serde default → empty).
+        let json = r#"{"uptime_secs":42,"devices":[]}"#;
+        let resp: HealthResponse = serde_json::from_str(json).expect("back-compat parse");
+        assert_eq!(resp.uptime_secs, 42);
+        assert!(resp.models.is_empty());
+    }
+
+    #[test]
+    fn health_response_round_trips_model_load() {
+        let resp = HealthResponse {
+            uptime_secs: 1,
+            devices: vec![],
+            activation: ActivationStatus::default(),
+            models: vec![ModelLoad {
+                id: "Qwen/Qwen3.6-27B".into(),
+                in_flight: 1,
+                queue_depth: 3,
+            }],
+        };
+        let s = serde_json::to_string(&resp).unwrap();
+        let back: HealthResponse = serde_json::from_str(&s).unwrap();
+        assert_eq!(back.models.len(), 1);
+        assert_eq!(back.models[0].in_flight, 1);
+        assert_eq!(back.models[0].queue_depth, 3);
+    }
 }
 
 /// High-level activation state of the neuron daemon. The HTTP listener

crates/cortex-core/src/entitlements.rs

@@ -0,0 +1,152 @@
+//! Identity and entitlement primitives for multi-tenant governance (#47).
+//!
+//! Identity is the shared substrate the whole epic hangs off:
+//! `identity (principal) → accounting (spend) → policy → enforcement`. This
+//! module defines the seam — the [`EntitlementProvider`] trait and its data
+//! types — so the local/static provider (operator-config caps, in
+//! cortex-gateway) can land the auth + per-key-cap + amplification fix
+//! *before* any upstream clearing house exists. The future helexa-upstream
+//! client (#57) is just another impl of this trait.
+//!
+//! The provider owns three jobs:
+//! 1. **resolve** a bearer key to a [`Principal`] (drives auth, #49);
+//! 2. **reserve → settle/release** token budget around a request so spend
+//!    can never overshoot a hard cap under concurrency (drives budget
+//!    enforcement, #52);
+//! 3. expose a [`BudgetSnapshot`] for metering/metrics (#51).
+//!
+//! [`BudgetError`] carries the cap-window semantics so the caller can pick
+//! the correct #63 rejection (`rate_limit_exceeded` + `Retry-After` for a
+//! resetting window vs `insufficient_quota` for a hard balance) without the
+//! provider knowing anything about HTTP.
+
+use async_trait::async_trait;
+use serde::{Deserialize, Serialize};
+
+/// Internal header carrying the resolved account id from cortex to neuron.
+/// neuron trusts these over the WireGuard link (#54); cortex **strips** any
+/// client-supplied copy before stamping the authoritative value, so a client
+/// can never assert a principal directly.
+pub const HEADER_ACCOUNT_ID: &str = "x-helexa-account-id";
+/// Internal header carrying the resolved key id from cortex to neuron.
+pub const HEADER_KEY_ID: &str = "x-helexa-key-id";
+
+/// Who a request is for. Resolved once at the edge from the bearer key and
+/// carried through the request context. `account_id` is the billable owner
+/// (spendable at any operator, by decision); `key_id` identifies the
+/// specific API key for per-key hard caps and ledger/metrics labels.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct Principal {
+    pub account_id: String,
+    pub key_id: String,
+}
+
+/// Cap-window semantics for a key's hard cap. Determines which #63 code an
+/// over-cap reservation maps to.
+#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(tag = "kind", rename_all = "snake_case")]
+pub enum CapWindow {
+    /// Hard balance — the cap never resets. Exhaustion is permanent
+    /// (`429 insufficient_quota`, no `Retry-After`).
+    #[default]
+    Balance,
+    /// Rolling window of `seconds` that resets. Exhaustion is transient
+    /// (`429 rate_limit_exceeded` + `Retry-After` until reset).
+    Rolling { seconds: u64 },
+}
+
+/// An outstanding budget reservation. The caller holds this opaque handle
+/// between [`EntitlementProvider::reserve`] and exactly one of
+/// [`EntitlementProvider::settle`] / [`EntitlementProvider::release`]. Not
+/// `Clone` — a reservation is consumed once.
+#[derive(Debug)]
+pub struct Reservation {
+    /// Provider-local handle; opaque to the caller.
+    pub id: u64,
+    /// The principal this reservation belongs to.
+    pub principal: Principal,
+    /// Tokens reserved against the cap.
+    pub reserved: u64,
+}
+
+/// A point-in-time view of a key's budget, for metering and metrics (#51).
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct BudgetSnapshot {
+    /// Hard cap in tokens. `None` means uncapped (e.g. an operator infra
+    /// key, #58).
+    pub hard_cap: Option<u64>,
+    /// Settled spend in the current window.
+    pub spent: u64,
+    /// Sum of outstanding (un-settled) reservations.
+    pub reserved: u64,
+}
+
+/// Authentication failure — the bearer key could not be resolved.
+#[derive(Debug, thiserror::Error)]
+pub enum AuthError {
+    /// The key is genuinely unknown → `401 invalid_api_key` (#49/#63).
+    #[error("invalid or unknown API key")]
+    InvalidKey,
+    /// The authority that could resolve the key is unreachable (e.g. the
+    /// helexa-upstream client failed, #57). Fail **closed** but distinctly:
+    /// a transient outage must surface as `503 service_unavailable` +
+    /// `Retry-After`, never `401` — a real key must not be rejected as
+    /// invalid during an upstream blip.
+    #[error("entitlement authority unavailable; retry in {retry_after_secs}s")]
+    Unavailable { retry_after_secs: u64 },
+}
+
+/// Why a reservation was refused. Carries enough for the caller to build the
+/// correct #63 envelope without the provider touching HTTP.
+#[derive(Debug, thiserror::Error)]
+pub enum BudgetError {
+    /// A resetting window is exhausted → `429 rate_limit_exceeded` +
+    /// `Retry-After: retry_after_secs`.
+    #[error(
+        "rolling-window budget exhausted ({requested} requested, {available} available); \
+         resets in {retry_after_secs}s"
+    )]
+    RateLimited {
+        requested: u64,
+        available: u64,
+        retry_after_secs: u64,
+    },
+    /// A hard balance is exhausted → `429 insufficient_quota` (no
+    /// `Retry-After`; the client surfaces and stops). Never `402`.
+    #[error("hard balance exhausted ({requested} requested, {available} available)")]
+    InsufficientQuota { requested: u64, available: u64 },
+}
+
+/// The seam between cortex's enforcement and whatever decides entitlement —
+/// a local/static config provider today (#50), the helexa-upstream client
+/// later (#57). All methods are async so the upstream impl can do network
+/// I/O; the local impl resolves in-process.
+#[async_trait]
+pub trait EntitlementProvider: Send + Sync {
+    /// Resolve a bearer API key to its principal. `Err(InvalidKey)` for an
+    /// unknown/empty key.
+    async fn resolve(&self, api_key: &str) -> Result<Principal, AuthError>;
+
+    /// Reserve up to `max_tokens` against the principal's cap. Returns a
+    /// handle on success, or a [`BudgetError`] (which the caller maps to a
+    /// #63 `429`) if the reservation would exceed the cap. Reserving the
+    /// *maximum* a request could consume before dispatch is what prevents
+    /// overshoot under concurrency.
+    async fn reserve(
+        &self,
+        principal: &Principal,
+        max_tokens: u64,
+    ) -> Result<Reservation, BudgetError>;
+
+    /// Settle a reservation with the tokens actually consumed, releasing the
+    /// unused remainder back to the cap.
+    async fn settle(&self, reservation: Reservation, actual_tokens: u64);
+
+    /// Release a reservation in full — e.g. dispatch failed before any
+    /// tokens were consumed.
+    async fn release(&self, reservation: Reservation);
+
+    /// Current budget snapshot for a principal, for metering/metrics.
+    /// `None` if the provider doesn't track this principal.
+    async fn snapshot(&self, principal: &Principal) -> Option<BudgetSnapshot>;
+}

crates/cortex-core/src/error_envelope.rs

@@ -0,0 +1,257 @@
+//! The OpenAI-standard error envelope (#60) and the rejection contract
+//! that rides on it (#63).
+//!
+//! Every non-2xx response cortex and neuron emit uses the shape
+//!
+//! ```json
+//! { "error": { "message": "...", "type": "...", "code": "...", "param": null } }
+//! ```
+//!
+//! because OpenAI-compatible clients (opencode, the AI SDK, litellm, the
+//! OpenAI SDKs) read `error.type` / `error.code` to decide what to do —
+//! most importantly `code == "context_length_exceeded"` triggers
+//! auto-compaction, and a `429` with `Retry-After` makes them back off and
+//! retry rather than surfacing an opaque failure. A flat `{"error":"..."}`
+//! string is invisible to that logic.
+//!
+//! This module is the single source of truth for that envelope. It is
+//! deliberately **axum-agnostic** — cortex-core is a pure types crate — so
+//! it carries the response as data (`status`, `body()`, `retry_after_secs`)
+//! and each HTTP crate (cortex-gateway, neuron) owns a tiny adapter that
+//! turns an [`OpenAiError`] into its framework's response type, setting the
+//! `Retry-After` header when present.
+//!
+//! Retryable conditions **must** carry `Retry-After` (per #63). The named
+//! constructors below encode that: [`OpenAiError::rate_limit_exceeded`] and
+//! [`OpenAiError::service_unavailable`] take a retry hint;
+//! [`OpenAiError::insufficient_quota`] (hard balance, no reset) and
+//! [`OpenAiError::context_length_exceeded`] / [`OpenAiError::invalid_api_key`]
+//! (permanent) do not. `402 Payment Required` is banned by the contract — use
+//! `429 insufficient_quota` for hard budget exhaustion.
+
+use serde_json::{Map, Value, json};
+
+/// A rejection rendered in the OpenAI error envelope.
+///
+/// Build with [`OpenAiError::new`] (or a named constructor), refine with the
+/// `with_*` builders, then hand to the consuming crate's adapter to turn into
+/// an HTTP response.
+#[derive(Debug, Clone)]
+pub struct OpenAiError {
+    /// HTTP status code (e.g. `401`, `429`, `503`).
+    pub status: u16,
+    /// Broad OpenAI category — `"invalid_request_error"`, `"api_error"`,
+    /// `"rate_limit_error"`, …
+    pub error_type: String,
+    /// Specific machine-readable code clients key on (`"invalid_api_key"`,
+    /// `"rate_limit_exceeded"`, `"context_length_exceeded"`, …). `None`
+    /// renders as JSON `null`.
+    pub code: Option<String>,
+    /// Human-readable, actionable message.
+    pub message: String,
+    /// OpenAI's `param` field — the offending request parameter, if any.
+    pub param: Option<String>,
+    /// Seconds to advertise in the `Retry-After` header. Set only on
+    /// retryable conditions; `None` means no header.
+    pub retry_after_secs: Option<u64>,
+    /// Diagnostic fields merged *inside* the `error` object (e.g.
+    /// `prompt_len`, `max`, `free_mb`) so they don't break the envelope
+    /// shape. Clients ignore unknown keys.
+    pub extra: Map<String, Value>,
+}
+
+impl OpenAiError {
+    /// Construct an envelope with an explicit code. For a `null` code use
+    /// [`OpenAiError::without_code`].
+    pub fn new(
+        status: u16,
+        error_type: impl Into<String>,
+        code: impl Into<String>,
+        message: impl Into<String>,
+    ) -> Self {
+        Self {
+            status,
+            error_type: error_type.into(),
+            code: Some(code.into()),
+            message: message.into(),
+            param: None,
+            retry_after_secs: None,
+            extra: Map::new(),
+        }
+    }
+
+    /// Construct an envelope whose `code` is `null` (e.g. an unclassified
+    /// internal error).
+    pub fn without_code(
+        status: u16,
+        error_type: impl Into<String>,
+        message: impl Into<String>,
+    ) -> Self {
+        Self {
+            status,
+            error_type: error_type.into(),
+            code: None,
+            message: message.into(),
+            param: None,
+            retry_after_secs: None,
+            extra: Map::new(),
+        }
+    }
+
+    /// Advertise a `Retry-After` (seconds). Use on retryable rejections.
+    pub fn with_retry_after(mut self, secs: u64) -> Self {
+        self.retry_after_secs = Some(secs);
+        self
+    }
+
+    /// Set the OpenAI `param` field.
+    pub fn with_param(mut self, param: impl Into<String>) -> Self {
+        self.param = Some(param.into());
+        self
+    }
+
+    /// Merge one diagnostic field into the error object.
+    pub fn with_extra(mut self, key: impl Into<String>, value: Value) -> Self {
+        self.extra.insert(key.into(), value);
+        self
+    }
+
+    /// Merge a bag of diagnostic fields into the error object.
+    pub fn with_extras(mut self, extras: Map<String, Value>) -> Self {
+        for (k, v) in extras {
+            self.extra.insert(k, v);
+        }
+        self
+    }
+
+    /// Render the `{ "error": { … } }` body. Field order is irrelevant to
+    /// clients (they parse JSON); the standard keys come first, then any
+    /// diagnostic extras.
+    pub fn body(&self) -> Value {
+        let mut error = Map::new();
+        error.insert("message".into(), Value::String(self.message.clone()));
+        error.insert("type".into(), Value::String(self.error_type.clone()));
+        error.insert(
+            "code".into(),
+            self.code.clone().map(Value::String).unwrap_or(Value::Null),
+        );
+        error.insert(
+            "param".into(),
+            self.param.clone().map(Value::String).unwrap_or(Value::Null),
+        );
+        for (k, v) in &self.extra {
+            error.insert(k.clone(), v.clone());
+        }
+        json!({ "error": Value::Object(error) })
+    }
+
+    // ── Named constructors for the #63 standard codes ──────────────────
+
+    /// `401 invalid_api_key` — missing/invalid bearer token (#49). Permanent.
+    pub fn invalid_api_key(message: impl Into<String>) -> Self {
+        Self::new(401, "invalid_request_error", "invalid_api_key", message)
+    }
+
+    /// `429 rate_limit_exceeded` + `Retry-After` — transient overload,
+    /// fair-share/in-flight cap, admission rejection, or a rolling budget
+    /// window that resets (#52/#53/#54/#55). Clients back off and retry.
+    pub fn rate_limit_exceeded(message: impl Into<String>, retry_after_secs: u64) -> Self {
+        Self::new(429, "rate_limit_error", "rate_limit_exceeded", message)
+            .with_retry_after(retry_after_secs)
+    }
+
+    /// `429 insufficient_quota` — hard balance exhausted, no reset (#52).
+    /// No `Retry-After`; the client surfaces and stops. (Never `402`.)
+    pub fn insufficient_quota(message: impl Into<String>) -> Self {
+        Self::new(429, "insufficient_quota", "insufficient_quota", message)
+    }
+
+    /// `400 context_length_exceeded` — prompt exceeds the model's context
+    /// window (#56/#60). Permanent for this request; opencode auto-compacts.
+    pub fn context_length_exceeded(message: impl Into<String>) -> Self {
+        Self::new(
+            400,
+            "invalid_request_error",
+            "context_length_exceeded",
+            message,
+        )
+    }
+
+    /// `503 service_unavailable` + optional `Retry-After` — transient
+    /// backend unavailability (no healthy nodes, recovery, fail-closed
+    /// upstream). Retryable when a hint is given.
+    pub fn service_unavailable(message: impl Into<String>, retry_after_secs: Option<u64>) -> Self {
+        let mut err = Self::new(503, "api_error", "service_unavailable", message);
+        err.retry_after_secs = retry_after_secs;
+        err
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn body_has_standard_envelope_shape() {
+        let env = OpenAiError::new(429, "rate_limit_error", "rate_limit_exceeded", "slow down");
+        let body = env.body();
+        let error = body.get("error").and_then(Value::as_object).unwrap();
+        assert_eq!(error["message"], "slow down");
+        assert_eq!(error["type"], "rate_limit_error");
+        assert_eq!(error["code"], "rate_limit_exceeded");
+        assert_eq!(error["param"], Value::Null);
+    }
+
+    #[test]
+    fn without_code_renders_null_code() {
+        let env = OpenAiError::without_code(500, "api_error", "kaboom");
+        assert_eq!(env.body()["error"]["code"], Value::Null);
+    }
+
+    #[test]
+    fn extras_ride_inside_the_error_object() {
+        let env = OpenAiError::context_length_exceeded("too long")
+            .with_extra("prompt_len", json!(60_000))
+            .with_extra("max", json!(49_152));
+        let error = &env.body()["error"];
+        assert_eq!(error["prompt_len"], 60_000);
+        assert_eq!(error["max"], 49_152);
+        assert_eq!(error["code"], "context_length_exceeded");
+    }
+
+    #[test]
+    fn rolling_window_rejection_carries_retry_after() {
+        let env = OpenAiError::rate_limit_exceeded("budget window", 30);
+        assert_eq!(env.status, 429);
+        assert_eq!(env.retry_after_secs, Some(30));
+    }
+
+    #[test]
+    fn hard_balance_rejection_has_no_retry_after() {
+        let env = OpenAiError::insufficient_quota("out of credit");
+        assert_eq!(env.status, 429);
+        assert_eq!(env.code.as_deref(), Some("insufficient_quota"));
+        assert_eq!(env.retry_after_secs, None);
+    }
+
+    #[test]
+    fn permanent_rejections_have_no_retry_after() {
+        assert_eq!(OpenAiError::invalid_api_key("nope").retry_after_secs, None);
+        assert_eq!(
+            OpenAiError::context_length_exceeded("too long").retry_after_secs,
+            None
+        );
+    }
+
+    #[test]
+    fn service_unavailable_retry_after_is_optional() {
+        assert_eq!(
+            OpenAiError::service_unavailable("recovering", Some(5)).retry_after_secs,
+            Some(5)
+        );
+        assert_eq!(
+            OpenAiError::service_unavailable("gone", None).retry_after_secs,
+            None
+        );
+    }
+}

crates/cortex-core/src/harness.rs

@@ -36,6 +36,60 @@ pub struct ModelSpec {
     pub devices: Option<Vec<u32>>,
 }
 
+/// Per-model token budget advertised by the catalogue or neuron.
+///
+/// `context` is the hard wall (the served max-seq-len).  `input` is the
+/// compaction trigger — when set, opencode treats it as "usable context =
+/// input − reserved".  When omitted, clients fall back to `context − output`.
+/// `output` is the maximum number of generation tokens.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ModelLimit {
+    /// Hard wall — served max-seq-len in tokens.
+    pub context: usize,
+    /// Compaction trigger / usable input budget.  When absent clients fall
+    /// back to `context − output`.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub input: Option<usize>,
+    /// Maximum number of generation tokens.
+    pub output: usize,
+}
+
+/// Operator-set pricing, **USD per 1,000,000 tokens, as JSON numbers**
+/// (`float`) — the models.dev/opencode `cost` convention, which is what
+/// helexa's primary client reads. NOT per-token, NOT decimal strings (that
+/// is OpenRouter's `pricing` shape, which helexa deliberately does not emit
+/// — see #68). A client must not rescale by 10⁶.
+///
+/// `cost` is sourced from the operator's `models.toml` catalogue profile and
+/// surfaced verbatim on `/v1/models`. The *absent* vs *zero* distinction is
+/// intentional and load-bearing (#68):
+/// - **`cost` absent** (the whole object omitted) — the model is **not
+///   priced**: the operator has not declared a rate. Clients should treat
+///   spend as unknown, not free.
+/// - **`cost` present with `input`/`output` = `0.0`** — the model is
+///   **intentionally free** (self-hosted, no charge). opencode renders `$0`.
+///
+/// Cache fields are optional — set them only when the backend supports a
+/// prefix-cache discount tier (relevant once cache-token reporting, #64,
+/// lands). The advertised rate here must equal the rate metering (#51) and
+/// reconciliation (#58/#59) bill against; today both read this catalogue
+/// value.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ModelCost {
+    /// USD per 1M input (prompt) tokens.
+    #[serde(default)]
+    pub input: f64,
+    /// USD per 1M output (completion) tokens.
+    #[serde(default)]
+    pub output: f64,
+    /// USD per 1M cache-hit tokens (optional).
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub cache_read: Option<f64>,
+    /// USD per 1M cache-write tokens (optional).
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub cache_write: Option<f64>,
+}
+
 /// A model as reported by a harness.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct ModelInfo {
@@ -46,14 +100,32 @@ pub struct ModelInfo {
     pub vram_used_mb: Option<u64>,
     /// Modalities this loaded model supports. Today: `["text"]` for
     /// text-only checkpoints, `["text", "vision"]` for vision-capable
-    /// ones (Stage B7 of the vision plan). Clients like litellm /
-    /// agent0 can gate `image_url` submission on the advertised set.
+    /// ones (Stage B7). Clients like litellm / agent0 can gate
+    /// `image_url` submission on the advertised set.
     ///
     /// Optional in the wire format so older clients that don't read
     /// it stay compatible. Default-empty for absent/older data, which
     /// callers can interpret as "text".
     #[serde(default, skip_serializing_if = "Vec::is_empty")]
     pub capabilities: Vec<String>,
+
+    // ── Enrichment (issue #62) ────────────────────────────────
+    /// Token budget advertised by the catalogue or discovered at load time.
+    /// `None` when neither the catalogue nor the loaded model can provide it.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub limit: Option<ModelLimit>,
+    /// Operator-set pricing — see [`ModelCost`] for units and the
+    /// absent (not priced) vs `0.0` (intentionally free) distinction.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub cost: Option<ModelCost>,
+    /// `true` when the model's tokenizer contains recognised tool-call
+    /// marker tokens (`<tool_call>` / `<\/tool_call>` convention).
+    #[serde(default)]
+    pub tool_call: bool,
+    /// `true` when the model's tokenizer contains recognised reasoning
+    /// marker tokens (`<think>` / `<\/think>` or similar).
+    #[serde(default)]
+    pub reasoning: bool,
 }
 
 /// What an inference harness must do, from neuron's perspective.

crates/cortex-core/src/lib.rs

@@ -3,6 +3,8 @@ pub mod build_info;
 pub mod catalogue;
 pub mod config;
 pub mod discovery;
+pub mod entitlements;
+pub mod error_envelope;
 pub mod harness;
 pub mod metrics;
 pub mod node;

crates/cortex-core/src/node.rs

@@ -1,4 +1,5 @@
-use crate::discovery::{ActivationStatus, DiscoveryResponse};
+use crate::discovery::{ActivationStatus, DiscoveryResponse, ModelLoad};
+use crate::harness::{ModelCost, ModelLimit};
 use chrono::{DateTime, Utc};
 use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
@@ -26,6 +27,17 @@ pub struct NodeState {
     /// to synthesize `Loading` locations so clients see a catalogued
     /// model that's mid-prewarm as "loading", not "missing".
     pub activation: Option<ActivationStatus>,
+    /// Last-seen per-model admission load from this neuron's `/health`
+    /// (#53), keyed by model id. The router (#55) reads it to pick the
+    /// least-busy replica when a model is loaded on more than one neuron.
+    /// Empty until the first /health poll reports load.
+    pub model_load: HashMap<String, ModelLoad>,
+    /// Consecutive failed `/models` polls. The poller marks a node
+    /// unhealthy only once this crosses a threshold, so a single transient
+    /// miss (e.g. a neuron momentarily slow to answer while busy) doesn't
+    /// yank the node — and all its models — out of routing. Reset to 0 on
+    /// any successful poll.
+    pub consecutive_poll_failures: u32,
 }
 
 /// A model registered on a node, with its runtime status.
@@ -43,6 +55,21 @@ pub struct ModelEntry {
     /// older persisted/serialised entries deserialisable.
     #[serde(default)]
     pub capabilities: Vec<String>,
+    /// Runtime-detected capability flags from the neuron's `/models`
+    /// response (`ModelInfo`). `false` when the neuron predates these
+    /// fields or hasn't reported them yet.
+    #[serde(default)]
+    pub tool_call: bool,
+    #[serde(default)]
+    pub reasoning: bool,
+    /// Self-derived token budget the neuron computed for this loaded
+    /// model (#67), copied from `ModelInfo.limit` at poll time. `None`
+    /// when the neuron doesn't compute one (arch without a context
+    /// profile, or derivation disabled). This is the authoritative
+    /// source the gateway advertises — operator-declared catalogue
+    /// limits are no longer consulted.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub limit: Option<ModelLimit>,
 }
 
 /// Model lifecycle status.
@@ -99,10 +126,60 @@ pub struct CortexModelEntry {
     pub locations: Vec<ModelLocation>,
     /// Union of the modalities advertised by every neuron that has this
     /// model loaded (e.g. `["text", "vision"]`). Empty for catalogue-only
-    /// entries with no loaded location — the catalogue profile doesn't
-    /// declare capabilities yet (tracked separately from C3).
+    /// entries with no loaded location — filled from catalogue profile
+    /// capabilities when available, then unioned with runtime-detected
+    /// values from loaded neurons.
     #[serde(default)]
     pub capabilities: Vec<String>,
+    // ── Enrichment (issue #62) ────────────────────────────────
+    /// Per-model token budget from the catalogue profile or discovered
+    /// at load time. `None` when neither source provides it.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub limit: Option<ModelLimit>,
+    /// Operator-set pricing from the catalogue profile — see
+    /// [`cortex_core::harness::ModelCost`] for units (USD per 1M tokens) and
+    /// the absent (not priced) vs `0.0` (intentionally free) distinction.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub cost: Option<ModelCost>,
+    /// `true` when any neuron reports this model supports tool calls.
+    #[serde(default)]
+    pub tool_call: bool,
+    /// `true` when any neuron reports this model supports reasoning tokens.
+    #[serde(default)]
+    pub reasoning: bool,
+    // ── Flat ecosystem context-window fields (issue #78) ──────
+    // Duplicates of `limit` under the flat, vLLM-convention key names
+    // (`max_model_len` et al.) that OpenAI-ecosystem clients (Hermes
+    // Agent, vLLM tooling) probe for — they cannot see `limit.context`.
+    // Additive: `limit` stays the opencode-oriented source of truth.
+    // Derived, never set directly — call [`sync_flat_limit`] after the
+    // final `limit` value is known. Omitted (not `0`) when the window
+    // is unknown; absent-vs-zero is load-bearing, as with `cost`.
+    //
+    // [`sync_flat_limit`]: CortexModelEntry::sync_flat_limit
+    /// Served max-seq-len in tokens — mirrors `limit.context`.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub max_model_len: Option<usize>,
+    /// Usable input budget — mirrors `limit.input` when present.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub max_input_tokens: Option<usize>,
+    /// Maximum generation tokens — mirrors `limit.output`.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub max_output_tokens: Option<usize>,
+}
+
+impl CortexModelEntry {
+    /// Re-derive the flat ecosystem fields (#78) from `limit`.
+    ///
+    /// Must run after the final `limit` is known (post merge/tightening),
+    /// immediately before serialization. Fully overwrites: a `None` limit
+    /// clears the flat fields, so stale values can't survive a merge that
+    /// dropped the limit.
+    pub fn sync_flat_limit(&mut self) {
+        self.max_model_len = self.limit.as_ref().map(|l| l.context);
+        self.max_input_tokens = self.limit.as_ref().and_then(|l| l.input);
+        self.max_output_tokens = self.limit.as_ref().map(|l| l.output);
+    }
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]

crates/cortex-core/src/openai.rs

@@ -106,6 +106,48 @@ pub struct Usage {
     pub prompt_tokens: u64,
     pub completion_tokens: u64,
     pub total_tokens: u64,
+    /// OpenAI-standard breakdown of `completion_tokens`. Optional and
+    /// additive — clients that don't read it are unaffected. Carries
+    /// `reasoning_tokens` for reasoning models (a sub-count of
+    /// `completion_tokens`, never added into `total_tokens`).
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub completion_tokens_details: Option<CompletionTokensDetails>,
+    /// OpenAI-standard breakdown of `prompt_tokens`. Populated once
+    /// prompt caching lands (#11); `None` until then.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub prompt_tokens_details: Option<PromptTokensDetails>,
+    /// helexa extension (non-OpenAI): server-measured prefill/decode
+    /// timing, so the bench harness can compute true prefill vs decode
+    /// tok/s instead of inferring both from client-side SSE arrival
+    /// (#85). Additive and optional — standard OpenAI clients ignore
+    /// it; cortex forwards usage verbatim so it survives proxying.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub helexa_timing: Option<HelexaTiming>,
+}
+
+/// helexa extension carried on [`Usage::helexa_timing`]. Mirrors
+/// neuron's internal `FinishTiming`. All fields are server-measured;
+/// `prefill_tokens` is the prefill-rate denominator.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct HelexaTiming {
+    pub prefill_ms: u64,
+    pub decode_ms: u64,
+    pub prefill_tokens: u64,
+}
+
+/// Sub-counts of `Usage::completion_tokens`.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct CompletionTokensDetails {
+    /// Tokens generated inside the model's reasoning span.
+    pub reasoning_tokens: u64,
+}
+
+/// Sub-counts of `Usage::prompt_tokens`.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct PromptTokensDetails {
+    /// Prompt tokens served from cache (cache-read rate). Populated
+    /// once prompt caching lands (#11).
+    pub cached_tokens: u64,
 }
 
 // ── Models list response ─────────────────────────────────────────────

crates/cortex-core/src/responses.rs

@@ -66,14 +66,48 @@ pub struct ResponsesRequest {
     pub extra: Value,
 }
 
-/// `input` is either a single string or an array of typed items.
+/// `input` is either a single string or an array of items.
 /// `#[serde(untagged)]` so the wire shape `"input": "hi"` and
 /// `"input": [{...}]` both deserialize.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 #[serde(untagged)]
 pub enum ResponsesInput {
     Text(String),
-    Items(Vec<ResponsesInputItem>),
+    Items(Vec<ResponsesInputElement>),
+}
+
+/// One element of an `input` array.
+///
+/// OpenAI's Responses API accepts three shapes here, and real clients
+/// use all of them — most notably agent-zero (via litellm), which
+/// sends the bare "easy message" form. We must tolerate every shape,
+/// because `input` is an `#[serde(untagged)]` array: a single element
+/// that matches no variant fails the *entire* request with a 422
+/// (`did not match any variant of untagged enum ResponsesInput`).
+///
+/// 1. [`Self::Typed`] — an item carrying an explicit `"type"`
+///    discriminant (`message`, `function_call`, `function_call_output`,
+///    `reasoning`).
+/// 2. [`Self::EasyMessage`] — a bare `{role, content}` with **no**
+///    `type` field. This is OpenAI's `EasyInputMessage` and what
+///    litellm emits for every turn. `content` is optional so an
+///    assistant turn carrying only tool calls (`content: null`) still
+///    parses.
+/// 3. [`Self::Other`] — anything else, captured as raw JSON and
+///    dropped during translation. This is the forward-compat escape
+///    hatch that mirrors [`ResponsesRequest::extra`] at the item
+///    level: an unmodeled item type can never again reject the whole
+///    request.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(untagged)]
+pub enum ResponsesInputElement {
+    Typed(ResponsesInputItem),
+    EasyMessage {
+        role: String,
+        #[serde(default, skip_serializing_if = "Option::is_none")]
+        content: Option<ResponsesMessageContent>,
+    },
+    Other(Value),
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -91,8 +125,11 @@ pub enum ResponsesInputItem {
         name: String,
         arguments: String,
     },
-    /// User is feeding a tool result back into the model.
-    FunctionCallOutput { call_id: String, output: String },
+    /// User is feeding a tool result back into the model. `output`
+    /// is a `Value` because OpenAI allows it to be either a plain
+    /// string or an array of content parts; the translator renders
+    /// either form to text rather than losing the tool result.
+    FunctionCallOutput { call_id: String, output: Value },
     /// Reasoning items emitted by o-series models. Accepted but
     /// not forwarded to the model — neuron's candle path doesn't
     /// surface reasoning separately yet.
@@ -132,6 +169,11 @@ pub enum ResponsesContentPart {
         #[serde(default, skip_serializing_if = "Vec::is_empty")]
         annotations: Vec<Value>,
     },
+    /// Any content-part type we don't model (e.g. `refusal`, audio).
+    /// Captured as a unit so an unknown part can't reject the whole
+    /// request; dropped during translation.
+    #[serde(other)]
+    Unknown,
 }
 
 // ── Response (non-streaming) ─────────────────────────────────────────
@@ -202,6 +244,30 @@ pub struct ResponsesUsage {
     pub input_tokens: u64,
     pub output_tokens: u64,
     pub total_tokens: u64,
+    /// OpenAI-standard breakdown of `output_tokens`. Optional and
+    /// additive. Carries `reasoning_tokens` for reasoning models (a
+    /// sub-count of `output_tokens`, never added into `total_tokens`).
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub output_tokens_details: Option<OutputTokensDetails>,
+    /// OpenAI-standard breakdown of `input_tokens`. Populated once
+    /// prompt caching lands (#11); `None` until then.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub input_tokens_details: Option<InputTokensDetails>,
+}
+
+/// Sub-counts of `ResponsesUsage::output_tokens`.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct OutputTokensDetails {
+    /// Tokens generated inside the model's reasoning span.
+    pub reasoning_tokens: u64,
+}
+
+/// Sub-counts of `ResponsesUsage::input_tokens`.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct InputTokensDetails {
+    /// Input tokens served from cache (cache-read rate). Populated
+    /// once prompt caching lands (#11).
+    pub cached_tokens: u64,
 }
 
 // ── Streaming event names ────────────────────────────────────────────
@@ -253,20 +319,116 @@ mod tests {
             ResponsesInput::Items(items) => {
                 assert_eq!(items.len(), 1);
                 match &items[0] {
-                    ResponsesInputItem::Message { role, content } => {
+                    ResponsesInputElement::Typed(ResponsesInputItem::Message { role, content }) => {
                         assert_eq!(role, "user");
                         match content {
                             ResponsesMessageContent::Text(t) => assert_eq!(t, "hi"),
                             other => panic!("expected Text content, got {other:?}"),
                         }
                     }
-                    other => panic!("expected Message item, got {other:?}"),
+                    other => panic!("expected typed Message item, got {other:?}"),
                 }
             }
             other => panic!("expected Items, got {other:?}"),
         }
     }
 
+    #[test]
+    fn deserialises_bare_easy_message_without_type() {
+        // The shape agent-zero (via litellm) actually sends: `input`
+        // items are bare `{role, content}` with NO `type` field. This
+        // is the exact payload that was returning 422.
+        let raw = r#"{
+            "model": "Qwen/Qwen3.6-27B",
+            "store": true,
+            "tools": [{"type": "function", "name": "x", "description": "d", "parameters": {}}],
+            "input": [
+                {"role": "system", "content": "you are helpful"},
+                {"role": "assistant", "content": "{\"tool_name\":\"response\"}"},
+                {"role": "user", "content": "hi"}
+            ]
+        }"#;
+        let req: ResponsesRequest = serde_json::from_str(raw).unwrap();
+        let items = match req.input {
+            ResponsesInput::Items(i) => i,
+            other => panic!("expected Items, got {other:?}"),
+        };
+        assert_eq!(items.len(), 3);
+        for el in &items {
+            assert!(
+                matches!(el, ResponsesInputElement::EasyMessage { .. }),
+                "expected EasyMessage, got {el:?}"
+            );
+        }
+        // `tools` / `store` ride through `extra`, not `input`.
+        assert!(req.extra.get("tools").is_some());
+        assert_eq!(req.extra.get("store"), Some(&Value::Bool(true)));
+    }
+
+    #[test]
+    fn tolerates_null_content_and_unknown_item_types() {
+        // An assistant turn carrying only tool calls has `content: null`;
+        // and a future/unmodeled item type must not 422 the request.
+        let raw = r#"{
+            "model": "m",
+            "input": [
+                {"role": "assistant", "content": null},
+                {"type": "item_reference", "id": "abc"},
+                {"type": "function_call_output", "call_id": "c1",
+                 "output": [{"type": "output_text", "text": "result"}]},
+                {"role": "user", "content": "go"}
+            ]
+        }"#;
+        let req: ResponsesRequest = serde_json::from_str(raw).unwrap();
+        let items = match req.input {
+            ResponsesInput::Items(i) => i,
+            other => panic!("expected Items, got {other:?}"),
+        };
+        assert_eq!(items.len(), 4);
+        assert!(matches!(
+            &items[0],
+            ResponsesInputElement::EasyMessage { content: None, .. }
+        ));
+        assert!(matches!(&items[1], ResponsesInputElement::Other(_)));
+        assert!(matches!(
+            &items[2],
+            ResponsesInputElement::Typed(ResponsesInputItem::FunctionCallOutput { .. })
+        ));
+        assert!(matches!(
+            &items[3],
+            ResponsesInputElement::EasyMessage { .. }
+        ));
+    }
+
+    #[test]
+    fn tolerates_unknown_content_part_type() {
+        // A `refusal` (or any unmodeled) content part must parse, not 422.
+        let raw = r#"{
+            "model": "m",
+            "input": [
+                {"role": "assistant", "content": [
+                    {"type": "refusal", "refusal": "no"},
+                    {"type": "output_text", "text": "ok"}
+                ]}
+            ]
+        }"#;
+        let req: ResponsesRequest = serde_json::from_str(raw).unwrap();
+        let items = match req.input {
+            ResponsesInput::Items(i) => i,
+            other => panic!("expected Items, got {other:?}"),
+        };
+        let parts = match &items[0] {
+            ResponsesInputElement::EasyMessage {
+                content: Some(ResponsesMessageContent::Parts(p)),
+                ..
+            } => p,
+            other => panic!("expected EasyMessage with Parts, got {other:?}"),
+        };
+        assert_eq!(parts.len(), 2);
+        assert!(matches!(&parts[0], ResponsesContentPart::Unknown));
+        assert!(matches!(&parts[1], ResponsesContentPart::OutputText { .. }));
+    }
+
     #[test]
     fn deserialises_input_with_image() {
         let raw = r#"{
@@ -284,10 +446,10 @@ mod tests {
             other => panic!("expected Items, got {other:?}"),
         };
         let parts = match &items[0] {
-            ResponsesInputItem::Message {
+            ResponsesInputElement::Typed(ResponsesInputItem::Message {
                 content: ResponsesMessageContent::Parts(p),
                 ..
-            } => p,
+            }) => p,
             other => panic!("expected Parts, got {other:?}"),
         };
         assert_eq!(parts.len(), 2);
@@ -336,6 +498,8 @@ mod tests {
                 input_tokens: 5,
                 output_tokens: 3,
                 total_tokens: 8,
+                output_tokens_details: None,
+                input_tokens_details: None,
             }),
         };
         let json = serde_json::to_string(&r).unwrap();

crates/cortex-core/src/translate.rs

@@ -11,47 +11,89 @@ use crate::openai::{
 use serde_json::{Value, json};
 
 /// Convert an Anthropic Messages request into an OpenAI ChatCompletion request.
+///
+/// This is the request half of the round trip Claude Code (and any
+/// Anthropic-native client pointed at cortex via `ANTHROPIC_BASE_URL`)
+/// exercises. The non-obvious work here is **tool translation**: the
+/// Anthropic and OpenAI tool shapes differ, and neuron feeds whatever
+/// `tools` array it receives straight into the HF chat template, which
+/// iterates the OpenAI shape (`tool.function.name`,
+/// `tool.function.parameters`). If we forwarded Anthropic-shaped tools
+/// (`{name, description, input_schema}`) verbatim the template would
+/// render empty/garbage definitions and the model would improvise an
+/// unparseable tool-call format — exactly the
+/// `<tool_use_name>…</tool_use_name>` text that leaks through to the
+/// client. So we reshape here:
+///
+/// - tool **definitions**: `{name, description, input_schema}` →
+///   `{type:"function", function:{name, description, parameters}}`
+/// - `tool_choice`: Anthropic `{type:"auto"|"any"|"tool", name}` →
+///   OpenAI `"auto"|"required"|{type:"function",function:{name}}`
+/// - assistant `tool_use` content blocks → an OpenAI assistant message
+///   carrying `tool_calls` (with `arguments` JSON-stringified)
+/// - user `tool_result` content blocks → standalone `role:"tool"`
+///   messages keyed by `tool_call_id`
 pub fn anthropic_to_openai(req: MessagesRequest) -> ChatCompletionRequest {
-    let mut messages = Vec::new();
-
-    // Anthropic `system` field becomes a system message.
+    // Collect ALL system content into a single leading system message.
+    // The top-level `system` field PLUS any `role:"system"` turns inside
+    // `messages` (Claude Code injects extra system-role messages beyond
+    // the top-level one) are merged into one message at index 0.
+    //
+    // This is load-bearing: most chat templates — Qwen3.6's among them —
+    // hard-reject a system message anywhere but the start
+    // (`raise_exception('System message must be at the beginning.')`),
+    // and on that render error neuron silently falls back to a
+    // template that renders NO tools at all, so the model gets zero
+    // tool-format guidance and improvises an unparseable tool syntax —
+    // tool calling breaks entirely. Merging keeps every system
+    // instruction while satisfying the template.
+    let mut system_parts: Vec<String> = Vec::new();
     if let Some(system) = req.system {
-        let content = match system {
+        system_parts.push(match system {
             SystemPrompt::Text(t) => t,
-            SystemPrompt::Blocks(blocks) => serde_json::to_string(&blocks).unwrap_or_default(),
-        };
+            SystemPrompt::Blocks(blocks) => system_blocks_to_text(&blocks),
+        });
+    }
+
+    // Translate the conversation. A single Anthropic message can fan out
+    // into several OpenAI messages (tool results split into their own
+    // `role:"tool"` turns); `role:"system"` turns are pulled into the
+    // accumulator above rather than emitted mid-stream.
+    let mut convo: Vec<ChatMessage> = Vec::new();
+    for msg in req.messages {
+        if msg.role == "system" {
+            system_parts.push(anthropic_content_to_text(msg.content));
+            continue;
+        }
+        push_translated_message(&mut convo, &msg.role, msg.content);
+    }
+
+    let mut messages = Vec::new();
+    if !system_parts.is_empty() {
         messages.push(ChatMessage {
             role: "system".into(),
-            content: MessageContent::Text(content),
+            content: MessageContent::Text(system_parts.join("\n\n")),
             extra: Value::Null,
         });
     }
+    messages.extend(convo);
 
-    // Convert message roles and content.
-    for msg in req.messages {
-        let content = match msg.content {
-            AnthropicContent::Text(t) => MessageContent::Text(t),
-            AnthropicContent::Blocks(blocks) => {
-                // For simple text-only blocks, extract the text.
-                // For mixed content (images, etc.), pass as parts.
-                if blocks.len() == 1 && blocks[0].block_type == "text" {
-                    let text = blocks[0]
-                        .data
-                        .get("text")
-                        .and_then(|v| v.as_str())
-                        .unwrap_or("")
-                        .to_string();
-                    MessageContent::Text(text)
-                } else {
-                    MessageContent::Parts(blocks.into_iter().map(|b| json!(b)).collect())
+    // Reshape `tools` / `tool_choice` (carried over from the request's
+    // flattened `extra`) into the OpenAI shape neuron's chat template
+    // expects. Computed-then-inserted to avoid borrowing `obj` across
+    // the mutation.
+    let mut extra = req.extra;
+    if let Value::Object(obj) = &mut extra {
+        let tools = obj.get("tools").and_then(anthropic_tools_to_openai);
+        if let Some(tools) = tools {
+            obj.insert("tools".into(), tools);
         }
+        let tool_choice = obj
+            .get("tool_choice")
+            .and_then(anthropic_tool_choice_to_openai);
+        if let Some(tc) = tool_choice {
+            obj.insert("tool_choice".into(), tc);
         }
-        };
-        messages.push(ChatMessage {
-            role: msg.role,
-            content,
-            extra: Value::Null,
-        });
     }
 
     ChatCompletionRequest {
@@ -61,7 +103,278 @@ pub fn anthropic_to_openai(req: MessagesRequest) -> ChatCompletionRequest {
         top_p: req.top_p,
         max_tokens: Some(req.max_tokens),
         stream: req.stream,
-        extra: req.extra,
+        extra,
+    }
+}
+
+/// Translate one Anthropic message into one-or-more OpenAI messages,
+/// appending them to `out`.
+fn push_translated_message(out: &mut Vec<ChatMessage>, role: &str, content: AnthropicContent) {
+    let blocks = match content {
+        AnthropicContent::Text(t) => {
+            out.push(ChatMessage {
+                role: role.into(),
+                content: MessageContent::Text(t),
+                extra: Value::Null,
+            });
+            return;
+        }
+        AnthropicContent::Blocks(blocks) => blocks,
+    };
+
+    let mut text_segments: Vec<String> = Vec::new();
+    let mut parts: Vec<Value> = Vec::new();
+    let mut has_nontext_part = false;
+    let mut tool_calls: Vec<Value> = Vec::new();
+    let mut tool_msgs: Vec<ChatMessage> = Vec::new();
+
+    for block in blocks {
+        match block.block_type.as_str() {
+            "text" => {
+                let t = block
+                    .data
+                    .get("text")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default()
+                    .to_string();
+                parts.push(json!({ "type": "text", "text": t }));
+                text_segments.push(t);
+            }
+            "tool_use" => {
+                let id = block
+                    .data
+                    .get("id")
+                    .and_then(Value::as_str)
+                    .unwrap_or("toolu_unknown");
+                let name = block
+                    .data
+                    .get("name")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default();
+                let input = block
+                    .data
+                    .get("input")
+                    .cloned()
+                    .unwrap_or_else(|| json!({}));
+                tool_calls.push(json!({
+                    "id": id,
+                    "type": "function",
+                    "function": {
+                        "name": name,
+                        // Arguments as the parsed OBJECT, not the OpenAI
+                        // JSON-string form. The Qwen3.6 chat template
+                        // iterates `tool_call.arguments | items` (treats
+                        // it as a dict); a string throws "cannot convert
+                        // value into pairs", making neuron fall back to a
+                        // tool-less prompt — which silently breaks
+                        // tool-call chaining the moment one tool call is
+                        // in the history.
+                        "arguments": input,
+                    }
+                }));
+            }
+            "tool_result" => {
+                let tool_use_id = block
+                    .data
+                    .get("tool_use_id")
+                    .and_then(Value::as_str)
+                    .unwrap_or("toolu_unknown");
+                tool_msgs.push(ChatMessage {
+                    role: "tool".into(),
+                    content: MessageContent::Text(tool_result_content_to_string(&block.data)),
+                    extra: json!({ "tool_call_id": tool_use_id }),
+                });
+            }
+            "image" => {
+                if let Some(part) = anthropic_image_to_openai(&block.data) {
+                    parts.push(part);
+                    has_nontext_part = true;
+                }
+            }
+            _ => {
+                // Unknown block kind: preserve it as a JSON part rather
+                // than silently dropping it.
+                parts.push(serde_json::to_value(&block).unwrap_or(Value::Null));
+                has_nontext_part = true;
+            }
+        }
+    }
+
+    // Tool results become standalone `role:"tool"` turns and must
+    // precede any residual content from the same Anthropic message.
+    out.append(&mut tool_msgs);
+
+    if !tool_calls.is_empty() {
+        // An assistant turn that invoked tools. OpenAI carries the calls
+        // in `tool_calls`; the visible text (if any) stays in `content`.
+        out.push(ChatMessage {
+            role: role.into(),
+            content: MessageContent::Text(text_segments.join("")),
+            extra: json!({ "tool_calls": tool_calls }),
+        });
+    } else if has_nontext_part {
+        // Mixed content (images): forward as OpenAI content parts.
+        out.push(ChatMessage {
+            role: role.into(),
+            content: MessageContent::Parts(parts),
+            extra: Value::Null,
+        });
+    } else if !text_segments.is_empty() {
+        out.push(ChatMessage {
+            role: role.into(),
+            content: MessageContent::Text(text_segments.join("")),
+            extra: Value::Null,
+        });
+    }
+    // else: the message was only tool_result blocks — already emitted
+    // as `role:"tool"` turns above, nothing residual to add.
+}
+
+/// Extract plain text from an Anthropic `tool_result` block's `content`
+/// (a string, or an array of `{type:"text", text}` blocks).
+fn tool_result_content_to_string(data: &Value) -> String {
+    match data.get("content") {
+        Some(Value::String(s)) => s.clone(),
+        Some(Value::Array(arr)) => arr
+            .iter()
+            .map(|b| {
+                if b.get("type").and_then(Value::as_str) == Some("text") {
+                    b.get("text")
+                        .and_then(Value::as_str)
+                        .unwrap_or_default()
+                        .to_string()
+                } else {
+                    b.to_string()
+                }
+            })
+            .collect::<Vec<_>>()
+            .join(""),
+        Some(other) => other.to_string(),
+        None => String::new(),
+    }
+}
+
+/// Convert an Anthropic image block's `data` (`{source:{...}}`) into an
+/// OpenAI `image_url` content part.
+fn anthropic_image_to_openai(data: &Value) -> Option<Value> {
+    let source = data.get("source")?;
+    match source
+        .get("type")
+        .and_then(Value::as_str)
+        .unwrap_or("base64")
+    {
+        "base64" => {
+            let media = source
+                .get("media_type")
+                .and_then(Value::as_str)
+                .unwrap_or("image/png");
+            let b64 = source
+                .get("data")
+                .and_then(Value::as_str)
+                .unwrap_or_default();
+            Some(json!({
+                "type": "image_url",
+                "image_url": { "url": format!("data:{media};base64,{b64}") }
+            }))
+        }
+        "url" => {
+            let url = source
+                .get("url")
+                .and_then(Value::as_str)
+                .unwrap_or_default();
+            Some(json!({ "type": "image_url", "image_url": { "url": url } }))
+        }
+        _ => None,
+    }
+}
+
+/// Reshape an Anthropic `tools` array into the OpenAI function-tool
+/// shape. Returns `None` if the value isn't an array (left untouched).
+fn anthropic_tools_to_openai(tools: &Value) -> Option<Value> {
+    let arr = tools.as_array()?;
+    let converted = arr
+        .iter()
+        .map(|t| {
+            // Already OpenAI-shaped (a client mixing conventions, or a
+            // re-translation): pass through unchanged.
+            if t.get("type").and_then(Value::as_str) == Some("function")
+                && t.get("function").is_some()
+            {
+                return t.clone();
+            }
+            let mut function = serde_json::Map::new();
+            function.insert("name".into(), t.get("name").cloned().unwrap_or(Value::Null));
+            if let Some(desc) = t.get("description") {
+                function.insert("description".into(), desc.clone());
+            }
+            function.insert(
+                "parameters".into(),
+                t.get("input_schema")
+                    .cloned()
+                    .unwrap_or_else(|| json!({ "type": "object" })),
+            );
+            json!({ "type": "function", "function": Value::Object(function) })
+        })
+        .collect();
+    Some(Value::Array(converted))
+}
+
+/// Map an Anthropic `tool_choice` to the OpenAI form.
+fn anthropic_tool_choice_to_openai(tc: &Value) -> Option<Value> {
+    match tc.get("type").and_then(Value::as_str)? {
+        "auto" => Some(json!("auto")),
+        "any" => Some(json!("required")),
+        "none" => Some(json!("none")),
+        "tool" => {
+            let name = tc.get("name").and_then(Value::as_str).unwrap_or_default();
+            Some(json!({ "type": "function", "function": { "name": name } }))
+        }
+        _ => None,
+    }
+}
+
+/// Flatten Anthropic system content blocks (`[{type:"text", text}]`)
+/// into a single string.
+fn system_blocks_to_text(blocks: &[Value]) -> String {
+    let joined = blocks
+        .iter()
+        .filter(|b| b.get("type").and_then(Value::as_str) == Some("text"))
+        .filter_map(|b| b.get("text").and_then(Value::as_str))
+        .collect::<Vec<_>>()
+        .join("\n");
+    if joined.is_empty() {
+        // Unusual shape — don't lose it.
+        serde_json::to_string(blocks).unwrap_or_default()
+    } else {
+        joined
+    }
+}
+
+/// Flatten an Anthropic message's content into plain text. Used to fold
+/// `role:"system"` conversation turns into the leading system message;
+/// non-text blocks (rare in a system turn) are JSON-stringified rather
+/// than dropped.
+fn anthropic_content_to_text(content: AnthropicContent) -> String {
+    match content {
+        AnthropicContent::Text(t) => t,
+        AnthropicContent::Blocks(blocks) => blocks
+            .iter()
+            .map(|b| {
+                if b.block_type == "text" {
+                    b.data
+                        .get("text")
+                        .and_then(Value::as_str)
+                        .unwrap_or_default()
+                        .to_string()
+                } else {
+                    serde_json::to_value(b)
+                        .ok()
+                        .map(|v| v.to_string())
+                        .unwrap_or_default()
+                }
+            })
+            .collect::<Vec<_>>()
+            .join("\n"),
     }
 }
 
@@ -85,6 +398,9 @@ pub fn openai_to_anthropic(resp: ChatCompletionResponse) -> MessagesResponse {
         prompt_tokens: 0,
         completion_tokens: 0,
         total_tokens: 0,
+        completion_tokens_details: None,
+        prompt_tokens_details: None,
+        helexa_timing: None,
     });
 
     MessagesResponse {
@@ -455,6 +771,9 @@ mod stream_tests {
             prompt_tokens: 225,
             completion_tokens: 42,
             total_tokens: 267,
+            completion_tokens_details: None,
+            prompt_tokens_details: None,
+            helexa_timing: None,
         });
         t.on_chunk(&usage_chunk);
         let fin = t.finish();
@@ -475,3 +794,238 @@ mod stream_tests {
         assert!(t2.finish().is_empty(), "second finish must emit nothing");
     }
 }
+
+#[cfg(test)]
+mod request_tests {
+    use super::*;
+    use crate::openai::MessageContent;
+
+    fn req(value: Value) -> MessagesRequest {
+        serde_json::from_value(value).expect("valid MessagesRequest")
+    }
+
+    #[test]
+    fn tool_definitions_reshape_to_openai_function_shape() {
+        let r = req(json!({
+            "model": "Qwen/Qwen3.6-27B",
+            "max_tokens": 1024,
+            "messages": [{"role": "user", "content": "read the file"}],
+            "tools": [{
+                "name": "Read",
+                "description": "Read a file",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {"path": {"type": "string"}},
+                    "required": ["path"]
+                }
+            }]
+        }));
+        let openai = anthropic_to_openai(r);
+        let tools = openai
+            .extra
+            .get("tools")
+            .and_then(Value::as_array)
+            .expect("tools array");
+        assert_eq!(tools.len(), 1);
+        let t = &tools[0];
+        assert_eq!(t["type"], "function");
+        assert_eq!(t["function"]["name"], "Read");
+        assert_eq!(t["function"]["description"], "Read a file");
+        // input_schema is renamed to parameters, contents preserved.
+        assert_eq!(
+            t["function"]["parameters"]["properties"]["path"]["type"],
+            "string"
+        );
+        assert!(t["function"].get("input_schema").is_none());
+    }
+
+    #[test]
+    fn tool_choice_maps_each_variant() {
+        let mk = |tc: Value| {
+            let r = req(json!({
+                "model": "m", "max_tokens": 8,
+                "messages": [{"role": "user", "content": "hi"}],
+                "tool_choice": tc
+            }));
+            anthropic_to_openai(r)
+                .extra
+                .get("tool_choice")
+                .cloned()
+                .unwrap()
+        };
+        assert_eq!(mk(json!({"type": "auto"})), json!("auto"));
+        assert_eq!(mk(json!({"type": "any"})), json!("required"));
+        assert_eq!(mk(json!({"type": "none"})), json!("none"));
+        assert_eq!(
+            mk(json!({"type": "tool", "name": "Read"})),
+            json!({"type": "function", "function": {"name": "Read"}})
+        );
+    }
+
+    #[test]
+    fn assistant_tool_use_block_becomes_openai_tool_calls() {
+        let r = req(json!({
+            "model": "m", "max_tokens": 8,
+            "messages": [{
+                "role": "assistant",
+                "content": [
+                    {"type": "text", "text": "Let me read it."},
+                    {"type": "tool_use", "id": "toolu_1", "name": "Read",
+                     "input": {"path": "/etc/hosts"}}
+                ]
+            }]
+        }));
+        let openai = anthropic_to_openai(r);
+        // One assistant message carrying both the text and the call.
+        let m = openai.messages.last().expect("a message");
+        assert_eq!(m.role, "assistant");
+        match &m.content {
+            MessageContent::Text(t) => assert_eq!(t, "Let me read it."),
+            other => panic!("expected text content, got {other:?}"),
+        }
+        let calls = m
+            .extra
+            .get("tool_calls")
+            .and_then(Value::as_array)
+            .expect("tool_calls");
+        assert_eq!(calls[0]["id"], "toolu_1");
+        assert_eq!(calls[0]["type"], "function");
+        assert_eq!(calls[0]["function"]["name"], "Read");
+        // arguments is the parsed object (Qwen3.6 template iterates it).
+        assert_eq!(
+            calls[0]["function"]["arguments"],
+            json!({"path": "/etc/hosts"})
+        );
+    }
+
+    #[test]
+    fn user_tool_result_block_becomes_role_tool_message() {
+        let r = req(json!({
+            "model": "m", "max_tokens": 8,
+            "messages": [{
+                "role": "user",
+                "content": [
+                    {"type": "tool_result", "tool_use_id": "toolu_1",
+                     "content": "127.0.0.1 localhost"}
+                ]
+            }]
+        }));
+        let openai = anthropic_to_openai(r);
+        assert_eq!(openai.messages.len(), 1);
+        let m = &openai.messages[0];
+        assert_eq!(m.role, "tool");
+        assert_eq!(m.extra["tool_call_id"], "toolu_1");
+        match &m.content {
+            MessageContent::Text(t) => assert_eq!(t, "127.0.0.1 localhost"),
+            other => panic!("expected text content, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn tool_result_with_block_array_content_is_flattened() {
+        let r = req(json!({
+            "model": "m", "max_tokens": 8,
+            "messages": [{
+                "role": "user",
+                "content": [
+                    {"type": "tool_result", "tool_use_id": "t",
+                     "content": [{"type": "text", "text": "line1"}, {"type": "text", "text": "line2"}]}
+                ]
+            }]
+        }));
+        let openai = anthropic_to_openai(r);
+        match &openai.messages[0].content {
+            MessageContent::Text(t) => assert_eq!(t, "line1line2"),
+            other => panic!("expected text, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn tool_result_then_text_emits_tool_turn_first() {
+        // A user turn that carries a tool result *and* a follow-up
+        // question must yield the tool message before the user text.
+        let r = req(json!({
+            "model": "m", "max_tokens": 8,
+            "messages": [{
+                "role": "user",
+                "content": [
+                    {"type": "tool_result", "tool_use_id": "t", "content": "ok"},
+                    {"type": "text", "text": "now what?"}
+                ]
+            }]
+        }));
+        let openai = anthropic_to_openai(r);
+        assert_eq!(openai.messages.len(), 2);
+        assert_eq!(openai.messages[0].role, "tool");
+        assert_eq!(openai.messages[1].role, "user");
+        match &openai.messages[1].content {
+            MessageContent::Text(t) => assert_eq!(t, "now what?"),
+            other => panic!("expected text, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn system_blocks_flatten_to_text_not_json() {
+        let r = req(json!({
+            "model": "m", "max_tokens": 8,
+            "system": [{"type": "text", "text": "You are helpful."}],
+            "messages": [{"role": "user", "content": "hi"}]
+        }));
+        let openai = anthropic_to_openai(r);
+        let sys = &openai.messages[0];
+        assert_eq!(sys.role, "system");
+        match &sys.content {
+            MessageContent::Text(t) => assert_eq!(t, "You are helpful."),
+            other => panic!("expected text, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn system_role_messages_merge_into_one_leading_system() {
+        // Claude Code's shape: a top-level `system` PLUS a `role:"system"`
+        // turn inside `messages` (which Qwen3.6's template rejects unless
+        // it's first). Both must merge into a single leading system msg.
+        let r = req(json!({
+            "model": "m", "max_tokens": 8,
+            "system": "TOP LEVEL SYSTEM",
+            "messages": [
+                {"role": "user", "content": "hello"},
+                {"role": "system", "content": "INJECTED SYSTEM"},
+                {"role": "user", "content": "do it"}
+            ],
+            "tools": [{"name": "noop", "input_schema": {"type": "object"}}]
+        }));
+        let openai = anthropic_to_openai(r);
+        // Exactly one system message, at index 0, merging both parts.
+        let systems: Vec<usize> = openai
+            .messages
+            .iter()
+            .enumerate()
+            .filter(|(_, m)| m.role == "system")
+            .map(|(i, _)| i)
+            .collect();
+        assert_eq!(systems, vec![0], "one system message, at the front");
+        match &openai.messages[0].content {
+            MessageContent::Text(t) => {
+                assert!(t.contains("TOP LEVEL SYSTEM"));
+                assert!(t.contains("INJECTED SYSTEM"));
+            }
+            other => panic!("expected text, got {other:?}"),
+        }
+        // The two real user turns survive, in order, after the system.
+        let roles: Vec<&str> = openai.messages.iter().map(|m| m.role.as_str()).collect();
+        assert_eq!(roles, vec!["system", "user", "user"]);
+    }
+
+    #[test]
+    fn already_openai_shaped_tools_pass_through() {
+        let r = req(json!({
+            "model": "m", "max_tokens": 8,
+            "messages": [{"role": "user", "content": "hi"}],
+            "tools": [{"type": "function", "function": {"name": "x", "parameters": {}}}]
+        }));
+        let openai = anthropic_to_openai(r);
+        let tools = openai.extra.get("tools").and_then(Value::as_array).unwrap();
+        assert_eq!(tools[0]["function"]["name"], "x");
+    }
+}

crates/cortex-gateway/Cargo.toml

@@ -6,6 +6,8 @@ license.workspace = true
 
 [dependencies]
 cortex-core.workspace = true
+helexa-stream = { path = "../helexa-stream" }
+async-trait.workspace = true
 tokio.workspace = true
 axum.workspace = true
 tower.workspace = true

crates/cortex-gateway/src/anthropic_sse.rs

@@ -32,6 +32,8 @@ pub async fn stream_translated(
     openai_body: axum::body::Bytes,
     model_id: &str,
     node_name: &str,
+    inbound_headers: &axum::http::HeaderMap,
+    usage_sink: Option<crate::metering::UsageSink>,
 ) -> Response {
     let url = format!("{endpoint}/v1/chat/completions");
     tracing::info!(
@@ -42,13 +44,14 @@ pub async fn stream_translated(
         "proxying streaming request (anthropic SSE translation)"
     );
 
-    let upstream = match client
+    let request = crate::auth::forward_principal_headers(
+        client
             .post(&url)
             .header("content-type", "application/json")
-        .body(openai_body)
-        .send()
-        .await
-    {
+            .body(openai_body),
+        inbound_headers,
+    );
+    let upstream = match request.send().await {
         Ok(r) => r,
         Err(e) => {
             tracing::warn!(
@@ -82,11 +85,22 @@ pub async fn stream_translated(
     // discipline as neuron's own projectors.
     let (tx, rx) = tokio::sync::mpsc::channel::<Result<Bytes, std::convert::Infallible>>(32);
     let node = node_name.to_string();
+    let model = model_id.to_string();
     tokio::spawn(async move {
         let mut upstream = upstream.bytes_stream();
         let mut translator = AnthropicStreamTranslator::new();
         let mut buf: Vec<u8> = Vec::new();
         let mut done = false;
+        // Wire-debug accounting for the stream summary emitted at the
+        // end: did the model emit a structured tool call, what was the
+        // final finish_reason, and how many upstream frames did we see.
+        let mut saw_tool_call = false;
+        let mut last_finish: Option<String> = None;
+        let mut frames = 0u64;
+        // Engine-truth usage for metering (#51), scanned from the upstream
+        // frames (neuron emits a final `usage` object on the stream, #48).
+        let mut usage_prompt = 0u64;
+        let mut usage_completion = 0u64;
 
         'outer: while let Some(block) = upstream.next().await {
             let block = match block {
@@ -113,10 +127,31 @@ pub async fn stream_translated(
                         }
                         continue;
                     }
+                    tracing::trace!(node = %node, frame = %data, "anthropic stream: upstream frame");
+                    // Capture usage for metering before translation — the
+                    // usage object rides on a late frame (often after the
+                    // last content delta).
+                    if let Some(p) = crate::proxy::last_count_for(data, "prompt_tokens") {
+                        usage_prompt = p;
+                    }
+                    if let Some(c) = crate::proxy::last_count_for(data, "completion_tokens") {
+                        usage_completion = c;
+                    }
                     let Ok(chunk) = serde_json::from_str::<ChatCompletionChunk>(data) else {
                         tracing::debug!(node = %node, "anthropic stream: unparsable upstream frame skipped");
                         continue;
                     };
+                    frames += 1;
+                    if chunk
+                        .choices
+                        .iter()
+                        .any(|c| c.delta.get("tool_calls").is_some())
+                    {
+                        saw_tool_call = true;
+                    }
+                    if let Some(fr) = chunk.choices.iter().find_map(|c| c.finish_reason.clone()) {
+                        last_finish = Some(fr);
+                    }
                     if !send_frames(&tx, translator.on_chunk(&chunk)).await {
                         break 'outer;
                     }
@@ -129,6 +164,28 @@ pub async fn stream_translated(
         if !done {
             let _ = send_frames(&tx, translator.finish()).await;
         }
+        // Stream summary: the streaming counterpart to the non-streaming
+        // handler's "upstream response" line. `upstream_tool_calls =
+        // false` on a tools-bearing request is the fingerprint of the
+        // model improvising an unparsed tool-call format.
+        tracing::debug!(
+            wire = "anthropic",
+            model = %model,
+            node = %node,
+            frames,
+            upstream_tool_calls = saw_tool_call,
+            finish_reason = ?last_finish,
+            terminated = done,
+            "anthropic stream complete"
+        );
+
+        // Settle metering with the observed usage (#51). Runs on every exit
+        // path of the pump — clean end, early break, or upstream error — so
+        // the reservation is always resolved. `(0, 0)` when no usage frame
+        // was seen, which releases without recording spend.
+        if let Some(sink) = usage_sink {
+            sink(usage_prompt, usage_completion);
+        }
     });
 
     Response::builder()

crates/cortex-gateway/src/auth.rs

@@ -0,0 +1,144 @@
+//! API-key authentication + principal resolution (#49).
+//!
+//! Identity rides standard bearer auth only — `Authorization: Bearer <key>`
+//! — which is what keeps every tier OpenAI-compatible by construction (no
+//! custom required headers or body fields, per #47). The middleware resolves
+//! the key to a [`Principal`] via the [`EntitlementProvider`], carries it in
+//! the request extensions for cortex-side metering/enforcement (#51/#52), and
+//! stamps it as internal headers on the request so it reaches neuron, which
+//! trusts cortex's assertion over WireGuard (#54).
+//!
+//! Anti-spoofing: any client-supplied principal header is **stripped** before
+//! the authoritative value is stamped, so a client can never assert a
+//! principal it didn't authenticate as.
+//!
+//! Rejection contract (#63): missing key under `require_auth`, or any present
+//! but unresolvable key, yields `401 invalid_api_key` in the #60 envelope.
+
+use crate::error::envelope_response;
+use crate::state::CortexState;
+use axum::extract::{Request, State};
+use axum::http::header::AUTHORIZATION;
+use axum::http::{HeaderMap, HeaderValue};
+use axum::middleware::Next;
+use axum::response::Response;
+use cortex_core::entitlements::{AuthError, HEADER_ACCOUNT_ID, HEADER_KEY_ID};
+use cortex_core::error_envelope::OpenAiError;
+use std::sync::Arc;
+
+/// Endpoints that never require auth: liveness/readiness probes. Everything
+/// else flows through resolution.
+fn is_public(path: &str) -> bool {
+    path == "/health" || path == "/"
+}
+
+/// Extract the bearer token from an `Authorization` header value, if present
+/// and well-formed. Scheme match is case-insensitive per RFC 7235.
+fn parse_bearer(headers: &HeaderMap) -> Option<String> {
+    let raw = headers.get(AUTHORIZATION)?.to_str().ok()?;
+    let (scheme, token) = raw.split_once(' ')?;
+    if scheme.eq_ignore_ascii_case("bearer") {
+        let token = token.trim();
+        (!token.is_empty()).then(|| token.to_string())
+    } else {
+        None
+    }
+}
+
+/// Axum middleware: resolve the bearer key, attach the principal, stamp the
+/// internal headers. Wired in `build_app` via `from_fn_with_state`.
+pub async fn require_principal(
+    State(fleet): State<Arc<CortexState>>,
+    mut req: Request,
+    next: Next,
+) -> Response {
+    if is_public(req.uri().path()) {
+        return next.run(req).await;
+    }
+
+    // Anti-spoof: drop any client-supplied principal headers up front.
+    {
+        let headers = req.headers_mut();
+        headers.remove(HEADER_ACCOUNT_ID);
+        headers.remove(HEADER_KEY_ID);
+    }
+
+    match parse_bearer(req.headers()) {
+        Some(key) => match fleet.entitlements.resolve(&key).await {
+            Ok(principal) => {
+                // Stamp the authoritative principal for neuron. Account/key
+                // ids come from operator config, so they're valid header
+                // values; guard anyway and skip a malformed one rather than
+                // panic.
+                if let (Ok(account), Ok(key_id)) = (
+                    HeaderValue::from_str(&principal.account_id),
+                    HeaderValue::from_str(&principal.key_id),
+                ) {
+                    let headers = req.headers_mut();
+                    headers.insert(HEADER_ACCOUNT_ID, account);
+                    headers.insert(HEADER_KEY_ID, key_id);
+                }
+                // Carry the typed principal for cortex-side metering (#51)
+                // and budget enforcement (#52).
+                req.extensions_mut().insert(principal);
+                next.run(req).await
+            }
+            // The entitlement authority is unreachable (upstream client
+            // blip, #57). Fail **closed but distinct**: a transient outage
+            // must not reject a real key as `401 invalid_api_key` — it's a
+            // retryable `503`. This holds regardless of require_auth: we
+            // can't safely serve a key we couldn't authorize.
+            Err(AuthError::Unavailable { retry_after_secs }) => {
+                envelope_response(OpenAiError::service_unavailable(
+                    "entitlement authority temporarily unavailable",
+                    Some(retry_after_secs),
+                ))
+            }
+            // A genuinely unrecognized key only hard-fails when auth is
+            // *required*. In allow-anonymous mode (the default) we IGNORE it
+            // and serve unauthenticated — otherwise the placeholder keys that
+            // OpenAI-compatible clients send by default (opencode, Open WebUI,
+            // Agent Zero, litellm) would all break though the operator never
+            // opted into auth. Pre-#49 the bearer was never inspected; this
+            // preserves that for require_auth=false.
+            Err(AuthError::InvalidKey) => {
+                if fleet.require_auth {
+                    unauthorized("invalid API key")
+                } else {
+                    tracing::debug!(
+                        "ignoring unrecognized bearer token (require_auth=false): serving anonymously"
+                    );
+                    next.run(req).await
+                }
+            }
+        },
+        None => {
+            if fleet.require_auth {
+                unauthorized("missing API key; supply 'Authorization: Bearer <key>'")
+            } else {
+                next.run(req).await
+            }
+        }
+    }
+}
+
+/// `401 invalid_api_key` in the standard envelope (#63).
+fn unauthorized(message: &str) -> Response {
+    envelope_response(OpenAiError::invalid_api_key(message))
+}
+
+/// Copy the cortex-stamped principal headers from an inbound [`HeaderMap`]
+/// onto an outbound reqwest builder. Used by the Anthropic proxy paths,
+/// which construct their own upstream requests instead of going through
+/// [`crate::proxy::forward_request`] (which forwards all headers verbatim).
+pub fn forward_principal_headers(
+    mut builder: reqwest::RequestBuilder,
+    headers: &HeaderMap,
+) -> reqwest::RequestBuilder {
+    for name in [HEADER_ACCOUNT_ID, HEADER_KEY_ID] {
+        if let Some(value) = headers.get(name) {
+            builder = builder.header(name, value);
+        }
+    }
+    builder
+}

crates/cortex-gateway/src/entitlements_chain.rs

@@ -0,0 +1,112 @@
+//! Chained entitlement provider (#57): operator-local keys first, mesh
+//! upstream for everything else.
+//!
+//! `resolve` tries the [`LocalEntitlementProvider`] (operator + infra keys —
+//! never a network hop); only a locally-unknown key falls through to
+//! [`UpstreamEntitlementProvider`]. Because the local provider treats an
+//! unconfigured principal as uncapped, reserve/settle/release/snapshot must
+//! **not** blindly hit local — they dispatch to whichever backend resolved
+//! that account, remembered in a map keyed by `account_id` (populated at
+//! resolve time).
+
+use crate::entitlements_local::LocalEntitlementProvider;
+use crate::entitlements_upstream::UpstreamEntitlementProvider;
+use async_trait::async_trait;
+use cortex_core::entitlements::{
+    AuthError, BudgetError, BudgetSnapshot, EntitlementProvider, Principal, Reservation,
+};
+use std::collections::HashMap;
+use tokio::sync::RwLock;
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+enum Backend {
+    Local,
+    Upstream,
+}
+
+pub struct ChainedEntitlementProvider {
+    local: LocalEntitlementProvider,
+    upstream: UpstreamEntitlementProvider,
+    /// account_id → which backend owns it, learned at resolve time.
+    backends: RwLock<HashMap<String, Backend>>,
+}
+
+impl ChainedEntitlementProvider {
+    pub fn new(local: LocalEntitlementProvider, upstream: UpstreamEntitlementProvider) -> Self {
+        Self {
+            local,
+            upstream,
+            backends: RwLock::new(HashMap::new()),
+        }
+    }
+
+    async fn record(&self, account_id: &str, backend: Backend) {
+        self.backends
+            .write()
+            .await
+            .insert(account_id.to_string(), backend);
+    }
+
+    /// The backend that owns `account_id`. Defaults to `Upstream` for an
+    /// account never resolved this process-lifetime (a resolve always
+    /// precedes reserve in a request, so this is just a safe fallback —
+    /// upstream fails closed if the account is bogus).
+    async fn backend_for(&self, account_id: &str) -> Backend {
+        self.backends
+            .read()
+            .await
+            .get(account_id)
+            .copied()
+            .unwrap_or(Backend::Upstream)
+    }
+}
+
+#[async_trait]
+impl EntitlementProvider for ChainedEntitlementProvider {
+    async fn resolve(&self, api_key: &str) -> Result<Principal, AuthError> {
+        match self.local.resolve(api_key).await {
+            Ok(p) => {
+                self.record(&p.account_id, Backend::Local).await;
+                Ok(p)
+            }
+            Err(AuthError::InvalidKey) => {
+                let p = self.upstream.resolve(api_key).await?;
+                self.record(&p.account_id, Backend::Upstream).await;
+                Ok(p)
+            }
+            Err(e) => Err(e),
+        }
+    }
+
+    async fn reserve(
+        &self,
+        principal: &Principal,
+        max_tokens: u64,
+    ) -> Result<Reservation, BudgetError> {
+        match self.backend_for(&principal.account_id).await {
+            Backend::Local => self.local.reserve(principal, max_tokens).await,
+            Backend::Upstream => self.upstream.reserve(principal, max_tokens).await,
+        }
+    }
+
+    async fn settle(&self, reservation: Reservation, actual_tokens: u64) {
+        match self.backend_for(&reservation.principal.account_id).await {
+            Backend::Local => self.local.settle(reservation, actual_tokens).await,
+            Backend::Upstream => self.upstream.settle(reservation, actual_tokens).await,
+        }
+    }
+
+    async fn release(&self, reservation: Reservation) {
+        match self.backend_for(&reservation.principal.account_id).await {
+            Backend::Local => self.local.release(reservation).await,
+            Backend::Upstream => self.upstream.release(reservation).await,
+        }
+    }
+
+    async fn snapshot(&self, principal: &Principal) -> Option<BudgetSnapshot> {
+        match self.backend_for(&principal.account_id).await {
+            Backend::Local => self.local.snapshot(principal).await,
+            Backend::Upstream => self.upstream.snapshot(principal).await,
+        }
+    }
+}

crates/cortex-gateway/src/entitlements_local.rs

@@ -0,0 +1,317 @@
+//! The local/static [`EntitlementProvider`] (#50).
+//!
+//! Accounts, keys, and hard caps come from operator config
+//! ([`cortex_core::config::EntitlementsConfig`]); reservations and settled
+//! spend are tracked in-process. This lands auth + per-key caps + the
+//! amplification fix before any upstream clearing house exists; the future
+//! helexa-upstream client (#57) implements the same trait.
+//!
+//! Budget math is serialized under a single [`std::sync::Mutex`] so
+//! reserve/settle/release are atomic — a key's `spent + reserved` can never
+//! exceed its hard cap even under concurrent requests (the #52 guarantee).
+//! The lock is held only for the in-memory arithmetic, never across an
+//! await.
+
+use cortex_core::config::{ApiKeyConfig, EntitlementsConfig};
+use cortex_core::entitlements::{
+    AuthError, BudgetError, BudgetSnapshot, CapWindow, EntitlementProvider, Principal, Reservation,
+};
+use std::collections::HashMap;
+use std::sync::Mutex;
+use std::sync::atomic::{AtomicU64, Ordering};
+use std::time::Instant;
+
+/// Per-key budget configuration (resolved from [`ApiKeyConfig`]).
+struct Budget {
+    hard_cap: Option<u64>,
+    window: CapWindow,
+}
+
+/// Live, mutable accounting for one key over its current window.
+#[derive(Default)]
+struct Ledger {
+    /// Settled spend in the current window.
+    spent: u64,
+    /// Sum of outstanding (un-settled) reservations.
+    reserved: u64,
+    /// Start of the current rolling window; `None` until the first reserve.
+    /// Unused for [`CapWindow::Balance`].
+    window_start: Option<Instant>,
+}
+
+pub struct LocalEntitlementProvider {
+    /// Bearer token → principal.
+    keys: HashMap<String, Principal>,
+    /// `key_id` → budget config.
+    budgets: HashMap<String, Budget>,
+    /// `key_id` → live ledger.
+    ledgers: Mutex<HashMap<String, Ledger>>,
+    /// Monotonic source of opaque reservation handles.
+    next_id: AtomicU64,
+}
+
+impl LocalEntitlementProvider {
+    /// Build from the `[entitlements]` config. A key without an explicit
+    /// `key_id` is tracked at `account_id` granularity (its secret is never
+    /// used as a label).
+    pub fn from_config(config: &EntitlementsConfig) -> Self {
+        let mut keys = HashMap::new();
+        let mut budgets = HashMap::new();
+        for ApiKeyConfig {
+            key,
+            account_id,
+            key_id,
+            hard_cap,
+            window,
+        } in &config.keys
+        {
+            let key_id = key_id.clone().unwrap_or_else(|| account_id.clone());
+            keys.insert(
+                key.clone(),
+                Principal {
+                    account_id: account_id.clone(),
+                    key_id: key_id.clone(),
+                },
+            );
+            budgets.insert(
+                key_id,
+                Budget {
+                    hard_cap: *hard_cap,
+                    window: window.clone(),
+                },
+            );
+        }
+        Self {
+            keys,
+            budgets,
+            ledgers: Mutex::new(HashMap::new()),
+            next_id: AtomicU64::new(1),
+        }
+    }
+}
+
+/// Tokens still available under `cap` given current `spent`/`reserved`.
+/// `None` cap = unlimited.
+fn available(cap: Option<u64>, spent: u64, reserved: u64) -> Option<u64> {
+    cap.map(|c| c.saturating_sub(spent).saturating_sub(reserved))
+}
+
+#[async_trait::async_trait]
+impl EntitlementProvider for LocalEntitlementProvider {
+    async fn resolve(&self, api_key: &str) -> Result<Principal, AuthError> {
+        self.keys.get(api_key).cloned().ok_or(AuthError::InvalidKey)
+    }
+
+    async fn reserve(
+        &self,
+        principal: &Principal,
+        max_tokens: u64,
+    ) -> Result<Reservation, BudgetError> {
+        // A principal with no configured budget (or an uncapped one) always
+        // reserves; we still track spend for metrics.
+        let budget = self.budgets.get(&principal.key_id);
+        let (cap, window) = match budget {
+            Some(b) => (b.hard_cap, b.window.clone()),
+            None => (None, CapWindow::Balance),
+        };
+
+        let mut ledgers = self.ledgers.lock().expect("ledger mutex poisoned");
+        let ledger = ledgers.entry(principal.key_id.clone()).or_default();
+
+        // Lazily reset a rolling window that has elapsed before checking.
+        let mut retry_after_secs = 0;
+        if let CapWindow::Rolling { seconds } = window {
+            let now = Instant::now();
+            match ledger.window_start {
+                Some(start) if now.duration_since(start).as_secs() < seconds => {
+                    retry_after_secs = seconds - now.duration_since(start).as_secs();
+                }
+                _ => {
+                    // First reserve, or the window has fully elapsed: reset.
+                    ledger.spent = 0;
+                    ledger.window_start = Some(now);
+                    retry_after_secs = seconds;
+                }
+            }
+        }
+
+        if let Some(avail) = available(cap, ledger.spent, ledger.reserved)
+            && max_tokens > avail
+        {
+            return Err(match window {
+                CapWindow::Rolling { .. } => BudgetError::RateLimited {
+                    requested: max_tokens,
+                    available: avail,
+                    // At least 1s so clients don't hot-loop on a sub-second
+                    // remainder.
+                    retry_after_secs: retry_after_secs.max(1),
+                },
+                CapWindow::Balance => BudgetError::InsufficientQuota {
+                    requested: max_tokens,
+                    available: avail,
+                },
+            });
+        }
+
+        ledger.reserved += max_tokens;
+        Ok(Reservation {
+            id: self.next_id.fetch_add(1, Ordering::Relaxed),
+            principal: principal.clone(),
+            reserved: max_tokens,
+        })
+    }
+
+    async fn settle(&self, reservation: Reservation, actual_tokens: u64) {
+        let mut ledgers = self.ledgers.lock().expect("ledger mutex poisoned");
+        if let Some(ledger) = ledgers.get_mut(&reservation.principal.key_id) {
+            ledger.reserved = ledger.reserved.saturating_sub(reservation.reserved);
+            ledger.spent += actual_tokens;
+        }
+    }
+
+    async fn release(&self, reservation: Reservation) {
+        let mut ledgers = self.ledgers.lock().expect("ledger mutex poisoned");
+        if let Some(ledger) = ledgers.get_mut(&reservation.principal.key_id) {
+            ledger.reserved = ledger.reserved.saturating_sub(reservation.reserved);
+        }
+    }
+
+    async fn snapshot(&self, principal: &Principal) -> Option<BudgetSnapshot> {
+        let ledgers = self.ledgers.lock().expect("ledger mutex poisoned");
+        let (spent, reserved) = ledgers
+            .get(&principal.key_id)
+            .map(|l| (l.spent, l.reserved))
+            .unwrap_or((0, 0));
+        let hard_cap = self.budgets.get(&principal.key_id).and_then(|b| b.hard_cap);
+        Some(BudgetSnapshot {
+            hard_cap,
+            spent,
+            reserved,
+        })
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn provider() -> LocalEntitlementProvider {
+        let config = EntitlementsConfig {
+            require_auth: true,
+            keys: vec![
+                ApiKeyConfig {
+                    key: "sk-balance".into(),
+                    account_id: "acct-a".into(),
+                    key_id: Some("key-balance".into()),
+                    hard_cap: Some(1_000),
+                    window: CapWindow::Balance,
+                },
+                ApiKeyConfig {
+                    key: "sk-rolling".into(),
+                    account_id: "acct-b".into(),
+                    key_id: Some("key-rolling".into()),
+                    hard_cap: Some(500),
+                    window: CapWindow::Rolling { seconds: 3_600 },
+                },
+                ApiKeyConfig {
+                    key: "sk-infra".into(),
+                    account_id: "operator".into(),
+                    key_id: Some("key-infra".into()),
+                    hard_cap: None,
+                    window: CapWindow::Balance,
+                },
+            ],
+        };
+        LocalEntitlementProvider::from_config(&config)
+    }
+
+    #[tokio::test]
+    async fn resolves_configured_key_to_principal() {
+        let p = provider();
+        let principal = p.resolve("sk-balance").await.expect("known key resolves");
+        assert_eq!(principal.account_id, "acct-a");
+        assert_eq!(principal.key_id, "key-balance");
+    }
+
+    #[tokio::test]
+    async fn unknown_key_is_invalid() {
+        let p = provider();
+        assert!(matches!(
+            p.resolve("sk-nope").await,
+            Err(AuthError::InvalidKey)
+        ));
+    }
+
+    #[tokio::test]
+    async fn reserve_settle_release_round_trip() {
+        let p = provider();
+        let principal = p.resolve("sk-balance").await.unwrap();
+
+        let r = p.reserve(&principal, 400).await.expect("within cap");
+        // Reserved, not yet spent.
+        let snap = p.snapshot(&principal).await.unwrap();
+        assert_eq!(snap.hard_cap, Some(1_000));
+        assert_eq!(snap.reserved, 400);
+        assert_eq!(snap.spent, 0);
+
+        // Used fewer tokens than reserved → remainder released, spend exact.
+        p.settle(r, 250).await;
+        let snap = p.snapshot(&principal).await.unwrap();
+        assert_eq!(snap.reserved, 0);
+        assert_eq!(snap.spent, 250);
+
+        // A reservation that is released contributes no spend.
+        let r2 = p.reserve(&principal, 100).await.unwrap();
+        p.release(r2).await;
+        let snap = p.snapshot(&principal).await.unwrap();
+        assert_eq!(snap.reserved, 0);
+        assert_eq!(snap.spent, 250);
+    }
+
+    #[tokio::test]
+    async fn balance_over_cap_is_insufficient_quota() {
+        let p = provider();
+        let principal = p.resolve("sk-balance").await.unwrap();
+        // Reserve most of the cap, then ask for more than remains.
+        let _r = p.reserve(&principal, 900).await.unwrap();
+        let err = p.reserve(&principal, 200).await.expect_err("over cap");
+        match err {
+            BudgetError::InsufficientQuota {
+                requested,
+                available,
+            } => {
+                assert_eq!(requested, 200);
+                assert_eq!(available, 100);
+            }
+            other => panic!("expected InsufficientQuota, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn rolling_over_cap_is_rate_limited_with_retry_after() {
+        let p = provider();
+        let principal = p.resolve("sk-rolling").await.unwrap();
+        let _r = p.reserve(&principal, 500).await.unwrap();
+        let err = p.reserve(&principal, 1).await.expect_err("over cap");
+        match err {
+            BudgetError::RateLimited {
+                retry_after_secs, ..
+            } => {
+                assert!(retry_after_secs >= 1, "must advertise a retry hint");
+                assert!(retry_after_secs <= 3_600);
+            }
+            other => panic!("expected RateLimited, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn uncapped_infra_key_never_refuses() {
+        let p = provider();
+        let principal = p.resolve("sk-infra").await.unwrap();
+        let r = p.reserve(&principal, 10_000_000).await.expect("uncapped");
+        p.settle(r, 10_000_000).await;
+        let snap = p.snapshot(&principal).await.unwrap();
+        assert_eq!(snap.hard_cap, None);
+        assert_eq!(snap.spent, 10_000_000);
+    }
+}

crates/cortex-gateway/src/entitlements_upstream.rs

@@ -0,0 +1,246 @@
+//! helexa-upstream client (#57): an [`EntitlementProvider`] that resolves
+//! keys and reserves/settles budget against the mesh authority's
+//! `/authz/v1` surface (B2). It is "just another impl of the trait" — cortex
+//! enforcement (`auth.rs`, `metering.rs`) is unchanged.
+//!
+//! **Fail closed.** When upstream is unreachable, `resolve` returns
+//! [`AuthError::Unavailable`] (→ `503`, never `401`) and `reserve` refuses
+//! with a retryable [`BudgetError::RateLimited`] — a request is never served
+//! on an un-authorized key, and a real key is never rejected as invalid
+//! during a blip.
+
+use async_trait::async_trait;
+use cortex_core::config::UpstreamClientConfig;
+use cortex_core::entitlements::{
+    AuthError, BudgetError, BudgetSnapshot, EntitlementProvider, Principal, Reservation,
+};
+use serde::Deserialize;
+use std::time::Duration;
+
+/// Retry-After (seconds) advertised when we fail closed on an upstream
+/// outage.
+const FAIL_CLOSED_RETRY_SECS: u64 = 5;
+
+pub struct UpstreamEntitlementProvider {
+    client: reqwest::Client,
+    base_url: String,
+    bearer: String,
+}
+
+#[derive(Deserialize)]
+struct PrincipalDto {
+    account_id: String,
+    key_id: String,
+}
+#[derive(Deserialize)]
+struct SnapshotDto {
+    hard_cap: Option<u64>,
+    spent: u64,
+    reserved: u64,
+}
+#[derive(Deserialize)]
+struct ResolveResp {
+    principal: PrincipalDto,
+    #[allow(dead_code)]
+    snapshot: Option<SnapshotDto>,
+}
+#[derive(Deserialize)]
+struct ReserveResp {
+    reservation_id: Option<i64>,
+    rejected: Option<Rejection>,
+}
+#[derive(Deserialize)]
+#[serde(tag = "kind", rename_all = "snake_case")]
+enum Rejection {
+    InsufficientQuota {
+        requested: u64,
+        available: u64,
+    },
+    RateLimited {
+        requested: u64,
+        available: u64,
+        retry_after_secs: u64,
+    },
+}
+
+impl UpstreamEntitlementProvider {
+    pub fn new(cfg: &UpstreamClientConfig) -> Self {
+        let client = reqwest::Client::builder()
+            .timeout(Duration::from_secs(cfg.timeout_secs))
+            .build()
+            .expect("failed to build upstream HTTP client");
+        Self {
+            client,
+            base_url: cfg.url.trim_end_matches('/').to_string(),
+            bearer: cfg.bearer.clone(),
+        }
+    }
+
+    fn url(&self, path: &str) -> String {
+        format!("{}{}", self.base_url, path)
+    }
+}
+
+#[async_trait]
+impl EntitlementProvider for UpstreamEntitlementProvider {
+    async fn resolve(&self, api_key: &str) -> Result<Principal, AuthError> {
+        let resp = self
+            .client
+            .post(self.url("/authz/v1/resolve"))
+            .bearer_auth(&self.bearer)
+            .json(&serde_json::json!({ "api_key": api_key }))
+            .send()
+            .await;
+        let resp = match resp {
+            Ok(r) => r,
+            Err(e) => {
+                tracing::warn!(error = %e, "upstream resolve unreachable; failing closed");
+                return Err(AuthError::Unavailable {
+                    retry_after_secs: FAIL_CLOSED_RETRY_SECS,
+                });
+            }
+        };
+        if resp.status().as_u16() == 401 {
+            return Err(AuthError::InvalidKey);
+        }
+        if !resp.status().is_success() {
+            return Err(AuthError::Unavailable {
+                retry_after_secs: FAIL_CLOSED_RETRY_SECS,
+            });
+        }
+        match resp.json::<ResolveResp>().await {
+            Ok(r) => Ok(Principal {
+                account_id: r.principal.account_id,
+                key_id: r.principal.key_id,
+            }),
+            Err(e) => {
+                tracing::warn!(error = %e, "upstream resolve: bad body; failing closed");
+                Err(AuthError::Unavailable {
+                    retry_after_secs: FAIL_CLOSED_RETRY_SECS,
+                })
+            }
+        }
+    }
+
+    async fn reserve(
+        &self,
+        principal: &Principal,
+        max_tokens: u64,
+    ) -> Result<Reservation, BudgetError> {
+        let fail_closed = || BudgetError::RateLimited {
+            requested: max_tokens,
+            available: 0,
+            retry_after_secs: FAIL_CLOSED_RETRY_SECS,
+        };
+        let resp = self
+            .client
+            .post(self.url("/authz/v1/reserve"))
+            .bearer_auth(&self.bearer)
+            .json(&serde_json::json!({
+                "account_id": principal.account_id,
+                "key_id": principal.key_id,
+                "max_tokens": max_tokens,
+            }))
+            .send()
+            .await;
+        let resp = match resp {
+            Ok(r) if r.status().is_success() => r,
+            Ok(r) => {
+                tracing::warn!(status = %r.status(), "upstream reserve non-2xx; failing closed");
+                return Err(fail_closed());
+            }
+            Err(e) => {
+                tracing::warn!(error = %e, "upstream reserve unreachable; failing closed");
+                return Err(fail_closed());
+            }
+        };
+        match resp.json::<ReserveResp>().await {
+            Ok(ReserveResp {
+                reservation_id: Some(id),
+                ..
+            }) => Ok(Reservation {
+                id: id as u64,
+                principal: principal.clone(),
+                reserved: max_tokens,
+            }),
+            Ok(ReserveResp {
+                rejected:
+                    Some(Rejection::InsufficientQuota {
+                        requested,
+                        available,
+                    }),
+                ..
+            }) => Err(BudgetError::InsufficientQuota {
+                requested,
+                available,
+            }),
+            Ok(ReserveResp {
+                rejected:
+                    Some(Rejection::RateLimited {
+                        requested,
+                        available,
+                        retry_after_secs,
+                    }),
+                ..
+            }) => Err(BudgetError::RateLimited {
+                requested,
+                available,
+                retry_after_secs,
+            }),
+            _ => Err(fail_closed()),
+        }
+    }
+
+    async fn settle(&self, reservation: Reservation, actual_tokens: u64) {
+        // Best-effort; a lost settle is reaped by the upstream sweeper (B2).
+        let _ = self
+            .client
+            .post(self.url("/authz/v1/settle"))
+            .bearer_auth(&self.bearer)
+            .json(&serde_json::json!({
+                "reservation_id": reservation.id as i64,
+                "actual_tokens": actual_tokens,
+            }))
+            .send()
+            .await
+            .inspect_err(
+                |e| tracing::warn!(error = %e, "upstream settle failed (sweeper will reap)"),
+            );
+    }
+
+    async fn release(&self, reservation: Reservation) {
+        let _ = self
+            .client
+            .post(self.url("/authz/v1/release"))
+            .bearer_auth(&self.bearer)
+            .json(&serde_json::json!({ "reservation_id": reservation.id as i64 }))
+            .send()
+            .await
+            .inspect_err(
+                |e| tracing::warn!(error = %e, "upstream release failed (sweeper will reap)"),
+            );
+    }
+
+    async fn snapshot(&self, principal: &Principal) -> Option<BudgetSnapshot> {
+        let resp = self
+            .client
+            .post(self.url("/authz/v1/snapshot"))
+            .bearer_auth(&self.bearer)
+            .json(&serde_json::json!({
+                "account_id": principal.account_id,
+                "key_id": principal.key_id,
+            }))
+            .send()
+            .await
+            .ok()?;
+        if !resp.status().is_success() {
+            return None;
+        }
+        let dto = resp.json::<SnapshotDto>().await.ok()?;
+        Some(BudgetSnapshot {
+            hard_cap: dto.hard_cap,
+            spent: dto.spent,
+            reserved: dto.reserved,
+        })
+    }
+}

crates/cortex-gateway/src/error.rs

@@ -0,0 +1,24 @@
+//! Gateway adapter that turns the shared, axum-agnostic
+//! [`cortex_core::error_envelope::OpenAiError`] into an axum [`Response`],
+//! setting the `Retry-After` header when the envelope carries one.
+//!
+//! cortex-core owns the envelope shape and the rejection contract (#60/#63);
+//! this is the only place the gateway crosses from that data into axum.
+
+use axum::http::{HeaderValue, StatusCode, header};
+use axum::response::{IntoResponse, Json, Response};
+use cortex_core::error_envelope::OpenAiError;
+
+/// Render an [`OpenAiError`] as an axum response (status + JSON envelope +
+/// optional `Retry-After`).
+pub fn envelope_response(err: OpenAiError) -> Response {
+    let status = StatusCode::from_u16(err.status).unwrap_or(StatusCode::INTERNAL_SERVER_ERROR);
+    let retry_after = err.retry_after_secs;
+    let mut response = (status, Json(err.body())).into_response();
+    if let Some(secs) = retry_after
+        && let Ok(value) = HeaderValue::from_str(&secs.to_string())
+    {
+        response.headers_mut().insert(header::RETRY_AFTER, value);
+    }
+    response
+}

crates/cortex-gateway/src/handlers.rs

@@ -11,6 +11,8 @@ use axum::http::HeaderMap;
 use axum::response::{IntoResponse, Json, Response};
 use axum::routing::{get, post};
 use chrono::Utc;
+use cortex_core::error_envelope::OpenAiError;
+use cortex_core::harness::ModelLimit;
 use cortex_core::node::{CortexModelEntry, ModelLocation};
 use serde_json::{Value, json};
 use std::sync::Arc;
@@ -33,6 +35,7 @@ async fn chat_completions(
     headers: HeaderMap,
     body: Bytes,
 ) -> Response {
+    log_inbound("openai-chat", "/v1/chat/completions", &body);
     let model_id = match extract_model(&body) {
         Some(m) => m,
         None => {
@@ -40,7 +43,12 @@ async fn chat_completions(
                 handler = "chat_completions",
                 "rejected: missing 'model' field in request body"
             );
-            return error_response(400, "missing 'model' field in request body");
+            return error_response(
+                400,
+                "invalid_request_error",
+                "missing_model_field",
+                "missing 'model' field in request body",
+            );
         }
     };
 
@@ -53,11 +61,7 @@ async fn chat_completions(
                 error = %e,
                 "route resolve failed"
             );
-            // RouteError's Display strings are short and informative
-            // ("model 'X' not found...", "no healthy nodes available")
-            // — fine to surface to the caller. The warn above carries
-            // any extra context for operators.
-            return error_response(e.http_status(), &e.to_string());
+            return route_error_response(&e);
         }
     };
 
@@ -89,6 +93,7 @@ async fn responses(
     headers: HeaderMap,
     body: Bytes,
 ) -> Response {
+    log_inbound("openai-responses", "/v1/responses", &body);
     let model_id = match extract_model(&body) {
         Some(m) => m,
         None => {
@@ -96,7 +101,12 @@ async fn responses(
                 handler = "responses",
                 "rejected: missing 'model' field in request body"
             );
-            return error_response(400, "missing 'model' field in request body");
+            return error_response(
+                400,
+                "invalid_request_error",
+                "missing_model_field",
+                "missing 'model' field in request body",
+            );
         }
     };
 
@@ -109,7 +119,7 @@ async fn responses(
                 error = %e,
                 "route resolve failed"
             );
-            return error_response(e.http_status(), &e.to_string());
+            return route_error_response(&e);
         }
     };
 
@@ -133,6 +143,7 @@ async fn completions(
     headers: HeaderMap,
     body: Bytes,
 ) -> Response {
+    log_inbound("openai-completions", "/v1/completions", &body);
     let model_id = match extract_model(&body) {
         Some(m) => m,
         None => {
@@ -140,7 +151,12 @@ async fn completions(
                 handler = "completions",
                 "rejected: missing 'model' field in request body"
             );
-            return error_response(400, "missing 'model' field in request body");
+            return error_response(
+                400,
+                "invalid_request_error",
+                "missing_model_field",
+                "missing 'model' field in request body",
+            );
         }
     };
 
@@ -153,11 +169,7 @@ async fn completions(
                 error = %e,
                 "route resolve failed"
             );
-            // RouteError's Display strings are short and informative
-            // ("model 'X' not found...", "no healthy nodes available")
-            // — fine to surface to the caller. The warn above carries
-            // any extra context for operators.
-            return error_response(e.http_status(), &e.to_string());
+            return route_error_response(&e);
         }
     };
 
@@ -178,7 +190,7 @@ async fn completions(
 /// `POST /v1/messages` — accept Anthropic format, translate, proxy, translate back.
 async fn anthropic_messages(
     State(fleet): State<Arc<CortexState>>,
-    _headers: HeaderMap,
+    headers: HeaderMap,
     body: Bytes,
 ) -> Response {
     // Parse as Anthropic request.
@@ -190,13 +202,48 @@ async fn anthropic_messages(
                 error = %e,
                 "rejected: invalid Anthropic request body"
             );
-            return error_response(400, "invalid Anthropic request body");
+            return error_response(
+                400,
+                "invalid_request_error",
+                "invalid_anthropic_body",
+                "invalid Anthropic request body",
+            );
         }
     };
 
     let model_id = anth_req.model.clone();
     let is_streaming = anth_req.stream.unwrap_or(false);
 
+    // Wire-debug: make the exercised path and request shape concrete
+    // rather than guesswork. `tool_history` flags whether the client is
+    // continuing a tool conversation (tool_use/tool_result blocks in the
+    // message history) vs. opening a fresh one. Full bodies ride at
+    // trace! (cortex/neuron ship at info; operator infra runs at debug).
+    if tracing::enabled!(tracing::Level::DEBUG) {
+        let n_tools = anth_req
+            .extra
+            .get("tools")
+            .and_then(Value::as_array)
+            .map(|a| a.len())
+            .unwrap_or(0);
+        let tool_history = anth_req
+            .messages
+            .iter()
+            .any(|m| anthropic_message_has_tool_blocks(&m.content));
+        tracing::debug!(
+            wire = "anthropic",
+            endpoint = "/v1/messages",
+            model = %model_id,
+            stream = is_streaming,
+            messages = anth_req.messages.len(),
+            tools = n_tools,
+            tool_history,
+            system = anth_req.system.is_some(),
+            "inbound request"
+        );
+    }
+    tracing::trace!(wire = "anthropic", body = %body_preview(&body), "inbound anthropic body");
+
     // Translate to OpenAI format.
     let openai_req = cortex_core::translate::anthropic_to_openai(anth_req);
     let openai_body = match serde_json::to_vec(&openai_req) {
@@ -208,7 +255,12 @@ async fn anthropic_messages(
                 error = %e,
                 "internal: failed to serialise translated OpenAI request"
             );
-            return error_response(500, "internal translation error");
+            return error_response(
+                500,
+                "api_error",
+                "internal_translation_error",
+                "internal translation error",
+            );
         }
     };
 
@@ -225,7 +277,7 @@ async fn anthropic_messages(
             // ("model 'X' not found...", "no healthy nodes available")
             // — fine to surface to the caller. The warn above carries
             // any extra context for operators.
-            return error_response(e.http_status(), &e.to_string());
+            return route_error_response(&e);
         }
     };
 
@@ -235,6 +287,14 @@ async fn anthropic_messages(
     // neuron's harness sees a model name that matches what it has
     // loaded.
     let openai_body = rewrite_model_in_body(openai_body, &route.resolved_model_id);
+    // The translated body is what neuron actually sees — the reshaped
+    // OpenAI-form tools live here. Tracing it makes "did the tool
+    // definitions survive translation?" a log line, not a guess.
+    tracing::trace!(
+        wire = "anthropic",
+        body = %body_preview(&openai_body),
+        "translated openai body (sent upstream)"
+    );
 
     let labels = [
         ("model", route.resolved_model_id.clone()),
@@ -246,6 +306,33 @@ async fn anthropic_messages(
     }
     let start = Instant::now();
 
+    // Per-request metering + budget enforcement (#51/#52), same lifecycle as
+    // the OpenAI paths. Estimate from the translated OpenAI body (what neuron
+    // sees). Refuse over-cap before dispatch via the #63 envelope; otherwise
+    // build the sink consumed by whichever branch runs below.
+    let usage_sink = match crate::metering::principal_from_headers(&headers) {
+        Some(principal) => {
+            let advertised =
+                advertised_output_limit(&fleet, &route.node_name, &route.resolved_model_id).await;
+            let max_tokens = crate::metering::reservation_estimate(&openai_body, advertised);
+            match crate::metering::reserve_or_reject(
+                Arc::clone(&fleet.entitlements),
+                &principal,
+                max_tokens,
+            )
+            .await
+            {
+                Ok(guard) => Some(crate::metering::usage_sink(
+                    principal,
+                    guard,
+                    std::sync::Arc::clone(&fleet.served_usage),
+                )),
+                Err(env) => return crate::error::envelope_response(env),
+            }
+        }
+        None => None,
+    };
+
     if is_streaming {
         // Anthropic SSE translation (#24): upstream speaks OpenAI SSE;
         // re-frame it event-by-event into Anthropic's message_start /
@@ -256,6 +343,8 @@ async fn anthropic_messages(
             openai_body,
             &model_id,
             &route.node_name,
+            &headers,
+            usage_sink,
         )
         .await;
         metrics::histogram!("cortex_request_duration_seconds", &labels)
@@ -275,11 +364,14 @@ async fn anthropic_messages(
             cold_start = route.cold_start,
             "proxying request"
         );
-        let upstream_resp = fleet
+        let upstream_resp = crate::auth::forward_principal_headers(
+            fleet
                 .http_client
                 .post(&target_url)
                 .body(openai_body)
-            .header("content-type", "application/json")
+                .header("content-type", "application/json"),
+            &headers,
+        )
         .send()
         .await;
 
@@ -295,7 +387,12 @@ async fn anthropic_messages(
                     error = %e,
                     "upstream request failed (network)"
                 );
-                return error_response(502, "upstream request failed");
+                return error_response(
+                    502,
+                    "api_error",
+                    "upstream_connection_error",
+                    "upstream request failed",
+                );
             }
         };
 
@@ -314,7 +411,12 @@ async fn anthropic_messages(
                 body = %body_snippet,
                 "upstream returned non-2xx"
             );
-            return error_response(status, &format!("upstream returned {status}"));
+            return error_response(
+                status,
+                "api_error",
+                "upstream_error",
+                &format!("upstream returned {status}"),
+            );
         }
 
         let body_bytes = match upstream_resp.bytes().await {
@@ -329,7 +431,12 @@ async fn anthropic_messages(
                     error = %e,
                     "failed to read upstream response body"
                 );
-                return error_response(502, "failed to read upstream response");
+                return error_response(
+                    502,
+                    "api_error",
+                    "upstream_connection_error",
+                    "failed to read upstream response",
+                );
             }
         };
 
@@ -351,17 +458,68 @@ async fn anthropic_messages(
                         body = %body_snippet,
                         "failed to parse upstream response as OpenAI ChatCompletionResponse"
                     );
-                    return error_response(502, "malformed upstream response");
+                    return error_response(
+                        502,
+                        "api_error",
+                        "upstream_malformed_response",
+                        "malformed upstream response",
+                    );
                 }
             };
 
         metrics::histogram!("cortex_request_duration_seconds", &labels)
             .record(start.elapsed().as_secs_f64());
+        // Settle metering with the upstream usage (#51). Scanned from the
+        // raw body — same engine-truth source as the streaming path — so we
+        // don't depend on the typed usage struct's optionality.
+        if let Some(sink) = usage_sink {
+            let tail = String::from_utf8_lossy(&body_bytes);
+            let prompt = proxy::last_count_for(&tail, "prompt_tokens").unwrap_or(0);
+            let completion = proxy::last_count_for(&tail, "completion_tokens").unwrap_or(0);
+            sink(prompt, completion);
+        }
+        // Did the model actually produce a structured tool call, or just
+        // text? This is the single most useful signal for "is tool
+        // calling working end-to-end" — a `false` here alongside a
+        // request that carried tools means the model improvised an
+        // unparsed format (the original failure mode).
+        let upstream_tool_calls = openai_resp.choices.iter().any(|c| {
+            c.message
+                .extra
+                .get("tool_calls")
+                .and_then(Value::as_array)
+                .map(|a| !a.is_empty())
+                .unwrap_or(false)
+        });
+        let finish_reason = openai_resp
+            .choices
+            .first()
+            .and_then(|c| c.finish_reason.clone());
+        tracing::debug!(
+            wire = "anthropic",
+            model = %model_id,
+            node = %route.node_name,
+            upstream_tool_calls,
+            finish_reason = ?finish_reason,
+            "upstream non-streaming response"
+        );
         let anthropic_resp = cortex_core::translate::openai_to_anthropic(openai_resp);
         Json(json!(anthropic_resp)).into_response()
     }
 }
 
+/// Combine two self-derived limits for the same model loaded on
+/// different neurons (#67): keep the tightest (smallest `context`) so a
+/// client sized against the advertised limit never overflows the
+/// most-constrained deployment that might serve the request. `None`
+/// means "that neuron reported no limit"; the present one wins.
+fn tightest_limit(a: Option<ModelLimit>, b: Option<ModelLimit>) -> Option<ModelLimit> {
+    match (a, b) {
+        (None, x) | (x, None) => x,
+        (Some(a), Some(b)) => Some(if b.context < a.context { b } else { a }),
+    }
+}
+
 /// `GET /v1/models` — union of (catalogue × topology feasibility) and
 /// (currently loaded somewhere). The result is what the fleet *could*
 /// serve, not just what's already loaded — so OpenAI-compatible tools
@@ -409,9 +567,25 @@ async fn list_models(State(fleet): State<Arc<CortexState>>) -> Json<Value> {
                 loaded: false,
                 feasible_on,
                 locations: Vec::new(),
-                // Catalogue profiles don't declare capabilities yet;
-                // the union is filled in Pass 2 from loaded locations.
-                capabilities: Vec::new(),
+                // Start with catalogue-declared capabilities; Pass 2 unions
+                // runtime-detected ones from loaded neurons.
+                capabilities: profile.capabilities.clone(),
+                // `limit` is no longer operator-declared (#67): the neuron
+                // self-derives it from live VRAM + throughput and reports it
+                // per loaded model — Pass 2 fills it from the neuron's
+                // ModelEntry. A catalogue `limit`, if present, is ignored
+                // (it can't track hot-swapped models or live capacity).
+                // `cost` stays operator-set and flows from the catalogue.
+                limit: None,
+                cost: profile.cost.clone(),
+                // Runtime-detected — will be OR-ed in Pass 2 from neuron data.
+                tool_call: false,
+                reasoning: false,
+                // Flat #78 fields are derived from `limit` in the final
+                // sync pass, once merging is done.
+                max_model_len: None,
+                max_input_tokens: None,
+                max_output_tokens: None,
             },
         );
     }
@@ -444,6 +618,15 @@ async fn list_models(State(fleet): State<Arc<CortexState>>) -> Json<Value> {
                             e.capabilities.push(cap.clone());
                         }
                     }
+                    // OR-in runtime-detected capability flags from the neuron.
+                    e.tool_call = e.tool_call || entry.tool_call;
+                    e.reasoning = e.reasoning || entry.reasoning;
+                    // Adopt the neuron's self-derived limit (#67). When a
+                    // model is loaded on several neurons with different
+                    // headroom, advertise the tightest (smallest context)
+                    // so a client never overflows the most-constrained
+                    // deployment that might serve it.
+                    e.limit = tightest_limit(e.limit.take(), entry.limit.clone());
                 })
                 .or_insert_with(|| CortexModelEntry {
                     id: model_id.clone(),
@@ -456,6 +639,13 @@ async fn list_models(State(fleet): State<Arc<CortexState>>) -> Json<Value> {
                     feasible_on: Vec::new(),
                     locations: vec![location],
                     capabilities: entry.capabilities.clone(),
+                    limit: entry.limit.clone(),
+                    cost: None,
+                    tool_call: entry.tool_call,
+                    reasoning: entry.reasoning,
+                    max_model_len: None,
+                    max_input_tokens: None,
+                    max_output_tokens: None,
                 });
         }
     }
@@ -508,6 +698,13 @@ async fn list_models(State(fleet): State<Arc<CortexState>>) -> Json<Value> {
                     // A model that's only mid-prewarm has no loaded
                     // location to read capabilities from yet.
                     capabilities: Vec::new(),
+                    limit: None,
+                    cost: None,
+                    tool_call: false,
+                    reasoning: false,
+                    max_model_len: None,
+                    max_input_tokens: None,
+                    max_output_tokens: None,
                 });
         }
     }
@@ -538,11 +735,28 @@ async fn list_models(State(fleet): State<Arc<CortexState>>) -> Json<Value> {
                 feasible_on: target_entry.feasible_on,
                 locations: target_entry.locations,
                 capabilities: target_entry.capabilities,
+                limit: target_entry.limit.clone(),
+                cost: target_entry.cost.clone(),
+                tool_call: target_entry.tool_call,
+                reasoning: target_entry.reasoning,
+                max_model_len: None,
+                max_input_tokens: None,
+                max_output_tokens: None,
             },
         );
     }
 
-    let data: Vec<Value> = entries.values().map(|e| json!(e)).collect();
+    // Final pass: derive the flat ecosystem context-window fields (#78)
+    // from each entry's now-settled `limit`, so vLLM-convention clients
+    // (Hermes Agent et al.) can read the window without knowing helexa's
+    // `limit` schema.
+    let data: Vec<Value> = entries
+        .values_mut()
+        .map(|e| {
+            e.sync_flat_limit();
+            json!(e)
+        })
+        .collect();
     Json(json!({
         "object": "list",
         "data": data,
@@ -575,6 +789,19 @@ async fn proxy_with_metrics(
     body: Bytes,
     model_id: &str,
 ) -> Response {
+    // Fail-fast prompt pre-validation (#56): refuse a prompt that already
+    // exceeds the model's advertised context window *before* dispatching to
+    // neuron — the same `400 context_length_exceeded` neuron would emit on
+    // overflow, just earlier and without burning a cold-load/queue slot.
+    // cortex has no tokenizer, so the estimate under-counts and neuron stays
+    // the exact wall; we only catch gross overages (the A0 failure mode).
+    if let Some(context) = advertised_context(fleet, &route.node_name, model_id).await {
+        let est = estimate_prompt_tokens(&body);
+        if est > context {
+            return context_length_exceeded_response(context, est, &headers);
+        }
+    }
+
     let labels = [
         ("model", model_id.to_string()),
         ("node", route.node_name.clone()),
@@ -585,9 +812,46 @@ async fn proxy_with_metrics(
         metrics::counter!("cortex_cold_starts_total", &labels).increment(1);
     }
 
+    // Per-request metering + budget enforcement (#51/#52): reconstruct the
+    // principal from the middleware-stamped headers, reserve the request's
+    // upper-bound cost (prompt estimate + max output), and build the
+    // completion sink that settles actual spend when the response finishes.
+    // A reservation over the hard cap is refused *before* dispatch with the
+    // #63 envelope. Anonymous requests skip all of this. Must happen before
+    // `headers`/`body` are moved into the proxy.
+    let usage_sink = match crate::metering::principal_from_headers(&headers) {
+        Some(principal) => {
+            let advertised = advertised_output_limit(fleet, &route.node_name, model_id).await;
+            let max_tokens = crate::metering::reservation_estimate(&body, advertised);
+            match crate::metering::reserve_or_reject(
+                Arc::clone(&fleet.entitlements),
+                &principal,
+                max_tokens,
+            )
+            .await
+            {
+                Ok(guard) => Some(crate::metering::usage_sink(
+                    principal,
+                    guard,
+                    std::sync::Arc::clone(&fleet.served_usage),
+                )),
+                Err(env) => return crate::error::envelope_response(env),
+            }
+        }
+        None => None,
+    };
+
     let start = Instant::now();
-    let result =
-        proxy::forward_request(&fleet.http_client, route, path, headers, body, model_id).await;
+    let result = proxy::forward_request(
+        &fleet.http_client,
+        route,
+        path,
+        headers,
+        body,
+        model_id,
+        usage_sink,
+    )
+    .await;
     let duration = start.elapsed();
 
     match result {
@@ -606,6 +870,117 @@ async fn proxy_with_metrics(
     }
 }
 
+/// The model's advertised `limit.output` (#62) on a given node, used as the
+/// default output budget for budget reservations (#52) when the request
+/// omits `max_(completion_)tokens`. `None` when the node/model/limit is
+/// unknown — callers fall back to [`crate::metering::FALLBACK_MAX_OUTPUT`].
+async fn advertised_output_limit(
+    fleet: &CortexState,
+    node_name: &str,
+    model_id: &str,
+) -> Option<u64> {
+    let nodes = fleet.nodes.read().await;
+    nodes
+        .get(node_name)?
+        .models
+        .get(model_id)?
+        .limit
+        .as_ref()
+        .map(|l| l.output as u64)
+}
+
+/// The model's advertised hard context window (`limit.context`, #62/#67) on a
+/// node, used for fail-fast prompt pre-validation (#56). `None` when no limit
+/// is known — pre-validation is then skipped and neuron remains the wall.
+async fn advertised_context(fleet: &CortexState, node_name: &str, model_id: &str) -> Option<u64> {
+    let nodes = fleet.nodes.read().await;
+    nodes
+        .get(node_name)?
+        .models
+        .get(model_id)?
+        .limit
+        .as_ref()
+        .map(|l| l.context as u64)
+}
+
+/// Conservative prompt-token estimate (~4 chars/token over message text).
+/// cortex has no tokenizer; under-counting is the safe direction — we only
+/// pre-reject gross overages (#56), and neuron enforces the exact wall.
+fn estimate_prompt_tokens(body: &[u8]) -> u64 {
+    let Ok(v) = serde_json::from_slice::<Value>(body) else {
+        return (body.len() as u64 / 4).max(1);
+    };
+    let mut chars = 0usize;
+    if let Some(messages) = v.get("messages").and_then(Value::as_array) {
+        for m in messages {
+            match m.get("content") {
+                Some(Value::String(s)) => chars += s.len(),
+                Some(Value::Array(parts)) => {
+                    for p in parts {
+                        if let Some(t) = p.get("text").and_then(Value::as_str) {
+                            chars += t.len();
+                        }
+                    }
+                }
+                _ => {}
+            }
+            chars += 8; // rough per-message role/formatting overhead
+        }
+    } else if let Some(prompt) = v.get("prompt").and_then(Value::as_str) {
+        chars += prompt.len(); // legacy /v1/completions
+    } else {
+        return (body.len() as u64 / 4).max(1);
+    }
+    (chars as u64 / 4).max(1)
+}
+
+/// Client-specific, advisory guidance for an over-long prompt (#56),
+/// fingerprinted from `User-Agent`. Strictly advisory: it rides the
+/// `X-Helexa-Advice` header only, never the error envelope, and behaviour
+/// never depends on it. Unknown clients get nothing.
+fn client_advice(headers: &HeaderMap) -> Option<&'static str> {
+    let ua = headers
+        .get(axum::http::header::USER_AGENT)?
+        .to_str()
+        .ok()?
+        .to_ascii_lowercase();
+    if ua.contains("litellm") {
+        Some(
+            "litellm forwards the full context; lower the configured context window or enable client-side compaction",
+        )
+    } else if ua.contains("agent-zero") || ua.contains("agent zero") {
+        Some("reduce the conversation/context size or summarize earlier turns before resending")
+    } else if ua.contains("zed") {
+        Some("reduce the assistant context window in Zed's settings")
+    } else {
+        None
+    }
+}
+
+/// `400 context_length_exceeded` for an over-long prompt caught at the edge
+/// (#56), in the #60 envelope — the same shape neuron emits on overflow, so
+/// clients (opencode auto-compacts) handle it identically. Attaches the
+/// advisory `X-Helexa-Advice` header for fingerprinted clients.
+fn context_length_exceeded_response(
+    context: u64,
+    prompt_est: u64,
+    headers: &HeaderMap,
+) -> Response {
+    let env = OpenAiError::context_length_exceeded(format!(
+        "This model's maximum context length is {context} tokens. Your request is \
+         estimated at ~{prompt_est} tokens. Please reduce the length of the messages."
+    ))
+    .with_extra("max", json!(context))
+    .with_extra("estimated_prompt_tokens", json!(prompt_est));
+    let mut response = crate::error::envelope_response(env);
+    if let Some(advice) = client_advice(headers)
+        && let Ok(value) = axum::http::HeaderValue::from_str(advice)
+    {
+        response.headers_mut().insert("x-helexa-advice", value);
+    }
+    response
+}
+
 /// Update `last_accessed` timestamp for a model on a node (drives LRU eviction).
 async fn touch_model(fleet: &CortexState, node_name: &str, model_id: &str) {
     let mut nodes = fleet.nodes.write().await;
@@ -621,6 +996,57 @@ fn extract_model(body: &[u8]) -> Option<String> {
     v.get("model")?.as_str().map(|s| s.to_string())
 }
 
+/// Emit a uniform wire-debug summary for an OpenAI-family inbound
+/// request (chat/completions, completions, responses). Makes which
+/// surface a client exercised — and whether it sent tools / asked for
+/// streaming — a concrete log line. The full body rides at trace!.
+///
+/// Parsing is gated on the debug level being enabled so info-level
+/// deployments pay nothing.
+fn log_inbound(wire: &str, endpoint: &str, body: &[u8]) {
+    if tracing::enabled!(tracing::Level::DEBUG) {
+        let v: Value = match serde_json::from_slice(body) {
+            Ok(v) => v,
+            Err(_) => return,
+        };
+        let model = v.get("model").and_then(Value::as_str).unwrap_or("?");
+        let stream = v.get("stream").and_then(Value::as_bool).unwrap_or(false);
+        let tools = v
+            .get("tools")
+            .and_then(Value::as_array)
+            .map(|a| a.len())
+            .unwrap_or(0);
+        tracing::debug!(wire, endpoint, model, stream, tools, "inbound request");
+    }
+    tracing::trace!(wire, endpoint, body = %body_preview(body), "inbound body");
+}
+
+/// True if an Anthropic message's content carries any `tool_use` or
+/// `tool_result` block — i.e. the client is mid tool-conversation.
+fn anthropic_message_has_tool_blocks(content: &cortex_core::anthropic::AnthropicContent) -> bool {
+    use cortex_core::anthropic::AnthropicContent;
+    match content {
+        AnthropicContent::Text(_) => false,
+        AnthropicContent::Blocks(blocks) => blocks
+            .iter()
+            .any(|b| matches!(b.block_type.as_str(), "tool_use" | "tool_result")),
+    }
+}
+
+/// Render a UTF-8-safe, length-capped preview of a request/response
+/// body for trace logging. Caps by characters (not bytes) so the slice
+/// can never split a multi-byte codepoint.
+fn body_preview(body: &[u8]) -> String {
+    const MAX_CHARS: usize = 8192;
+    let text = String::from_utf8_lossy(body);
+    if text.chars().count() > MAX_CHARS {
+        let head: String = text.chars().take(MAX_CHARS).collect();
+        format!("{head}…<truncated, {} bytes total>", body.len())
+    } else {
+        text.into_owned()
+    }
+}
+
 /// Rewrite the `model` field of an OpenAI-style JSON request body to
 /// the resolved concrete id. Returns the original bytes if `new_model`
 /// matches what's already there or the body fails to parse — the
@@ -653,14 +1079,16 @@ fn rewrite_model_in_body(body: Bytes, new_model: &str) -> Bytes {
     }
 }
 
-fn error_response(status: u16, message: &str) -> Response {
-    let code = axum::http::StatusCode::from_u16(status)
-        .unwrap_or(axum::http::StatusCode::INTERNAL_SERVER_ERROR);
-    let body = json!({
-        "error": {
-            "message": message,
-            "type": "gateway_error",
+fn error_response(status: u16, typ: &str, code: &str, message: &str) -> Response {
+    crate::error::envelope_response(OpenAiError::new(status, typ, code, message))
 }
-    });
-    (code, Json(body)).into_response()
+
+/// Render a [`RouteError`] in the standard envelope, attaching `Retry-After`
+/// for its transient variants (#63).
+fn route_error_response(e: &router::RouteError) -> Response {
+    let mut env = OpenAiError::new(e.http_status(), e.broad_type(), e.code(), e.to_string());
+    if let Some(secs) = e.retry_after_secs() {
+        env = env.with_retry_after(secs);
+    }
+    crate::error::envelope_response(env)
 }

crates/cortex-gateway/src/lib.rs

@@ -1,23 +1,41 @@
 pub mod anthropic_sse;
+pub mod auth;
+pub mod entitlements_chain;
+pub mod entitlements_local;
+pub mod entitlements_upstream;
+pub mod error;
 pub mod evictor;
 pub mod handlers;
+pub mod metering;
 pub mod metrics;
 pub mod poller;
 pub mod proxy;
 pub mod router;
+pub mod served_usage;
 pub mod state;
 
 use anyhow::Result;
 use axum::Router;
+use axum::middleware::from_fn_with_state;
 use cortex_core::config::GatewayConfig;
 use std::sync::Arc;
 use tower_http::cors::CorsLayer;
 use tower_http::trace::TraceLayer;
 
 /// Build the Axum application router with all routes wired up.
+///
+/// Layer order (outermost first): trace → CORS → auth → handlers. CORS is
+/// outer to auth so preflight `OPTIONS` short-circuits before resolution;
+/// auth (`require_principal`) resolves the bearer key, attaches the
+/// principal, and stamps the internal principal headers before any handler
+/// runs.
 pub fn build_app(fleet: Arc<state::CortexState>) -> Router {
     Router::new()
         .merge(handlers::api_routes())
+        .layer(from_fn_with_state(
+            Arc::clone(&fleet),
+            auth::require_principal,
+        ))
         .layer(CorsLayer::permissive())
         .layer(TraceLayer::new_for_http())
         .with_state(fleet)
@@ -40,6 +58,28 @@ pub async fn run(config: GatewayConfig) -> Result<()> {
         evictor::eviction_loop(evictor_fleet).await;
     });
 
+    // Served-usage reporter (#58): when this operator is part of the mesh,
+    // periodically flush absolute per-principal served-token counters to
+    // upstream for reconciliation.
+    if config.upstream.enabled {
+        let su_fleet = Arc::clone(&fleet);
+        let url = config.upstream.url.clone();
+        let bearer = config.upstream.bearer.clone();
+        let interval =
+            std::time::Duration::from_secs(config.upstream.served_usage_report_interval_secs);
+        tokio::spawn(async move {
+            loop {
+                tokio::time::sleep(interval).await;
+                let rows = su_fleet.served_usage.snapshot();
+                if let Err(e) =
+                    served_usage::report(&su_fleet.http_client, &url, &bearer, &rows).await
+                {
+                    tracing::warn!(error = %e, "served-usage report failed (will retry)");
+                }
+            }
+        });
+    }
+
     let app = build_app(Arc::clone(&fleet));
 
     let listen_addr = config.gateway.listen.parse::<std::net::SocketAddr>()?;

crates/cortex-gateway/src/metering.rs

@@ -0,0 +1,231 @@
+//! Per-request token metering (#51).
+//!
+//! Captures the real `(prompt, completion)` usage of every request and feeds
+//! it to two places: the [`EntitlementProvider`] spend ledger (via
+//! reserve→settle) and per-principal Prometheus counters. The principal is
+//! reconstructed from the internal headers the auth middleware stamped (#49),
+//! so this works uniformly across every proxy path without threading the
+//! typed principal through each handler.
+//!
+//! The reserve→settle lifecycle is established here but, in this phase,
+//! reserves **zero** tokens — metering only, no enforcement. Budget
+//! enforcement (#52) flips the reserved amount to the real
+//! `prompt + max_output` and handles the [`BudgetError`] rejection; the
+//! settle/release plumbing is identical, so that change is localized.
+//!
+//! [`ReservationGuard`] makes leaks impossible: settling records actual
+//! spend and releases the unused remainder; dropping a guard that was never
+//! settled releases the whole reservation. So an early return, error path,
+//! or dropped stream can't strand a reservation.
+
+use axum::http::HeaderMap;
+use cortex_core::entitlements::{
+    BudgetError, EntitlementProvider, HEADER_ACCOUNT_ID, HEADER_KEY_ID, Principal,
+};
+use cortex_core::error_envelope::OpenAiError;
+use std::sync::Arc;
+
+/// Fallback output-token budget when neither the request nor the model's
+/// advertised limit gives one. Bounds the reservation so a capped key is
+/// still gated even on under-specified requests (#52).
+pub const FALLBACK_MAX_OUTPUT: u64 = 4096;
+
+/// Invoked exactly once at request completion with best-effort
+/// `(prompt_tokens, completion_tokens)`. When no usage could be observed
+/// (e.g. a pre-dispatch failure or a dropped stream) it is dropped unused —
+/// which releases the held reservation via [`ReservationGuard`]'s `Drop`.
+pub type UsageSink = Box<dyn FnOnce(u64, u64) + Send>;
+
+/// Reconstruct the principal from the cortex-stamped internal headers. The
+/// auth middleware strips any client copy and stamps the authoritative value,
+/// so these headers are trustworthy within cortex. `None` for anonymous
+/// (unauthenticated) requests.
+pub fn principal_from_headers(headers: &HeaderMap) -> Option<Principal> {
+    let account_id = headers.get(HEADER_ACCOUNT_ID)?.to_str().ok()?.to_string();
+    let key_id = headers.get(HEADER_KEY_ID)?.to_str().ok()?.to_string();
+    Some(Principal { account_id, key_id })
+}
+
+/// Emit per-principal spend counters (#51). Labelled by account/key only —
+/// both are operator-bounded, so cardinality is controlled.
+pub fn record_spend(principal: &Principal, prompt: u64, completion: u64) {
+    let labels = [
+        ("account", principal.account_id.clone()),
+        ("key", principal.key_id.clone()),
+    ];
+    metrics::counter!("cortex_spend_tokens_total", &labels).increment(prompt + completion);
+    metrics::counter!("cortex_spend_prompt_tokens_total", &labels).increment(prompt);
+    metrics::counter!("cortex_spend_completion_tokens_total", &labels).increment(completion);
+}
+
+/// Holds a budget reservation for the life of a request. [`settle`] records
+/// actual spend and releases the remainder; an un-settled guard releases the
+/// whole reservation when dropped. Anonymous requests carry an empty guard,
+/// where every operation is a no-op.
+///
+/// [`settle`]: ReservationGuard::settle
+pub struct ReservationGuard {
+    provider: Arc<dyn EntitlementProvider>,
+    reservation: Option<cortex_core::entitlements::Reservation>,
+}
+
+impl ReservationGuard {
+    /// An empty guard for an anonymous request — no reservation to resolve.
+    pub fn anonymous(provider: Arc<dyn EntitlementProvider>) -> Self {
+        Self {
+            provider,
+            reservation: None,
+        }
+    }
+
+    /// Wrap an already-acquired reservation.
+    fn held(
+        provider: Arc<dyn EntitlementProvider>,
+        reservation: cortex_core::entitlements::Reservation,
+    ) -> Self {
+        Self {
+            provider,
+            reservation: Some(reservation),
+        }
+    }
+
+    /// Settle with the tokens actually consumed, disarming the drop-release.
+    /// Spawns the (fast, in-process for the local provider) settle so the
+    /// caller — which may be a sync stream-completion callback — needn't
+    /// await.
+    pub fn settle(mut self, actual_tokens: u64) {
+        if let Some(reservation) = self.reservation.take() {
+            let provider = Arc::clone(&self.provider);
+            tokio::spawn(async move {
+                provider.settle(reservation, actual_tokens).await;
+            });
+        }
+    }
+}
+
+impl Drop for ReservationGuard {
+    fn drop(&mut self) {
+        if let Some(reservation) = self.reservation.take() {
+            let provider = Arc::clone(&self.provider);
+            tokio::spawn(async move {
+                provider.release(reservation).await;
+            });
+        }
+    }
+}
+
+/// Build the completion sink for an authenticated request: record spend and
+/// settle the reservation with the observed total. Dropping it unused (no
+/// usage observed) releases the reservation via the guard.
+pub fn usage_sink(
+    principal: Principal,
+    guard: ReservationGuard,
+    served_usage: std::sync::Arc<crate::served_usage::ServedUsage>,
+) -> UsageSink {
+    Box::new(move |prompt, completion| {
+        record_spend(&principal, prompt, completion);
+        // Per-principal served-usage tally for #58 reconciliation. Recorded
+        // for every metered (authenticated) request; the flush task reports
+        // it to upstream when the operator is part of the mesh.
+        served_usage.add(
+            &principal.account_id,
+            &principal.key_id,
+            prompt + completion,
+        );
+        guard.settle(prompt + completion);
+    })
+}
+
+/// Reserve the request's upper-bound token cost for the principal, refusing
+/// *before* dispatch if it would exceed the hard cap (#52). On success
+/// returns a guard the caller settles with actual usage; on refusal returns
+/// the #63 envelope (`rate_limit_exceeded` + `Retry-After` for a resetting
+/// window, `insufficient_quota` for a hard balance — never `402`).
+pub async fn reserve_or_reject(
+    provider: Arc<dyn EntitlementProvider>,
+    principal: &Principal,
+    max_tokens: u64,
+) -> Result<ReservationGuard, OpenAiError> {
+    match provider.reserve(principal, max_tokens).await {
+        Ok(reservation) => Ok(ReservationGuard::held(provider, reservation)),
+        Err(err) => Err(budget_error_to_envelope(err)),
+    }
+}
+
+/// Map a [`BudgetError`] to the #63 envelope. The provider chose the window
+/// semantics; this only translates them to HTTP.
+fn budget_error_to_envelope(err: BudgetError) -> OpenAiError {
+    match err {
+        BudgetError::RateLimited {
+            retry_after_secs, ..
+        } => OpenAiError::rate_limit_exceeded(err.to_string(), retry_after_secs),
+        BudgetError::InsufficientQuota { .. } => OpenAiError::insufficient_quota(err.to_string()),
+    }
+}
+
+/// Upper-bound tokens to reserve for a request (#52): an over-estimate of
+/// the prompt plus the maximum output. `advertised_output` is the model's
+/// `limit.output` (#62), used when the request omits `max_(completion_)tokens`.
+/// Over-reserving is safe — settle corrects spend to the actual usage.
+pub fn reservation_estimate(body: &[u8], advertised_output: Option<u64>) -> u64 {
+    let max_output = requested_max_output(body)
+        .or(advertised_output)
+        .unwrap_or(FALLBACK_MAX_OUTPUT);
+    estimate_prompt_tokens(body).saturating_add(max_output)
+}
+
+/// The client's requested output cap, from `max_completion_tokens` (or the
+/// legacy `max_tokens`). `None` when unspecified.
+fn requested_max_output(body: &[u8]) -> Option<u64> {
+    let v: serde_json::Value = serde_json::from_slice(body).ok()?;
+    v.get("max_completion_tokens")
+        .or_else(|| v.get("max_tokens"))
+        .and_then(serde_json::Value::as_u64)
+}
+
+/// Rough prompt-token estimate at ~4 chars/token over the whole body. cortex
+/// has no tokenizer; JSON overhead makes this a conservative over-estimate,
+/// and neuron remains the exact context wall (#56/#60). Settle reconciles to
+/// the real usage afterward.
+fn estimate_prompt_tokens(body: &[u8]) -> u64 {
+    (body.len() as u64 / 4).max(1)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn requested_max_output_prefers_max_completion_tokens() {
+        let body = br#"{"model":"m","max_completion_tokens":256,"max_tokens":99}"#;
+        assert_eq!(requested_max_output(body), Some(256));
+    }
+
+    #[test]
+    fn requested_max_output_falls_back_to_legacy_max_tokens() {
+        let body = br#"{"model":"m","max_tokens":128}"#;
+        assert_eq!(requested_max_output(body), Some(128));
+    }
+
+    #[test]
+    fn estimate_uses_requested_output_when_present() {
+        // Requested output dominates; prompt estimate is small for a tiny body.
+        let body = br#"{"model":"m","max_tokens":1000}"#;
+        let est = reservation_estimate(body, Some(8192));
+        assert!(est >= 1000 && est < 1100, "est was {est}");
+    }
+
+    #[test]
+    fn estimate_uses_advertised_output_when_request_omits_it() {
+        let body = br#"{"model":"m","messages":[]}"#;
+        let est = reservation_estimate(body, Some(8192));
+        assert!(est >= 8192, "est was {est}");
+    }
+
+    #[test]
+    fn estimate_falls_back_when_nothing_advertised() {
+        let body = br#"{"model":"m"}"#;
+        let est = reservation_estimate(body, None);
+        assert!(est >= FALLBACK_MAX_OUTPUT, "est was {est}");
+    }
+}

crates/cortex-gateway/src/metrics.rs

@@ -63,4 +63,16 @@ fn describe_metrics() {
         "cortex_cold_starts_total",
         "Total number of cold-start model loads"
     );
+    metrics::describe_counter!(
+        "cortex_spend_tokens_total",
+        "Total metered tokens (prompt + completion) per principal, labelled by account/key (#51)"
+    );
+    metrics::describe_counter!(
+        "cortex_spend_prompt_tokens_total",
+        "Metered prompt tokens per principal, labelled by account/key (#51)"
+    );
+    metrics::describe_counter!(
+        "cortex_spend_completion_tokens_total",
+        "Metered completion tokens per principal, labelled by account/key (#51)"
+    );
 }

crates/cortex-gateway/src/poller.rs

@@ -5,12 +5,29 @@ use crate::state::CortexState;
 use chrono::Utc;
 use cortex_core::discovery::{DiscoveryResponse, HealthResponse};
 use cortex_core::harness::ModelInfo;
-use cortex_core::node::{ModelEntry, ModelStatus};
+use cortex_core::node::{ModelEntry, ModelStatus, NodeState};
 use std::sync::Arc;
 use std::time::Duration;
 
 const POLL_INTERVAL: Duration = Duration::from_secs(10);
 
+/// Consecutive failed `/models` polls before a node is marked unhealthy.
+/// Debounces transient misses (a busy neuron briefly slow to answer) so a
+/// single blip can't yank a node — and its models — out of routing. At the
+/// 10s poll interval this tolerates ~20s of flapping before evicting.
+const POLL_FAILURE_THRESHOLD: u32 = 3;
+
+/// Record a failed poll for `node`, marking it unhealthy only once failures
+/// reach [`POLL_FAILURE_THRESHOLD`]. Below the threshold the node keeps its
+/// last-known health, riding over transient misses. A successful poll resets
+/// the counter (see the success arm in `poll_once`).
+fn record_poll_failure(node: &mut NodeState) {
+    node.consecutive_poll_failures = node.consecutive_poll_failures.saturating_add(1);
+    if node.consecutive_poll_failures >= POLL_FAILURE_THRESHOLD {
+        node.healthy = false;
+    }
+}
+
 /// Runs forever, polling all neurons on a fixed interval.
 pub async fn poll_loop(fleet: Arc<CortexState>) {
     loop {
@@ -26,14 +43,23 @@ pub async fn poll_once(fleet: &CortexState) {
     }
 }
 
-/// One-shot fetch of `GET /discovery`. Cached on the NodeState forever
-/// after the first success — topology is invariant for a given neuron
-/// process. Skipped when the cache is already populated.
+/// Fetch `GET /discovery` and cache it on the NodeState — topology is
+/// invariant for a given neuron process, so a successful fetch is kept.
+/// Re-polled only while `max_prompt_tokens` is still unknown (0): on a
+/// rolling deploy cortex can win the race and cache a neuron's discovery
+/// before that neuron reports the field (it deserialises to 0). Re-polling
+/// until a real cap arrives self-heals that without periodic polling.
 async fn maybe_poll_discovery(fleet: &CortexState, name: &str, endpoint: &str) {
     {
         let nodes = fleet.nodes.read().await;
         match nodes.get(name) {
-            Some(n) if n.discovery.is_some() => return,
+            Some(n)
+                if n.discovery
+                    .as_ref()
+                    .is_some_and(|d| d.max_prompt_tokens > 0) =>
+            {
+                return;
+            }
             _ => {}
         }
     }
@@ -108,6 +134,11 @@ async fn poll_neuron(fleet: &CortexState, name: &str, endpoint: &str) {
                                 e.status = status;
                                 e.vram_estimate_mb = upstream.vram_used_mb;
                                 e.capabilities = upstream.capabilities.clone();
+                                e.tool_call = upstream.tool_call;
+                                e.reasoning = upstream.reasoning;
+                                // Neuron's self-derived limit (#67) — the
+                                // authoritative source the gateway advertises.
+                                e.limit = upstream.limit.clone();
                             })
                             .or_insert_with(|| ModelEntry {
                                 id: upstream.id.clone(),
@@ -115,19 +146,23 @@ async fn poll_neuron(fleet: &CortexState, name: &str, endpoint: &str) {
                                 last_accessed: None,
                                 vram_estimate_mb: upstream.vram_used_mb,
                                 capabilities: upstream.capabilities.clone(),
+                                tool_call: upstream.tool_call,
+                                reasoning: upstream.reasoning,
+                                limit: upstream.limit.clone(),
                             });
                     }
 
                     // Remove models no longer reported by the neuron.
                     node.models.retain(|id, _| seen.contains(id));
 
+                    node.consecutive_poll_failures = 0;
                     node.healthy = true;
                     node.last_poll = Some(Utc::now());
                     tracing::debug!(node = name, models = models.len(), "poll ok");
                 }
                 Err(e) => {
                     tracing::warn!(node = name, error = %e, "failed to parse /models response");
-                    node.healthy = false;
+                    record_poll_failure(node);
                 }
             }
         }
@@ -137,11 +172,11 @@ async fn poll_neuron(fleet: &CortexState, name: &str, endpoint: &str) {
                 status = %resp.status(),
                 "neuron returned non-success status"
             );
-            node.healthy = false;
+            record_poll_failure(node);
         }
         Err(e) => {
             tracing::warn!(node = name, error = %e, "failed to reach neuron");
-            node.healthy = false;
+            record_poll_failure(node);
         }
     }
 
@@ -183,6 +218,9 @@ async fn poll_health(fleet: &CortexState, name: &str, endpoint: &str) {
             let mut nodes = fleet.nodes.write().await;
             if let Some(node) = nodes.get_mut(name) {
                 node.activation = Some(h.activation);
+                // Per-model admission load (#53) → keyed by id for the
+                // load-aware router (#55).
+                node.model_load = h.models.into_iter().map(|m| (m.id.clone(), m)).collect();
             }
         }
         Err(e) => {

crates/cortex-gateway/src/proxy.rs

@@ -1,21 +1,27 @@
 //! Streaming HTTP reverse proxy to neuron backends.
 //!
-//! For streaming requests, SSE chunks are forwarded as they arrive.
-//! The proxy captures timing information for metrics but does not
-//! buffer the full response.
+//! The streaming *mechanism* — forward an SSE body chunk-for-chunk without
+//! buffering, observing the bytes for metrics — lives in the shared
+//! [`helexa_stream`] crate (#71), so cortex and helexa-router use one
+//! implementation. This module supplies cortex's *policy*: the
+//! [`CortexMetrics`] observer (per-request token metrics + per-principal
+//! reservation settle), cortex's logging contract, and the cortex error
+//! envelope. The usage-extraction helper is re-exported from the shared
+//! crate so existing call sites keep working.
 
 use crate::router::RouteDecision;
-use anyhow::Result;
-use axum::body::Body;
-use axum::http::{HeaderMap, StatusCode};
+use axum::http::HeaderMap;
+use axum::http::StatusCode;
 use axum::response::{IntoResponse, Response};
-use futures::Stream;
-use futures::stream::BoxStream;
+use helexa_stream::{BodyTail, ChunkObserver, StreamError};
 use reqwest::Client;
-use std::pin::Pin;
-use std::task::{Context, Poll};
 use std::time::Instant;
 
+/// Re-export the shared usage-extraction helper. Several cortex modules
+/// (`handlers`, `anthropic_sse`) pull token counts out of a buffered body
+/// tail via this function; it lives in `helexa-stream` now.
+pub use helexa_stream::last_count_for;
+
 /// Proxy a request body to the resolved backend node and stream the response.
 ///
 /// Logging contract: every call emits exactly one structured event at
@@ -31,6 +37,7 @@ pub async fn forward_request(
     headers: HeaderMap,
     body: bytes::Bytes,
     model_id: &str,
+    usage_sink: Option<crate::metering::UsageSink>,
 ) -> Result<Response, ProxyError> {
     let request_start = Instant::now();
     let url = format!("{}{}", route.endpoint, path);
@@ -41,66 +48,41 @@ pub async fn forward_request(
         "proxying request"
     );
 
-    let mut req_builder = client.post(&url).body(body);
+    let observer = CortexMetrics::new(model_id, &route.node_name, request_start, usage_sink);
 
-    // Forward relevant headers.
-    for (key, value) in headers.iter() {
-        if key == "host" || key == "content-length" {
-            continue; // reqwest sets these
-        }
-        req_builder = req_builder.header(key, value);
-    }
-
-    let upstream_resp = match req_builder.send().await {
-        Ok(r) => r,
-        Err(e) => {
-            tracing::warn!(
+    let response = helexa_stream::forward_streaming(client, &url, headers, body, observer)
+        .await
+        .map_err(|e| {
+            match &e {
+                StreamError::Upstream(err) => tracing::warn!(
                     node = %route.node_name,
                     url = %url,
-                error = %e,
+                    error = %err,
                     "proxy: upstream request failed (network)"
-            );
-            return Err(ProxyError::Upstream(e));
+                ),
+                StreamError::ResponseBuild(err) => tracing::warn!(
+                    node = %route.node_name,
+                    url = %url,
+                    error = %err,
+                    "proxy: failed to build response"
+                ),
             }
-    };
+            ProxyError::from(e)
+        })?;
 
-    let upstream_status = upstream_resp.status();
-    if !upstream_status.is_success() {
+    if !response.status().is_success() {
         // Streaming body — can't snippet without breaking the stream
         // pass-through. Log status + URL; the client still gets the
         // upstream status, just without the leaked body.
         tracing::warn!(
             node = %route.node_name,
             url = %url,
-            status = upstream_status.as_u16(),
+            status = response.status().as_u16(),
             "proxy: upstream returned non-2xx"
         );
     }
 
-    let status = StatusCode::from_u16(upstream_status.as_u16()).unwrap_or(StatusCode::BAD_GATEWAY);
-
-    let resp_headers = upstream_resp.headers().clone();
-    let stream = TokenMetricsStream::new(
-        Box::pin(upstream_resp.bytes_stream()),
-        TokenMetrics::new(model_id, &route.node_name, request_start),
-    );
-
-    let body = Body::from_stream(stream);
-
-    let mut response = Response::builder().status(status);
-    for (key, value) in resp_headers.iter() {
-        response = response.header(key, value);
-    }
-
-    response.body(body).map_err(|e| {
-        tracing::warn!(
-            node = %route.node_name,
-            url = %url,
-            error = %e,
-            "proxy: failed to build response"
-        );
-        ProxyError::ResponseBuild(e.to_string())
-    })
+    Ok(response)
 }
 
 #[derive(Debug, thiserror::Error)]
@@ -111,22 +93,35 @@ pub enum ProxyError {
     ResponseBuild(String),
 }
 
+impl From<StreamError> for ProxyError {
+    fn from(e: StreamError) -> Self {
+        match e {
+            StreamError::Upstream(err) => ProxyError::Upstream(err),
+            StreamError::ResponseBuild(msg) => ProxyError::ResponseBuild(msg),
+        }
+    }
+}
+
 impl IntoResponse for ProxyError {
     fn into_response(self) -> Response {
-        let (status, message) = match &self {
-            ProxyError::Upstream(_) => (StatusCode::BAD_GATEWAY, "upstream request failed"),
+        let (status, code, message) = match &self {
+            ProxyError::Upstream(_) => (
+                StatusCode::BAD_GATEWAY,
+                "upstream_connection_error",
+                "upstream request failed",
+            ),
             ProxyError::ResponseBuild(_) => (
                 StatusCode::INTERNAL_SERVER_ERROR,
+                "internal_server_error",
                 "failed to build response",
             ),
         };
-        let body = serde_json::json!({
-            "error": {
-                "message": message,
-                "type": "proxy_error",
-            }
-        });
-        (status, axum::Json(body)).into_response()
+        crate::error::envelope_response(cortex_core::error_envelope::OpenAiError::new(
+            status.as_u16(),
+            "api_error",
+            code,
+            message,
+        ))
     }
 }
 
@@ -134,9 +129,10 @@ impl IntoResponse for ProxyError {
 //
 // The proxy never buffers or re-serialises the upstream body — chunks
 // are forwarded verbatim. For metrics it observes each chunk's arrival
-// time and keeps a bounded tail of the body text, from which the final
-// OpenAI `usage` object (present on the last SSE chunk and on
-// non-streaming JSON bodies alike) yields engine-truth token counts.
+// time and keeps a bounded tail of the body text (via the shared
+// `helexa_stream::BodyTail`), from which the final OpenAI `usage` object
+// (present on the last SSE chunk and on non-streaming JSON bodies alike)
+// yields engine-truth token counts.
 //
 // Emitted per request, labelled {model, node}:
 //   cortex_time_to_first_token_seconds  (histogram) — first body chunk
@@ -150,42 +146,29 @@ impl IntoResponse for ProxyError {
 /// non-streaming bodies.
 const TAIL_CAP_BYTES: usize = 64 * 1024;
 
-/// Find the value of the LAST `"key": <integer>` occurrence in `tail`.
-/// Pure and chunk-boundary-safe (the tail is contiguous appended text).
-/// The quoted-needle form means `completion_tokens` never matches
-/// `completion_tokens_details`.
-pub(crate) fn last_count_for(tail: &str, key: &str) -> Option<u64> {
-    let needle = format!("\"{key}\"");
-    let mut result = None;
-    for (idx, _) in tail.match_indices(&needle) {
-        let rest = tail[idx + needle.len()..].trim_start();
-        let Some(rest) = rest.strip_prefix(':') else {
-            continue;
-        };
-        let rest = rest.trim_start();
-        let digits: &str = &rest[..rest
-            .char_indices()
-            .find(|(_, c)| !c.is_ascii_digit())
-            .map(|(i, _)| i)
-            .unwrap_or(rest.len())];
-        if let Ok(v) = digits.parse::<u64>() {
-            result = Some(v);
-        }
-    }
-    result
-}
-
-struct TokenMetrics {
+/// cortex's [`ChunkObserver`]: per-request token metrics plus the
+/// per-principal reservation settle. Drives cortex policy over the shared
+/// streaming mechanism.
+struct CortexMetrics {
     labels: [(&'static str, String); 2],
     request_start: Instant,
     first_chunk: Option<Instant>,
     last_chunk: Option<Instant>,
-    tail: String,
+    tail: BodyTail,
     finished: bool,
+    /// Per-principal metering hook (#51). Invoked exactly once in `finish`
+    /// with the observed `(prompt, completion)` so the reservation can be
+    /// settled and spend recorded. `None` for anonymous requests.
+    usage_sink: Option<crate::metering::UsageSink>,
 }
 
-impl TokenMetrics {
-    fn new(model_id: &str, node_name: &str, request_start: Instant) -> Self {
+impl CortexMetrics {
+    fn new(
+        model_id: &str,
+        node_name: &str,
+        request_start: Instant,
+        usage_sink: Option<crate::metering::UsageSink>,
+    ) -> Self {
         Self {
             labels: [
                 ("model", model_id.to_string()),
@@ -194,25 +177,19 @@ impl TokenMetrics {
             request_start,
             first_chunk: None,
             last_chunk: None,
-            tail: String::new(),
+            tail: BodyTail::new(TAIL_CAP_BYTES),
             finished: false,
+            usage_sink,
+        }
     }
 }
 
+impl ChunkObserver for CortexMetrics {
     fn observe(&mut self, chunk: &[u8]) {
         let now = Instant::now();
         self.first_chunk.get_or_insert(now);
         self.last_chunk = Some(now);
-        self.tail.push_str(&String::from_utf8_lossy(chunk));
-        if self.tail.len() > TAIL_CAP_BYTES {
-            // Keep the newest half; the usage object is always at the
-            // very end of the body. Split at a char boundary.
-            let mut cut = self.tail.len() - TAIL_CAP_BYTES / 2;
-            while !self.tail.is_char_boundary(cut) {
-                cut += 1;
-            }
-            self.tail.drain(..cut);
-        }
+        self.tail.push(chunk);
     }
 
     /// Emit the metrics exactly once — called on clean stream end and
@@ -223,28 +200,28 @@ impl TokenMetrics {
             return;
         }
         self.finished = true;
-        let Some(first) = self.first_chunk else {
-            return; // no body ever arrived — nothing to record
-        };
+
+        let prompt = last_count_for(self.tail.as_str(), "prompt_tokens");
+        let completion = last_count_for(self.tail.as_str(), "completion_tokens");
+
+        // Per-model metrics — only when body chunks actually arrived.
+        if let Some(first) = self.first_chunk {
             let ttft = first.duration_since(self.request_start).as_secs_f64();
             metrics::histogram!("cortex_time_to_first_token_seconds", &self.labels).record(ttft);
 
-        if let Some(prompt) = last_count_for(&self.tail, "prompt_tokens") {
+            if let Some(prompt) = prompt {
                 metrics::counter!("cortex_prompt_tokens_total", &self.labels).increment(prompt);
             }
-        let Some(completion) = last_count_for(&self.tail, "completion_tokens") else {
-            return;
-        };
-        if completion == 0 {
-            return;
-        }
-        metrics::counter!("cortex_completion_tokens_total", &self.labels).increment(completion);
+            if let Some(completion) = completion.filter(|c| *c > 0) {
+                metrics::counter!("cortex_completion_tokens_total", &self.labels)
+                    .increment(completion);
 
                 let last = self.last_chunk.unwrap_or(first);
                 let decode_window = last.duration_since(first).as_secs_f64();
                 // Streaming: rate over the decode window (first→last chunk).
-        // Non-streaming bodies arrive as ~one chunk (window ≈ 0), where
-        // the only honest denominator is the full request duration.
+                // Non-streaming bodies arrive as ~one chunk (window ≈ 0),
+                // where the only honest denominator is the full request
+                // duration.
                 let secs = if decode_window >= 0.1 {
                     decode_window
                 } else {
@@ -257,96 +234,11 @@ impl TokenMetrics {
             }
         }
 
-/// Pass-through stream wrapper that feeds [`TokenMetrics`]. Emits on
-/// clean end-of-stream; the Drop impl covers client disconnects.
-struct TokenMetricsStream {
-    inner: BoxStream<'static, Result<bytes::Bytes, reqwest::Error>>,
-    metrics: TokenMetrics,
-}
-
-impl TokenMetricsStream {
-    fn new(
-        inner: BoxStream<'static, Result<bytes::Bytes, reqwest::Error>>,
-        metrics: TokenMetrics,
-    ) -> Self {
-        Self { inner, metrics }
-    }
-}
-
-impl Stream for TokenMetricsStream {
-    type Item = Result<bytes::Bytes, reqwest::Error>;
-
-    fn poll_next(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
-        let this = self.get_mut();
-        match this.inner.as_mut().poll_next(cx) {
-            Poll::Ready(Some(Ok(chunk))) => {
-                this.metrics.observe(&chunk);
-                Poll::Ready(Some(Ok(chunk)))
-            }
-            Poll::Ready(Some(Err(e))) => Poll::Ready(Some(Err(e))),
-            Poll::Ready(None) => {
-                this.metrics.finish();
-                Poll::Ready(None)
-            }
-            Poll::Pending => Poll::Pending,
+        // Per-principal metering + reservation settle (#51). Always runs so
+        // the reservation is resolved even when no usage/body was observed
+        // (sink with (0, 0) → settle 0 → release).
+        if let Some(sink) = self.usage_sink.take() {
+            sink(prompt.unwrap_or(0), completion.unwrap_or(0));
         }
     }
 }
-
-impl Drop for TokenMetricsStream {
-    fn drop(&mut self) {
-        self.metrics.finish();
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::last_count_for;
-
-    #[test]
-    fn extracts_counts_from_final_sse_usage_chunk() {
-        let tail = concat!(
-            "data: {\"choices\":[{\"delta\":{\"content\":\"hi\"}}]}\n\n",
-            "data: {\"choices\":[],\"usage\":{\"prompt_tokens\":225,",
-            "\"completion_tokens\":42,\"total_tokens\":267}}\n\n",
-            "data: [DONE]\n\n"
-        );
-        assert_eq!(last_count_for(tail, "prompt_tokens"), Some(225));
-        assert_eq!(last_count_for(tail, "completion_tokens"), Some(42));
-    }
-
-    #[test]
-    fn extracts_counts_from_non_streaming_body() {
-        let tail = "{\"choices\":[{\"message\":{\"content\":\"hi\"}}],\
-                    \"usage\":{\"prompt_tokens\": 12, \"completion_tokens\": 7}}";
-        assert_eq!(last_count_for(tail, "prompt_tokens"), Some(12));
-        assert_eq!(last_count_for(tail, "completion_tokens"), Some(7));
-    }
-
-    #[test]
-    fn ignores_details_variants_and_takes_last_occurrence() {
-        // completion_tokens_details must not shadow completion_tokens,
-        // and the LAST usage object wins (matters when content echoes
-        // a usage-shaped string earlier in the stream).
-        let tail = concat!(
-            "data: {\"usage\":{\"completion_tokens\":1}}\n\n",
-            "data: {\"usage\":{\"completion_tokens\":99,",
-            "\"completion_tokens_details\":{\"reasoning_tokens\":3}}}\n\n"
-        );
-        assert_eq!(last_count_for(tail, "completion_tokens"), Some(99));
-    }
-
-    #[test]
-    fn absent_keys_yield_none() {
-        assert_eq!(
-            last_count_for("data: [DONE]\n\n", "completion_tokens"),
-            None
-        );
-        assert_eq!(last_count_for("", "prompt_tokens"), None);
-        // key present but non-numeric value
-        assert_eq!(
-            last_count_for("\"completion_tokens\": null", "completion_tokens"),
-            None
-        );
-    }
-}

crates/cortex-gateway/src/router.rs

@@ -50,6 +50,10 @@ pub enum RouteError {
         "model '{model_id}' is in the catalogue but no healthy neuron's topology satisfies its constraints"
     )]
     NoFeasibleNeuron { model_id: String },
+    #[error(
+        "model '{model_id}' is feasible on a neuron that is currently unhealthy — retry shortly"
+    )]
+    FeasibleNodeUnhealthy { model_id: String },
     #[error("cold-load of '{model_id}' on '{node}' failed: {message}")]
     ColdLoadFailed {
         model_id: String,
@@ -63,15 +67,57 @@ pub enum RouteError {
 }
 
 impl RouteError {
-    /// HTTP status the gateway should answer with. `ModelRecovering`
-    /// is the one transient case (503, retry the same request);
-    /// everything else keeps the long-standing 404 behaviour.
+    /// HTTP status the gateway should answer with. `NoHealthyNodes` and
+    /// `ModelRecovering` are the transient cases (503 service_unavailable,
+    /// safe to retry the same request); everything else is 404.
     pub fn http_status(&self) -> u16 {
         match self {
-            RouteError::ModelRecovering { .. } => 503,
+            RouteError::NoHealthyNodes
+            | RouteError::ModelRecovering { .. }
+            | RouteError::FeasibleNodeUnhealthy { .. } => 503,
             _ => 404,
         }
     }
+
+    /// Broad OpenAI error category for the JSON envelope.
+    pub fn broad_type(&self) -> &'static str {
+        match self {
+            RouteError::ModelNotFound(_) => "invalid_request_error",
+            RouteError::NoHealthyNodes
+            | RouteError::EndpointResolveFailed(_, _)
+            | RouteError::NoFeasibleNeuron { .. }
+            | RouteError::ColdLoadFailed { .. }
+            | RouteError::ModelRecovering { .. }
+            | RouteError::FeasibleNodeUnhealthy { .. } => "api_error",
+        }
+    }
+
+    /// Specific machine-readable error code.
+    pub fn code(&self) -> &'static str {
+        match self {
+            RouteError::ModelNotFound(_) => "model_not_found",
+            RouteError::NoHealthyNodes => "service_unavailable",
+            RouteError::EndpointResolveFailed(_, _) => "service_unavailable",
+            RouteError::NoFeasibleNeuron { .. } => "service_unavailable",
+            RouteError::ColdLoadFailed { .. } => "service_unavailable",
+            RouteError::ModelRecovering { .. } => "service_unavailable",
+            RouteError::FeasibleNodeUnhealthy { .. } => "service_unavailable",
+        }
+    }
+
+    /// Seconds to advertise in `Retry-After` for the transient variants
+    /// (#63). `NoHealthyNodes` may clear once the poller re-marks a node
+    /// healthy; `ModelRecovering` clears once the device context finishes
+    /// rebuilding — both are safe to retry. Everything else is permanent
+    /// for this request (404) and carries no hint.
+    pub fn retry_after_secs(&self) -> Option<u64> {
+        match self {
+            RouteError::ModelRecovering { .. } => Some(2),
+            RouteError::FeasibleNodeUnhealthy { .. } => Some(3),
+            RouteError::NoHealthyNodes => Some(5),
+            _ => None,
+        }
+    }
 }
 
 /// Resolve which node should serve a request for the given model.
@@ -95,7 +141,9 @@ pub async fn resolve(
     // Snapshot loaded / unloaded / recovering state from the poller cache.
     let (loaded_route, unloaded_route, recovering_node, any_healthy) = {
         let nodes = fleet.nodes.read().await;
-        let mut loaded_route = None;
+        // All healthy nodes with the model loaded, each with its current
+        // admission load (#53) so we can pick the least-busy replica (#55).
+        let mut loaded_candidates: Vec<(String, String, usize)> = Vec::new();
         let mut unloaded_route = None;
         let mut recovering_node = None;
         let mut any_healthy = false;
@@ -107,8 +155,15 @@ pub async fn resolve(
             if let Some(entry) = node.models.get(model_id) {
                 match entry.status {
                     ModelStatus::Loaded | ModelStatus::Reloading => {
-                        loaded_route = Some((node.name.clone(), node.endpoint.clone(), false));
-                        break;
+                        // Least-busy score: in-flight + queued from the
+                        // neuron's last /health (#53). Unknown load (no poll
+                        // yet) scores 0 so the replica stays eligible.
+                        let score = node
+                            .model_load
+                            .get(model_id)
+                            .map(|l| l.in_flight + l.queue_depth)
+                            .unwrap_or(0);
+                        loaded_candidates.push((node.name.clone(), node.endpoint.clone(), score));
                     }
                     ModelStatus::Unloaded => {
                         if unloaded_route.is_none() {
@@ -138,6 +193,12 @@ pub async fn resolve(
                 }
             }
         }
+        // Pick the least-busy loaded replica; ties break by node name for
+        // deterministic routing. `false` = not a cold start.
+        let loaded_route = loaded_candidates
+            .into_iter()
+            .min_by(|a, b| a.2.cmp(&b.2).then_with(|| a.0.cmp(&b.0)))
+            .map(|(name, endpoint, _score)| (name, endpoint, false));
         (loaded_route, unloaded_route, recovering_node, any_healthy)
     };
 
@@ -200,11 +261,32 @@ async fn pick_feasible_neuron(
         b.2.cmp(&a.2) // pinned first (true > false)
             .then(a.0.cmp(&b.0))
     });
-    let pick = candidates.into_iter().next();
-    pick.map(|(n, e, _)| (n, e))
-        .ok_or_else(|| RouteError::NoFeasibleNeuron {
+    if let Some((n, e, _)) = candidates.into_iter().next() {
+        return Ok((n, e));
+    }
+
+    // No *healthy* feasible neuron. Distinguish a transient outage from a
+    // permanent misconfiguration: if some neuron is topologically feasible
+    // but currently unhealthy (e.g. it briefly missed polls while busy),
+    // this is retryable — return 503 + Retry-After so the client backs off
+    // and retries instead of treating a 404 as a hard failure. Only when no
+    // neuron could *ever* satisfy the topology is it a permanent 404.
+    let feasible_but_unhealthy = nodes.values().any(|node| {
+        !node.healthy
+            && node
+                .discovery
+                .as_ref()
+                .is_some_and(|disc| profile.is_feasible_on(&node.name, &disc.devices))
+    });
+    if feasible_but_unhealthy {
+        Err(RouteError::FeasibleNodeUnhealthy {
             model_id: profile.id.clone(),
         })
+    } else {
+        Err(RouteError::NoFeasibleNeuron {
+            model_id: profile.id.clone(),
+        })
+    }
 }
 
 /// Issue `POST {endpoint}/models/load` for this profile on this neuron,
@@ -281,6 +363,9 @@ async fn cold_load(
                     last_accessed: Some(chrono::Utc::now()),
                     vram_estimate_mb: profile.vram_mb,
                     capabilities: Vec::new(),
+                    tool_call: false,
+                    reasoning: false,
+                    limit: None,
                 },
             );
         }
@@ -440,6 +525,9 @@ mod tests {
             min_device_vram_mb: None,
             pinned_on: vec![],
             source: source.map(String::from),
+            limit: None,
+            cost: None,
+            capabilities: vec![],
         }
     }
 

crates/cortex-gateway/src/served_usage.rs

@@ -0,0 +1,105 @@
+//! Served-usage ledger (#58): cortex meters, per principal and per UTC day,
+//! the tokens it has served on behalf of mesh accounts, and periodically
+//! reports **absolute** cumulative counters to helexa-upstream for
+//! reconciliation (operators are compensated for served tokens).
+//!
+//! Counters are cumulative-since-process-start for the current period;
+//! upstream upserts them monotonically (GREATEST), so re-sending the same
+//! value is idempotent and a flush that races another is harmless. (A
+//! process restart resets the in-memory counter; the monotonic upsert keeps
+//! upstream from regressing — at most it under-counts the restarted window,
+//! acceptable for beta. One cortex per operator token is assumed.)
+
+use serde::Serialize;
+use std::collections::HashMap;
+use std::sync::Mutex;
+
+#[derive(Debug, Clone, Serialize, PartialEq, Eq)]
+pub struct ServedRow {
+    pub account_id: String,
+    pub key_id: String,
+    pub period: String, // YYYY-MM-DD (UTC)
+    pub served_tokens: u64,
+}
+
+#[derive(Default)]
+pub struct ServedUsage {
+    inner: Mutex<HashMap<(String, String, String), u64>>,
+}
+
+impl ServedUsage {
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    /// Add served tokens for a principal in today's (UTC) period.
+    pub fn add(&self, account_id: &str, key_id: &str, tokens: u64) {
+        if tokens == 0 {
+            return;
+        }
+        let period = chrono::Utc::now().format("%Y-%m-%d").to_string();
+        let mut m = self.inner.lock().expect("served-usage lock");
+        *m.entry((account_id.to_string(), key_id.to_string(), period))
+            .or_insert(0) += tokens;
+    }
+
+    /// Absolute cumulative counters, for a flush to upstream.
+    pub fn snapshot(&self) -> Vec<ServedRow> {
+        let m = self.inner.lock().expect("served-usage lock");
+        m.iter()
+            .map(|((account_id, key_id, period), &served_tokens)| ServedRow {
+                account_id: account_id.clone(),
+                key_id: key_id.clone(),
+                period: period.clone(),
+                served_tokens,
+            })
+            .collect()
+    }
+}
+
+/// POST the absolute counters to upstream's `/authz/v1/served-usage`.
+pub async fn report(
+    client: &reqwest::Client,
+    base_url: &str,
+    bearer: &str,
+    rows: &[ServedRow],
+) -> Result<(), reqwest::Error> {
+    if rows.is_empty() {
+        return Ok(());
+    }
+    let url = format!("{}/authz/v1/served-usage", base_url.trim_end_matches('/'));
+    client
+        .post(url)
+        .bearer_auth(bearer)
+        .json(&serde_json::json!({ "rows": rows }))
+        .send()
+        .await?
+        .error_for_status()?;
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn accumulates_per_principal_and_period() {
+        let su = ServedUsage::new();
+        su.add("acct", "key", 10);
+        su.add("acct", "key", 5);
+        su.add("acct", "other", 7);
+        su.add("acct", "key", 0); // no-op
+        let mut rows = su.snapshot();
+        rows.sort_by(|a, b| a.key_id.cmp(&b.key_id));
+        assert_eq!(rows.len(), 2);
+        let key_row = rows.iter().find(|r| r.key_id == "key").unwrap();
+        assert_eq!(key_row.served_tokens, 15);
+        assert_eq!(
+            rows.iter()
+                .find(|r| r.key_id == "other")
+                .unwrap()
+                .served_tokens,
+            7
+        );
+    }
+}

crates/cortex-gateway/src/state.rs

@@ -1,7 +1,12 @@
+use crate::entitlements_chain::ChainedEntitlementProvider;
+use crate::entitlements_local::LocalEntitlementProvider;
+use crate::entitlements_upstream::UpstreamEntitlementProvider;
 use cortex_core::catalogue::ModelCatalogue;
 use cortex_core::config::{EvictionSettings, GatewayConfig, NeuronEndpoint};
+use cortex_core::entitlements::EntitlementProvider;
 use cortex_core::node::NodeState;
 use std::collections::HashMap;
+use std::sync::Arc;
 use tokio::sync::RwLock;
 
 /// Shared fleet state, protected by a RwLock for concurrent reader access.
@@ -11,6 +16,15 @@ pub struct CortexState {
     pub eviction: EvictionSettings,
     pub catalogue: ModelCatalogue,
     pub http_client: reqwest::Client,
+    /// Resolves bearer keys to principals and enforces token budgets (#47).
+    /// A local/static provider today (#50); the upstream client later (#57).
+    pub entitlements: Arc<dyn EntitlementProvider>,
+    /// Whether to reject unauthenticated requests (#49). Read by the auth
+    /// middleware once it lands.
+    pub require_auth: bool,
+    /// Per-principal served-token tally (#58), reported to upstream for
+    /// operator reconciliation by the flush task when upstream is enabled.
+    pub served_usage: Arc<crate::served_usage::ServedUsage>,
 }
 
 impl CortexState {
@@ -28,12 +42,29 @@ impl CortexState {
                     last_poll: None,
                     discovery: None,
                     activation: None,
+                    model_load: HashMap::new(),
+                    consecutive_poll_failures: 0,
                 },
             );
         }
 
         let catalogue = ModelCatalogue::load(&config.models_config);
 
+        // Local provider always handles operator + infra keys. When the
+        // upstream client is enabled (#57), wrap it in the chain so locally
+        // unknown keys fall through to the mesh authority; otherwise stay
+        // purely local.
+        let local = LocalEntitlementProvider::from_config(&config.entitlements);
+        let entitlements: Arc<dyn EntitlementProvider> = if config.upstream.enabled {
+            tracing::info!(url = %config.upstream.url, "upstream entitlement client enabled");
+            Arc::new(ChainedEntitlementProvider::new(
+                local,
+                UpstreamEntitlementProvider::new(&config.upstream),
+            ))
+        } else {
+            Arc::new(local)
+        };
+
         Self {
             nodes: RwLock::new(nodes),
             neuron_configs: config.neurons.clone(),
@@ -43,6 +74,9 @@ impl CortexState {
                 .timeout(std::time::Duration::from_secs(300))
                 .build()
                 .expect("failed to build HTTP client"),
+            entitlements,
+            require_auth: config.entitlements.require_auth,
+            served_usage: Arc::new(crate::served_usage::ServedUsage::new()),
         }
     }
 }

crates/cortex-gateway/tests/aliases.rs

@@ -56,6 +56,8 @@ async fn test_alias_resolves_in_chat_completions() {
             endpoint: mock_url,
         }],
         models_config: models_path.to_string_lossy().to_string(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -75,6 +77,9 @@ async fn test_alias_resolves_in_chat_completions() {
                 last_accessed: None,
                 vram_estimate_mb: None,
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
     }
@@ -138,6 +143,8 @@ async fn test_aliases_surface_in_v1_models() {
             endpoint: mock_url,
         }],
         models_config: models_path.to_string_lossy().to_string(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -156,6 +163,9 @@ async fn test_aliases_surface_in_v1_models() {
                 last_accessed: None,
                 vram_estimate_mb: Some(2000),
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
     }
@@ -223,6 +233,8 @@ async fn test_alias_falls_through_for_unmapped_model() {
             endpoint: mock_url,
         }],
         models_config: models_path.to_string_lossy().to_string(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -238,6 +250,9 @@ async fn test_alias_falls_through_for_unmapped_model() {
                 last_accessed: None,
                 vram_estimate_mb: None,
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
     }

crates/cortex-gateway/tests/anthropic.rs

@@ -124,6 +124,94 @@ async fn test_anthropic_invalid_request() {
     assert_eq!(resp.status(), 400);
 }
 
+/// Tool round-trip: an Anthropic `/v1/messages` request carrying tools
+/// (the Claude Code shape: `{name, description, input_schema}`) must
+/// reach the upstream neuron reshaped into OpenAI function-tool form,
+/// and tool history (`tool_use` / `tool_result` blocks) must become
+/// `tool_calls` / `role:"tool"` messages. This is the fix for the
+/// failure where the model received malformed tool defs and improvised
+/// an unparseable `<tool_use_name>` format.
+#[tokio::test]
+async fn test_anthropic_tools_reshaped_for_upstream() {
+    let (mock_url, captured) = common::spawn_capturing_mock_neuron().await;
+    let gw_url = common::spawn_gateway(&mock_url).await;
+
+    let client = reqwest::Client::new();
+    let resp = client
+        .post(format!("{gw_url}/v1/messages"))
+        .header("content-type", "application/json")
+        .json(&json!({
+            "model": "test-model",
+            "max_tokens": 100,
+            "tools": [{
+                "name": "Read",
+                "description": "Read a file from disk",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {"path": {"type": "string"}},
+                    "required": ["path"]
+                }
+            }],
+            "tool_choice": {"type": "auto"},
+            "messages": [
+                {"role": "user", "content": "read /etc/hosts"},
+                {"role": "assistant", "content": [
+                    {"type": "text", "text": "Reading it."},
+                    {"type": "tool_use", "id": "toolu_42", "name": "Read",
+                     "input": {"path": "/etc/hosts"}}
+                ]},
+                {"role": "user", "content": [
+                    {"type": "tool_result", "tool_use_id": "toolu_42",
+                     "content": "127.0.0.1 localhost"}
+                ]}
+            ]
+        }))
+        .send()
+        .await
+        .expect("request should succeed");
+    assert_eq!(resp.status(), 200);
+
+    let forwarded = {
+        let guard = captured.lock().unwrap();
+        guard.last().cloned().expect("upstream received a request")
+    };
+
+    // Tool definitions reshaped to OpenAI function form.
+    let tools = forwarded["tools"].as_array().expect("tools array");
+    assert_eq!(tools[0]["type"], "function");
+    assert_eq!(tools[0]["function"]["name"], "Read");
+    assert_eq!(
+        tools[0]["function"]["parameters"]["properties"]["path"]["type"],
+        "string"
+    );
+    assert!(tools[0]["function"].get("input_schema").is_none());
+
+    // tool_choice mapped.
+    assert_eq!(forwarded["tool_choice"], "auto");
+
+    // Message history: user, assistant(+tool_calls), tool, user.
+    let msgs = forwarded["messages"].as_array().expect("messages array");
+    let assistant = msgs
+        .iter()
+        .find(|m| m["role"] == "assistant")
+        .expect("assistant turn");
+    assert_eq!(assistant["tool_calls"][0]["id"], "toolu_42");
+    assert_eq!(assistant["tool_calls"][0]["function"]["name"], "Read");
+    // arguments is the parsed object, not a JSON string — the Qwen3.6
+    // chat template iterates `tool_call.arguments | items`.
+    assert_eq!(
+        assistant["tool_calls"][0]["function"]["arguments"],
+        json!({"path": "/etc/hosts"})
+    );
+
+    let tool_msg = msgs
+        .iter()
+        .find(|m| m["role"] == "tool")
+        .expect("tool result turn");
+    assert_eq!(tool_msg["tool_call_id"], "toolu_42");
+    assert_eq!(tool_msg["content"], "127.0.0.1 localhost");
+}
+
 /// #24: a streaming Anthropic request gets a translated Anthropic SSE
 /// stream — not raw OpenAI frames. Verifies the full event sequence,
 /// text reassembly, and the content type.

crates/cortex-gateway/tests/auth.rs

@@ -0,0 +1,273 @@
+//! Integration tests for API-key auth + principal resolution (#49).
+//!
+//! Verifies the #63 rejection contract (401 invalid_api_key via the #60
+//! envelope) and that an authenticated request reaches neuron carrying the
+//! internal principal headers — while a client-supplied principal header is
+//! stripped (anti-spoofing).
+
+use axum::Json;
+use axum::extract::Path;
+use axum::http::HeaderMap;
+use axum::routing::{get, post};
+use cortex_core::config::{
+    ApiKeyConfig, EntitlementsConfig, EvictionSettings, EvictionStrategy, GatewayConfig,
+    GatewaySettings, NeuronEndpoint,
+};
+use cortex_core::entitlements::{CapWindow, HEADER_ACCOUNT_ID, HEADER_KEY_ID};
+use cortex_core::node::{ModelEntry, ModelStatus};
+use cortex_gateway::state::CortexState;
+use serde_json::{Value, json};
+use std::sync::{Arc, Mutex};
+use tokio::net::TcpListener;
+
+/// What the mock neuron observed on the inbound `/v1/chat/completions`
+/// request: the principal headers cortex stamped (or didn't).
+#[derive(Default)]
+struct Seen {
+    account_id: Option<String>,
+    key_id: Option<String>,
+}
+
+/// Spawn a mock neuron that records the principal headers it receives and
+/// returns a trivial chat completion. Returns (base_url, observed).
+async fn spawn_capturing_neuron() -> (String, Arc<Mutex<Seen>>) {
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    let base_url = format!("http://{addr}");
+    let inference_url = base_url.clone();
+    let seen: Arc<Mutex<Seen>> = Arc::new(Mutex::new(Seen::default()));
+    let sink = Arc::clone(&seen);
+
+    let app = axum::Router::new()
+        .route(
+            "/models/{model_id}/endpoint",
+            get(move |Path(_): Path<String>| {
+                let url = inference_url.clone();
+                async move { Json(json!({ "url": url })) }
+            }),
+        )
+        .route(
+            "/v1/chat/completions",
+            post(move |headers: HeaderMap, Json(body): Json<Value>| {
+                let sink = Arc::clone(&sink);
+                async move {
+                    {
+                        let mut s = sink.lock().unwrap();
+                        s.account_id = headers
+                            .get(HEADER_ACCOUNT_ID)
+                            .and_then(|v| v.to_str().ok())
+                            .map(str::to_string);
+                        s.key_id = headers
+                            .get(HEADER_KEY_ID)
+                            .and_then(|v| v.to_str().ok())
+                            .map(str::to_string);
+                    }
+                    let model = body.get("model").and_then(Value::as_str).unwrap_or("m");
+                    Json(json!({
+                        "id": "chatcmpl-auth-001",
+                        "object": "chat.completion",
+                        "created": 1700000000_u64,
+                        "model": model,
+                        "choices": [{
+                            "index": 0,
+                            "message": {"role": "assistant", "content": "ok"},
+                            "finish_reason": "stop"
+                        }],
+                        "usage": {"prompt_tokens": 3, "completion_tokens": 1, "total_tokens": 4}
+                    }))
+                }
+            }),
+        )
+        .with_state(());
+
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+
+    (base_url, seen)
+}
+
+/// Spawn a gateway with the given entitlements config, a single neuron, and
+/// `test-model` seeded as loaded (build_app spawns no poller).
+async fn spawn_gateway(neuron_url: &str, entitlements: EntitlementsConfig) -> String {
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        neurons: vec![NeuronEndpoint {
+            name: "mock-node".into(),
+            endpoint: neuron_url.to_string(),
+        }],
+        models_config: "/dev/null".into(),
+        entitlements,
+        upstream: Default::default(),
+    };
+
+    let fleet = Arc::new(CortexState::from_config(&config));
+    {
+        let mut nodes = fleet.nodes.write().await;
+        let node = nodes.get_mut("mock-node").unwrap();
+        node.healthy = true;
+        node.models.insert(
+            "test-model".into(),
+            ModelEntry {
+                id: "test-model".into(),
+                status: ModelStatus::Loaded,
+                last_accessed: None,
+                vram_estimate_mb: Some(8000),
+                capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
+            },
+        );
+    }
+
+    let app = cortex_gateway::build_app(Arc::clone(&fleet));
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+    format!("http://{addr}")
+}
+
+fn one_key_config(require_auth: bool) -> EntitlementsConfig {
+    EntitlementsConfig {
+        require_auth,
+        keys: vec![ApiKeyConfig {
+            key: "sk-good".into(),
+            account_id: "acct-1".into(),
+            key_id: Some("key-1".into()),
+            hard_cap: None,
+            window: CapWindow::Balance,
+        }],
+    }
+}
+
+fn chat_body() -> Value {
+    json!({
+        "model": "test-model",
+        "messages": [{"role": "user", "content": "hi"}]
+    })
+}
+
+#[tokio::test]
+async fn missing_key_when_required_is_401_invalid_api_key() {
+    let (neuron, _seen) = spawn_capturing_neuron().await;
+    let gateway = spawn_gateway(&neuron, one_key_config(true)).await;
+
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .json(&chat_body())
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::UNAUTHORIZED);
+    let body: Value = resp.json().await.unwrap();
+    assert_eq!(body["error"]["code"], "invalid_api_key");
+    assert_eq!(body["error"]["type"], "invalid_request_error");
+}
+
+#[tokio::test]
+async fn unrecognized_key_is_ignored_when_auth_not_required() {
+    let (neuron, seen) = spawn_capturing_neuron().await;
+    // allow-anonymous mode: a placeholder/unknown bearer (as opencode,
+    // Open WebUI, Agent Zero, litellm all send by default) must NOT be
+    // rejected — it's ignored and the request is served anonymously.
+    let gateway = spawn_gateway(&neuron, one_key_config(false)).await;
+
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .bearer_auth("sk-dummy-placeholder")
+        .json(&chat_body())
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::OK);
+    let _ = resp.bytes().await.unwrap();
+    // Served, but anonymous — no principal stamped from the bogus key.
+    assert!(seen.lock().unwrap().account_id.is_none());
+}
+
+#[tokio::test]
+async fn invalid_key_is_401_when_auth_required() {
+    let (neuron, seen) = spawn_capturing_neuron().await;
+    // With auth required, a present-but-wrong credential is rejected.
+    let gateway = spawn_gateway(&neuron, one_key_config(true)).await;
+
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .bearer_auth("sk-wrong")
+        .json(&chat_body())
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::UNAUTHORIZED);
+    let body: Value = resp.json().await.unwrap();
+    assert_eq!(body["error"]["code"], "invalid_api_key");
+    // Rejected before dispatch — neuron never saw the request.
+    assert!(seen.lock().unwrap().account_id.is_none());
+}
+
+#[tokio::test]
+async fn valid_key_reaches_neuron_with_principal_headers() {
+    let (neuron, seen) = spawn_capturing_neuron().await;
+    let gateway = spawn_gateway(&neuron, one_key_config(true)).await;
+
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .bearer_auth("sk-good")
+        // A spoofed principal header must be stripped, not forwarded.
+        .header(HEADER_ACCOUNT_ID, "attacker")
+        .json(&chat_body())
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::OK);
+    let s = seen.lock().unwrap();
+    assert_eq!(s.account_id.as_deref(), Some("acct-1"));
+    assert_eq!(s.key_id.as_deref(), Some("key-1"));
+}
+
+#[tokio::test]
+async fn anonymous_allowed_when_auth_not_required() {
+    let (neuron, seen) = spawn_capturing_neuron().await;
+    let gateway = spawn_gateway(&neuron, EntitlementsConfig::default()).await;
+
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .json(&chat_body())
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::OK);
+    // No principal resolved → no principal headers stamped.
+    let s = seen.lock().unwrap();
+    assert!(s.account_id.is_none());
+    assert!(s.key_id.is_none());
+}
+
+#[tokio::test]
+async fn health_is_public_even_when_auth_required() {
+    let (neuron, _seen) = spawn_capturing_neuron().await;
+    let gateway = spawn_gateway(&neuron, one_key_config(true)).await;
+
+    let resp = reqwest::Client::new()
+        .get(format!("{gateway}/health"))
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::OK);
+}

crates/cortex-gateway/tests/budget_enforcement.rs

@@ -0,0 +1,254 @@
+//! Integration tests for budget enforcement (#52) — the A0 seatbelt.
+//!
+//! A reservation over the key's hard cap is refused *before* neuron is hit,
+//! with the #63 code matching the cap-window semantics (rate_limit_exceeded
+//! + Retry-After for a resetting window, insufficient_quota for a hard
+//! balance). Spend never exceeds the cap. No 402, ever.
+
+use axum::Json;
+use axum::extract::Path;
+use axum::routing::{get, post};
+use cortex_core::config::{
+    ApiKeyConfig, EntitlementsConfig, EvictionSettings, EvictionStrategy, GatewayConfig,
+    GatewaySettings, NeuronEndpoint,
+};
+use cortex_core::entitlements::{CapWindow, Principal};
+use cortex_core::node::{ModelEntry, ModelStatus};
+use cortex_gateway::state::CortexState;
+use serde_json::{Value, json};
+use std::sync::Arc;
+use std::sync::atomic::{AtomicU64, Ordering};
+use tokio::net::TcpListener;
+
+/// Mock neuron with a hit counter on the inference path, so a test can prove
+/// a request was (or wasn't) dispatched.
+async fn spawn_counting_neuron() -> (String, Arc<AtomicU64>) {
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    let base_url = format!("http://{addr}");
+    let inference_url = base_url.clone();
+    let hits = Arc::new(AtomicU64::new(0));
+    let sink = Arc::clone(&hits);
+
+    let app = axum::Router::new()
+        .route(
+            "/models/{model_id}/endpoint",
+            get(move |Path(_): Path<String>| {
+                let url = inference_url.clone();
+                async move { Json(json!({ "url": url })) }
+            }),
+        )
+        .route(
+            "/v1/chat/completions",
+            post(move |Json(body): Json<Value>| {
+                let sink = Arc::clone(&sink);
+                async move {
+                    sink.fetch_add(1, Ordering::SeqCst);
+                    let model = body.get("model").and_then(Value::as_str).unwrap_or("m");
+                    Json(json!({
+                        "id": "chatcmpl-budget",
+                        "object": "chat.completion",
+                        "created": 1700000000_u64,
+                        "model": model,
+                        "choices": [{"index": 0, "message": {"role": "assistant", "content": "ok"}, "finish_reason": "stop"}],
+                        "usage": {"prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15}
+                    }))
+                }
+            }),
+        );
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+    (base_url, hits)
+}
+
+async fn spawn_gateway(neuron_url: &str, key: ApiKeyConfig) -> (Arc<CortexState>, String) {
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        neurons: vec![NeuronEndpoint {
+            name: "mock-node".into(),
+            endpoint: neuron_url.to_string(),
+        }],
+        models_config: "/dev/null".into(),
+        entitlements: EntitlementsConfig {
+            require_auth: true,
+            keys: vec![key],
+        },
+        upstream: Default::default(),
+    };
+    let fleet = Arc::new(CortexState::from_config(&config));
+    {
+        let mut nodes = fleet.nodes.write().await;
+        let node = nodes.get_mut("mock-node").unwrap();
+        node.healthy = true;
+        node.models.insert(
+            "test-model".into(),
+            ModelEntry {
+                id: "test-model".into(),
+                status: ModelStatus::Loaded,
+                last_accessed: None,
+                vram_estimate_mb: Some(8000),
+                capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
+            },
+        );
+    }
+    let app = cortex_gateway::build_app(Arc::clone(&fleet));
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+    (fleet, format!("http://{addr}"))
+}
+
+fn key(window: CapWindow, hard_cap: u64) -> ApiKeyConfig {
+    ApiKeyConfig {
+        key: "sk-cap".into(),
+        account_id: "acct-cap".into(),
+        key_id: Some("key-cap".into()),
+        hard_cap: Some(hard_cap),
+        window,
+    }
+}
+
+fn chat(max_tokens: u64) -> Value {
+    json!({
+        "model": "test-model",
+        "max_tokens": max_tokens,
+        "messages": [{"role": "user", "content": "hi"}]
+    })
+}
+
+#[tokio::test]
+async fn balance_over_cap_is_429_insufficient_quota_before_dispatch() {
+    let (neuron, hits) = spawn_counting_neuron().await;
+    // Cap far below a single request's reservation (max_tokens 1000).
+    let (_fleet, gateway) = spawn_gateway(&neuron, key(CapWindow::Balance, 10)).await;
+
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .bearer_auth("sk-cap")
+        .json(&chat(1000))
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::TOO_MANY_REQUESTS);
+    // Hard balance → no Retry-After.
+    assert!(resp.headers().get(reqwest::header::RETRY_AFTER).is_none());
+    let body: Value = resp.json().await.unwrap();
+    assert_eq!(body["error"]["code"], "insufficient_quota");
+    // Refused before dispatch — neuron never saw it.
+    assert_eq!(hits.load(Ordering::SeqCst), 0);
+}
+
+#[tokio::test]
+async fn rolling_over_cap_is_429_rate_limited_with_retry_after() {
+    let (neuron, hits) = spawn_counting_neuron().await;
+    let (_fleet, gateway) =
+        spawn_gateway(&neuron, key(CapWindow::Rolling { seconds: 3600 }, 10)).await;
+
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .bearer_auth("sk-cap")
+        .json(&chat(1000))
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::TOO_MANY_REQUESTS);
+    let retry = resp
+        .headers()
+        .get(reqwest::header::RETRY_AFTER)
+        .expect("rolling-window rejection must carry Retry-After");
+    assert!(retry.to_str().unwrap().parse::<u64>().unwrap() >= 1);
+    let body: Value = resp.json().await.unwrap();
+    assert_eq!(body["error"]["code"], "rate_limit_exceeded");
+    assert_eq!(hits.load(Ordering::SeqCst), 0);
+}
+
+#[tokio::test]
+async fn within_cap_is_served() {
+    let (neuron, hits) = spawn_counting_neuron().await;
+    let (_fleet, gateway) = spawn_gateway(&neuron, key(CapWindow::Balance, 1_000_000)).await;
+
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .bearer_auth("sk-cap")
+        .json(&chat(50))
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::OK);
+    let _ = resp.bytes().await.unwrap();
+    assert_eq!(hits.load(Ordering::SeqCst), 1);
+}
+
+#[tokio::test]
+async fn a0_seatbelt_caps_a_runaway_fan_out() {
+    // An Agent-Zero-style key with a modest cap: a burst of requests drains
+    // it, then further requests are refused — the account stops draining and
+    // spend never exceeds the cap.
+    let (neuron, hits) = spawn_counting_neuron().await;
+    let (fleet, gateway) = spawn_gateway(&neuron, key(CapWindow::Balance, 100)).await;
+    let client = reqwest::Client::new();
+
+    let mut ok = 0;
+    let mut refused = 0;
+    for _ in 0..20 {
+        let resp = client
+            .post(format!("{gateway}/v1/chat/completions"))
+            .bearer_auth("sk-cap")
+            .json(&chat(20))
+            .send()
+            .await
+            .unwrap();
+        match resp.status() {
+            reqwest::StatusCode::OK => {
+                ok += 1;
+                let _ = resp.bytes().await.unwrap();
+            }
+            reqwest::StatusCode::TOO_MANY_REQUESTS => {
+                refused += 1;
+                let body: Value = resp.json().await.unwrap();
+                assert_eq!(body["error"]["code"], "insufficient_quota");
+            }
+            other => panic!("unexpected status {other}"),
+        }
+    }
+
+    assert!(ok >= 1, "some requests should be served");
+    assert!(refused >= 1, "the cap must eventually refuse the fan-out");
+    assert_eq!(
+        hits.load(Ordering::SeqCst),
+        ok,
+        "refused requests never dispatched"
+    );
+
+    // Spend never exceeded the hard cap (reservation prevents overshoot).
+    // Poll briefly for in-flight settles to land.
+    let principal = Principal {
+        account_id: "acct-cap".into(),
+        key_id: "key-cap".into(),
+    };
+    for _ in 0..50 {
+        let snap = fleet.entitlements.snapshot(&principal).await.unwrap();
+        if snap.reserved == 0 {
+            break;
+        }
+        tokio::time::sleep(std::time::Duration::from_millis(20)).await;
+    }
+    let snap = fleet.entitlements.snapshot(&principal).await.unwrap();
+    assert!(snap.spent <= 100, "spent {} exceeded cap", snap.spent);
+}

crates/cortex-gateway/tests/common/mod.rs

@@ -54,9 +54,64 @@ pub async fn spawn_mock_neuron() -> String {
     base_url
 }
 
+/// Like [`spawn_mock_neuron`] but captures the JSON body of every
+/// `POST /v1/chat/completions` it receives into the returned handle, so
+/// a test can assert what the gateway *actually forwarded upstream*
+/// (e.g. that Anthropic-shaped tools were reshaped to OpenAI form).
+pub async fn spawn_capturing_mock_neuron() -> (String, Arc<std::sync::Mutex<Vec<Value>>>) {
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    let base_url = format!("http://{addr}");
+    let inference_url = base_url.clone();
+    let captured: Arc<std::sync::Mutex<Vec<Value>>> = Arc::new(std::sync::Mutex::new(Vec::new()));
+    let sink = captured.clone();
+
+    let app = Router::new()
+        .route("/models", get(mock_neuron_list_models))
+        .route(
+            "/models/{model_id}/endpoint",
+            get(move |Path(_): Path<String>| {
+                let url = inference_url.clone();
+                async move { Json(json!({"url": url})) }
+            }),
+        )
+        .route(
+            "/v1/chat/completions",
+            post(move |Json(body): Json<Value>| {
+                let sink = sink.clone();
+                async move {
+                    let model = body
+                        .get("model")
+                        .and_then(|v| v.as_str())
+                        .unwrap_or("unknown");
+                    let resp = json!({
+                        "id": "chatcmpl-capture-001",
+                        "object": "chat.completion",
+                        "created": 1700000000_u64,
+                        "model": model,
+                        "choices": [{
+                            "index": 0,
+                            "message": {"role": "assistant", "content": "Hello from mock backend"},
+                            "finish_reason": "stop"
+                        }],
+                        "usage": {"prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15}
+                    });
+                    sink.lock().unwrap().push(body);
+                    Json(resp)
+                }
+            }),
+        );
+
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+
+    (base_url, captured)
+}
+
 async fn mock_neuron_list_models() -> Json<Value> {
     Json(json!([
-        {"id": "test-model", "harness": "candle", "status": "loaded", "devices": [0], "vram_used_mb": 8000}
+        {"id": "test-model", "harness": "candle", "status": "loaded", "devices": [0], "vram_used_mb": 8000, "capabilities": ["text"], "tool_call": false, "reasoning": false}
     ]))
 }
 
@@ -374,6 +429,8 @@ pub async fn spawn_gateway_with_state(mock_url: &str) -> (Arc<CortexState>, Stri
             endpoint: mock_url.to_string(),
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -391,6 +448,9 @@ pub async fn spawn_gateway_with_state(mock_url: &str) -> (Arc<CortexState>, Stri
                 last_accessed: None,
                 vram_estimate_mb: Some(8000),
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
     }

crates/cortex-gateway/tests/error_envelope.rs

@@ -0,0 +1,141 @@
+mod common;
+
+use serde_json::json;
+
+#[tokio::test]
+async fn error_response_model_not_found() {
+    let neuron_url = common::spawn_mock_neuron().await;
+    let gateway_url = common::spawn_gateway(&neuron_url).await;
+
+    let client = reqwest::Client::new();
+
+    // Request a model that isn't loaded on the mock neuron.
+    let resp = client
+        .post(format!("{gateway_url}/v1/chat/completions"))
+        .header("Content-Type", "application/json")
+        .json(&json!({
+            "model": "nonexistent-model",
+            "messages": [{"role": "user", "content": "hi"}]
+        }))
+        .send()
+        .await
+        .expect("request should succeed");
+
+    assert_eq!(resp.status(), axum::http::StatusCode::NOT_FOUND);
+
+    let body: serde_json::Value = resp.json().await.expect("valid json");
+    let err = body.get("error").expect("response has error object");
+
+    // Broad type categorization
+    assert_eq!(err.get("type").unwrap(), "invalid_request_error");
+    // Specific machine-readable code
+    assert_eq!(
+        err.get("code").unwrap().as_str().unwrap(),
+        "model_not_found"
+    );
+    // param is always null
+    assert!(err.get("param").unwrap().is_null());
+}
+
+#[tokio::test]
+async fn error_response_missing_model_field() {
+    let neuron_url = common::spawn_mock_neuron().await;
+    let gateway_url = common::spawn_gateway(&neuron_url).await;
+
+    let client = reqwest::Client::new();
+
+    // Request without the required `model` field.
+    let resp = client
+        .post(format!("{gateway_url}/v1/chat/completions"))
+        .header("Content-Type", "application/json")
+        .json(&json!({
+            "messages": [{"role": "user", "content": "hi"}]
+        }))
+        .send()
+        .await
+        .expect("request should succeed");
+
+    assert_eq!(resp.status(), axum::http::StatusCode::BAD_REQUEST);
+
+    let body: serde_json::Value = resp.json().await.expect("valid json");
+    let err = body.get("error").expect("response has error object");
+
+    assert_eq!(err.get("type").unwrap(), "invalid_request_error");
+    assert_eq!(
+        err.get("code").unwrap().as_str().unwrap(),
+        "missing_model_field"
+    );
+    assert!(err.get("param").unwrap().is_null());
+}
+
+#[tokio::test]
+async fn error_response_no_healthy_nodes() {
+    use cortex_core::config::{EvictionSettings, GatewayConfig, GatewaySettings, NeuronEndpoint};
+    use std::sync::Arc;
+
+    // Create a gateway config with a neuron pointing at an unreachable port so no node is ever healthy.
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: cortex_core::config::EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        neurons: vec![NeuronEndpoint {
+            name: "dead-node".into(),
+            endpoint: "http://127.0.0.1:1".into(),
+        }],
+        models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
+    };
+
+    let fleet = Arc::new(cortex_gateway::state::CortexState::from_config(&config));
+
+    let app = cortex_gateway::build_app(fleet);
+    let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+
+    // Allow the poller a moment to mark the node unhealthy.
+    tokio::time::sleep(std::time::Duration::from_millis(200)).await;
+
+    let client = reqwest::Client::new();
+    let resp = client
+        .post(format!("http://{addr}/v1/chat/completions"))
+        .header("Content-Type", "application/json")
+        .json(&json!({
+            "model": "any-model",
+            "messages": [{"role": "user", "content": "hi"}]
+        }))
+        .send()
+        .await
+        .expect("request should succeed");
+
+    assert_eq!(resp.status(), axum::http::StatusCode::SERVICE_UNAVAILABLE);
+
+    // Transient 503 — the gateway advertises Retry-After so OpenAI-compatible
+    // clients back off and retry rather than surfacing an opaque error (#63).
+    let retry_after = resp
+        .headers()
+        .get(reqwest::header::RETRY_AFTER)
+        .expect("transient 503 must carry Retry-After")
+        .to_str()
+        .unwrap()
+        .to_string();
+    assert_eq!(retry_after, "5");
+
+    let body: serde_json::Value = resp.json().await.expect("valid json");
+    let err = body.get("error").expect("response has error object");
+
+    assert_eq!(err.get("type").unwrap(), "api_error");
+    assert_eq!(
+        err.get("code").unwrap().as_str().unwrap(),
+        "service_unavailable"
+    );
+    assert!(err.get("param").unwrap().is_null());
+}

crates/cortex-gateway/tests/eviction.rs

@@ -71,6 +71,8 @@ fn make_fleet(endpoint: &str, defrag_after: u32) -> Arc<CortexState> {
             endpoint: endpoint.to_string(),
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
     Arc::new(CortexState::from_config(&config))
 }
@@ -92,6 +94,9 @@ async fn test_evict_lru_model() {
                 last_accessed: Some(Utc::now() - chrono::Duration::hours(2)),
                 vram_estimate_mb: Some(8000),
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
         node.models.insert(
@@ -102,6 +107,9 @@ async fn test_evict_lru_model() {
                 last_accessed: Some(Utc::now()),
                 vram_estimate_mb: Some(8000),
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
     }
@@ -166,6 +174,9 @@ async fn test_eviction_increments_lifecycle_cycles() {
                 last_accessed: None,
                 vram_estimate_mb: None,
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
     }

crates/cortex-gateway/tests/feasibility_routing.rs

@@ -0,0 +1,125 @@
+//! Router: a catalogued model whose only topologically-feasible neuron is
+//! currently unhealthy is a *transient* condition (retryable 503), not a
+//! permanent 404. This is the exact shape of the beast incident: benjy/
+//! quadbrat (1 GPU, healthy) can't host the 27B, and beast (2 GPU) — the
+//! sole feasible node — briefly drops out → clients must back off and retry,
+//! not hard-fail.
+
+use cortex_core::config::{
+    EvictionSettings, EvictionStrategy, GatewayConfig, GatewaySettings, NeuronEndpoint,
+};
+use cortex_core::discovery::{DeviceInfo, DiscoveryResponse};
+use cortex_gateway::router::{self, RouteError};
+use cortex_gateway::state::CortexState;
+use std::sync::Arc;
+
+fn devices(n: usize) -> Vec<DeviceInfo> {
+    (0..n)
+        .map(|i| DeviceInfo {
+            index: i as u32,
+            name: "RTX 5090".into(),
+            vram_total_mb: 32_768,
+            compute_capability: "9.0".into(),
+        })
+        .collect()
+}
+
+fn discovery(host: &str, n_devices: usize) -> DiscoveryResponse {
+    DiscoveryResponse {
+        hostname: host.into(),
+        os: "Linux".into(),
+        kernel: "7.0".into(),
+        cuda_version: Some("13.0".into()),
+        driver_version: Some("999".into()),
+        devices: devices(n_devices),
+        harnesses: vec!["candle".into()],
+        cuda_unavailable_reason: None,
+        max_prompt_tokens: 49_152,
+    }
+}
+
+/// Catalogue with one model needing 2 devices. Returns a temp path.
+fn write_catalogue() -> std::path::PathBuf {
+    let toml = r#"
+[[models]]
+id = "big-model"
+harness = "candle"
+min_devices = 2
+"#;
+    let path = std::env::temp_dir().join("cortex_test_feasibility_models.toml");
+    std::fs::write(&path, toml).unwrap();
+    path
+}
+
+async fn fleet_with(big_healthy: bool, big_devices: usize) -> Arc<CortexState> {
+    let cat = write_catalogue();
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        neurons: vec![
+            NeuronEndpoint {
+                name: "small".into(),
+                endpoint: "http://127.0.0.1:1".into(),
+            },
+            NeuronEndpoint {
+                name: "big".into(),
+                endpoint: "http://127.0.0.1:2".into(),
+            },
+        ],
+        models_config: cat.to_string_lossy().into_owned(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
+    };
+    let fleet = Arc::new(CortexState::from_config(&config));
+    {
+        let mut nodes = fleet.nodes.write().await;
+        // "small" is healthy but only has 1 GPU → not feasible for the model.
+        let small = nodes.get_mut("small").unwrap();
+        small.healthy = true;
+        small.discovery = Some(discovery("small", 1));
+        // "big" has enough GPUs but its health is the variable under test.
+        let big = nodes.get_mut("big").unwrap();
+        big.healthy = big_healthy;
+        big.discovery = Some(discovery("big", big_devices));
+    }
+    fleet
+}
+
+#[tokio::test]
+async fn feasible_node_unhealthy_is_transient_503() {
+    // big (2 GPU, the only feasible node) is unhealthy; small (1 GPU) is
+    // healthy but can't host the model → retryable, not a permanent 404.
+    let fleet = fleet_with(false, 2).await;
+    let err = router::resolve(&fleet, "big-model")
+        .await
+        .expect_err("model can't be served right now");
+    assert!(
+        matches!(err, RouteError::FeasibleNodeUnhealthy { .. }),
+        "expected FeasibleNodeUnhealthy, got {err:?}"
+    );
+    assert_eq!(err.http_status(), 503);
+    assert_eq!(err.retry_after_secs(), Some(3));
+    assert_eq!(err.code(), "service_unavailable");
+}
+
+#[tokio::test]
+async fn no_node_can_ever_satisfy_is_permanent_404() {
+    // big is healthy but only has 1 GPU now (e.g. topology genuinely can't
+    // satisfy min_devices=2 anywhere) → permanent, non-retryable 404.
+    let fleet = fleet_with(true, 1).await;
+    let err = router::resolve(&fleet, "big-model")
+        .await
+        .expect_err("no feasible topology");
+    assert!(
+        matches!(err, RouteError::NoFeasibleNeuron { .. }),
+        "expected NoFeasibleNeuron, got {err:?}"
+    );
+    assert_eq!(err.http_status(), 404);
+    assert_eq!(err.retry_after_secs(), None);
+}

crates/cortex-gateway/tests/load_routing.rs

@@ -0,0 +1,190 @@
+//! Load-aware routing across replicas (#55).
+//!
+//! When a model is loaded on more than one healthy neuron, the router picks
+//! the least-busy replica using the per-model admission load each neuron
+//! reports on `GET /health` (#53), rather than always taking the first.
+
+mod common;
+
+use axum::Json;
+use axum::extract::Path;
+use axum::http::{StatusCode, header};
+use axum::response::IntoResponse;
+use axum::routing::{get, post};
+use cortex_core::config::{
+    EvictionSettings, EvictionStrategy, GatewayConfig, GatewaySettings, NeuronEndpoint,
+};
+use cortex_core::discovery::ModelLoad;
+use cortex_core::node::{ModelEntry, ModelStatus};
+use cortex_gateway::state::CortexState;
+use serde_json::{Value, json};
+use std::sync::Arc;
+use tokio::net::TcpListener;
+
+/// Seed a node as healthy with `test-model` loaded and a given admission load.
+async fn seed_loaded(fleet: &CortexState, node: &str, in_flight: usize, queue_depth: usize) {
+    let mut nodes = fleet.nodes.write().await;
+    let n = nodes.get_mut(node).expect("node exists");
+    n.healthy = true;
+    n.models.insert(
+        "test-model".into(),
+        ModelEntry {
+            id: "test-model".into(),
+            status: ModelStatus::Loaded,
+            last_accessed: None,
+            vram_estimate_mb: Some(8000),
+            capabilities: Vec::new(),
+            tool_call: false,
+            reasoning: false,
+            limit: None,
+        },
+    );
+    n.model_load.insert(
+        "test-model".into(),
+        ModelLoad {
+            id: "test-model".into(),
+            in_flight,
+            queue_depth,
+        },
+    );
+}
+
+/// Build a gateway state over two mock neurons (no poller; we seed state).
+async fn two_neuron_fleet(endpoint_a: &str, endpoint_b: &str) -> Arc<CortexState> {
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        neurons: vec![
+            NeuronEndpoint {
+                name: "node-a".into(),
+                endpoint: endpoint_a.to_string(),
+            },
+            NeuronEndpoint {
+                name: "node-b".into(),
+                endpoint: endpoint_b.to_string(),
+            },
+        ],
+        models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
+    };
+    Arc::new(CortexState::from_config(&config))
+}
+
+#[tokio::test]
+async fn routes_to_least_busy_replica() {
+    let neuron_a = common::spawn_mock_neuron().await;
+    let neuron_b = common::spawn_mock_neuron().await;
+    let fleet = two_neuron_fleet(&neuron_a, &neuron_b).await;
+
+    // A is busy (1 running + 3 queued), B is idle.
+    seed_loaded(&fleet, "node-a", 1, 3).await;
+    seed_loaded(&fleet, "node-b", 0, 0).await;
+
+    let route = cortex_gateway::router::resolve(&fleet, "test-model")
+        .await
+        .expect("model is loaded on both nodes");
+    assert_eq!(route.node_name, "node-b", "should pick the idle replica");
+
+    // Flip the load: now B is the busy one.
+    seed_loaded(&fleet, "node-a", 0, 0).await;
+    seed_loaded(&fleet, "node-b", 1, 5).await;
+    let route = cortex_gateway::router::resolve(&fleet, "test-model")
+        .await
+        .expect("still loaded");
+    assert_eq!(route.node_name, "node-a", "should follow the lighter load");
+}
+
+/// Mock neuron whose inference endpoint always returns a #63 backpressure
+/// envelope (503 + Retry-After) — simulating a saturated neuron.
+async fn spawn_busy_neuron() -> String {
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    let base_url = format!("http://{addr}");
+    let inference_url = base_url.clone();
+    let app = axum::Router::new()
+        .route(
+            "/models/{model_id}/endpoint",
+            get(move |Path(_): Path<String>| {
+                let url = inference_url.clone();
+                async move { Json(json!({ "url": url })) }
+            }),
+        )
+        .route(
+            "/v1/chat/completions",
+            post(|| async {
+                let body = json!({"error": {
+                    "message": "model is busy (admission queue full); retry shortly",
+                    "type": "rate_limit_error",
+                    "code": "rate_limit_exceeded",
+                    "param": null
+                }});
+                (
+                    StatusCode::SERVICE_UNAVAILABLE,
+                    [(header::RETRY_AFTER, "6")],
+                    Json(body),
+                )
+                    .into_response()
+            }),
+        );
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+    base_url
+}
+
+#[tokio::test]
+async fn neuron_backpressure_is_propagated_intact() {
+    // A saturated neuron's 503 + Retry-After + envelope must reach the client
+    // verbatim — not unwrapped, remapped, or stripped (#55 / #63).
+    let neuron = spawn_busy_neuron().await;
+    let fleet = two_neuron_fleet(&neuron, &neuron).await;
+    seed_loaded(&fleet, "node-a", 1, 8).await;
+
+    let app = cortex_gateway::build_app(Arc::clone(&fleet));
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+
+    let resp = reqwest::Client::new()
+        .post(format!("http://{addr}/v1/chat/completions"))
+        .json(&json!({"model": "test-model", "messages": [{"role": "user", "content": "hi"}]}))
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::SERVICE_UNAVAILABLE);
+    assert_eq!(
+        resp.headers()
+            .get(reqwest::header::RETRY_AFTER)
+            .and_then(|v| v.to_str().ok()),
+        Some("6"),
+        "Retry-After must survive the proxy"
+    );
+    let body: Value = resp.json().await.unwrap();
+    assert_eq!(body["error"]["code"], "rate_limit_exceeded");
+}
+
+#[tokio::test]
+async fn ties_break_deterministically_by_name() {
+    let neuron_a = common::spawn_mock_neuron().await;
+    let neuron_b = common::spawn_mock_neuron().await;
+    let fleet = two_neuron_fleet(&neuron_a, &neuron_b).await;
+
+    // Equal load on both → stable pick (lowest node name).
+    seed_loaded(&fleet, "node-a", 0, 0).await;
+    seed_loaded(&fleet, "node-b", 0, 0).await;
+
+    let route = cortex_gateway::router::resolve(&fleet, "test-model")
+        .await
+        .expect("loaded");
+    assert_eq!(route.node_name, "node-a", "ties break by name");
+}

crates/cortex-gateway/tests/metering.rs

@@ -0,0 +1,209 @@
+//! Integration tests for per-request token metering (#51).
+//!
+//! Drives authenticated requests through the gateway to a mock neuron that
+//! reports a fixed `usage` object, then asserts the EntitlementProvider's
+//! spend ledger reflects cumulative per-key spend and that reservations
+//! settle to actual (no outstanding reserved tokens once requests complete).
+
+mod common;
+
+use cortex_core::config::{
+    ApiKeyConfig, EntitlementsConfig, EvictionSettings, EvictionStrategy, GatewayConfig,
+    GatewaySettings, NeuronEndpoint,
+};
+use cortex_core::entitlements::{CapWindow, Principal};
+use cortex_core::node::{ModelEntry, ModelStatus};
+use cortex_gateway::state::CortexState;
+use serde_json::json;
+use std::sync::Arc;
+use std::time::Duration;
+use tokio::net::TcpListener;
+
+const ACCOUNT: &str = "acct-meter";
+const KEY_ID: &str = "key-meter";
+const BEARER: &str = "sk-meter";
+
+/// The mock neuron (common::spawn_mock_neuron) reports this fixed usage on
+/// every chat completion.
+const PROMPT_PER_REQ: u64 = 10;
+const COMPLETION_PER_REQ: u64 = 5;
+
+async fn spawn_metered_gateway(neuron_url: &str) -> (Arc<CortexState>, String) {
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        neurons: vec![NeuronEndpoint {
+            name: "mock-node".into(),
+            endpoint: neuron_url.to_string(),
+        }],
+        models_config: "/dev/null".into(),
+        entitlements: EntitlementsConfig {
+            require_auth: true,
+            keys: vec![ApiKeyConfig {
+                key: BEARER.into(),
+                account_id: ACCOUNT.into(),
+                key_id: Some(KEY_ID.into()),
+                hard_cap: Some(1_000_000),
+                window: CapWindow::Balance,
+            }],
+        },
+        upstream: Default::default(),
+    };
+
+    let fleet = Arc::new(CortexState::from_config(&config));
+    {
+        let mut nodes = fleet.nodes.write().await;
+        let node = nodes.get_mut("mock-node").unwrap();
+        node.healthy = true;
+        node.models.insert(
+            "test-model".into(),
+            ModelEntry {
+                id: "test-model".into(),
+                status: ModelStatus::Loaded,
+                last_accessed: None,
+                vram_estimate_mb: Some(8000),
+                capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
+            },
+        );
+    }
+
+    let app = cortex_gateway::build_app(Arc::clone(&fleet));
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+    (fleet, format!("http://{addr}"))
+}
+
+fn principal() -> Principal {
+    Principal {
+        account_id: ACCOUNT.into(),
+        key_id: KEY_ID.into(),
+    }
+}
+
+/// Poll the provider ledger until settled spend reaches `expected` (settle
+/// runs in a spawned task after the response stream finishes) or time out.
+async fn await_spent(fleet: &CortexState, expected: u64) -> u64 {
+    let principal = principal();
+    for _ in 0..100 {
+        let snap = fleet.entitlements.snapshot(&principal).await.unwrap();
+        if snap.spent >= expected {
+            return snap.spent;
+        }
+        tokio::time::sleep(Duration::from_millis(20)).await;
+    }
+    fleet.entitlements.snapshot(&principal).await.unwrap().spent
+}
+
+#[tokio::test]
+async fn cumulative_spend_is_metered_per_key() {
+    let neuron = common::spawn_mock_neuron().await;
+    let (fleet, gateway) = spawn_metered_gateway(&neuron).await;
+    let client = reqwest::Client::new();
+
+    const N: u64 = 3;
+    for _ in 0..N {
+        let resp = client
+            .post(format!("{gateway}/v1/chat/completions"))
+            .bearer_auth(BEARER)
+            .json(&json!({"model": "test-model", "messages": [{"role": "user", "content": "hi"}]}))
+            .send()
+            .await
+            .unwrap();
+        assert_eq!(resp.status(), reqwest::StatusCode::OK);
+        // Drain the body so the response stream finishes and metering settles.
+        let _ = resp.bytes().await.unwrap();
+    }
+
+    let expected = N * (PROMPT_PER_REQ + COMPLETION_PER_REQ);
+    let spent = await_spent(&fleet, expected).await;
+    assert_eq!(
+        spent, expected,
+        "ledger must reflect cumulative per-key spend"
+    );
+
+    // Reservations settled to actual — nothing left outstanding.
+    let snap = fleet.entitlements.snapshot(&principal()).await.unwrap();
+    assert_eq!(snap.reserved, 0, "all reservations must settle/release");
+    assert_eq!(snap.hard_cap, Some(1_000_000));
+}
+
+#[tokio::test]
+async fn anonymous_request_records_no_spend() {
+    // require_auth=false so the unauthenticated request is served, but with
+    // no principal it must not touch any ledger.
+    let neuron = common::spawn_mock_neuron().await;
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        neurons: vec![NeuronEndpoint {
+            name: "mock-node".into(),
+            endpoint: neuron.clone(),
+        }],
+        models_config: "/dev/null".into(),
+        entitlements: EntitlementsConfig::default(),
+        upstream: Default::default(),
+    };
+    let fleet = Arc::new(CortexState::from_config(&config));
+    {
+        let mut nodes = fleet.nodes.write().await;
+        let node = nodes.get_mut("mock-node").unwrap();
+        node.healthy = true;
+        node.models.insert(
+            "test-model".into(),
+            ModelEntry {
+                id: "test-model".into(),
+                status: ModelStatus::Loaded,
+                last_accessed: None,
+                vram_estimate_mb: Some(8000),
+                capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
+            },
+        );
+    }
+    let app = cortex_gateway::build_app(Arc::clone(&fleet));
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+
+    let resp = reqwest::Client::new()
+        .post(format!("http://{addr}/v1/chat/completions"))
+        .json(&json!({"model": "test-model", "messages": [{"role": "user", "content": "hi"}]}))
+        .send()
+        .await
+        .unwrap();
+    assert_eq!(resp.status(), reqwest::StatusCode::OK);
+    let _ = resp.bytes().await.unwrap();
+
+    // An unconfigured principal has a zeroed snapshot — nothing was metered.
+    let snap = fleet
+        .entitlements
+        .snapshot(&Principal {
+            account_id: "nobody".into(),
+            key_id: "nobody".into(),
+        })
+        .await
+        .unwrap();
+    assert_eq!(snap.spent, 0);
+}

crates/cortex-gateway/tests/model_cost.rs

@@ -0,0 +1,132 @@
+//! Issue #68: the `cost` wire contract on `GET /v1/models`.
+//!
+//! `cost` is operator-set pricing sourced from the `models.toml` catalogue
+//! profile (the source of truth today; the marketplace clearing house #59
+//! later — both must read the same value metering/#51 bills against). The
+//! shape is the models.dev/opencode convention: **USD per 1,000,000 tokens,
+//! as JSON numbers**, with optional `cache_read`/`cache_write` tiers. This
+//! test pins:
+//!   - the units/shape (per-million floats, not per-token, not strings);
+//!   - that cache fields flow through when present and are omitted otherwise;
+//!   - the load-bearing **absent vs `0.0`** distinction (#68): a model with
+//!     no catalogue `cost` omits the key entirely (price unknown), distinct
+//!     from an explicit `0.0` (intentionally free).
+//!
+//! Catalogue-only models surface via Pass 1 of `list_models` even with no
+//! feasible neuron, so this is hermetic — no nodes or poller needed.
+
+use cortex_core::config::{
+    EvictionSettings, EvictionStrategy, GatewayConfig, GatewaySettings, NeuronEndpoint,
+};
+use cortex_gateway::state::CortexState;
+use std::sync::Arc;
+use tokio::net::TcpListener;
+
+#[tokio::test]
+async fn v1_models_cost_units_shape_and_absent_vs_zero() {
+    // Three catalogue models exercise the whole contract: a priced model
+    // with cache tiers, an intentionally-free model (explicit 0.0), and an
+    // unpriced model (no `cost` block at all).
+    let models_toml = r#"
+[[models]]
+id = "priced-model"
+harness = "candle"
+cost.input = 0.5
+cost.output = 1.5
+cost.cache_read = 0.05
+cost.cache_write = 0.6
+
+[[models]]
+id = "free-model"
+harness = "candle"
+cost.input = 0.0
+cost.output = 0.0
+
+[[models]]
+id = "unpriced-model"
+harness = "candle"
+"#;
+    let cat_path = std::env::temp_dir().join("cortex_test_issue68_models.toml");
+    std::fs::write(&cat_path, models_toml).unwrap();
+
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        // Never contacted: build_app does not spawn the poller, so the
+        // catalogue alone drives /v1/models.
+        neurons: vec![NeuronEndpoint {
+            name: "mock-node".into(),
+            endpoint: "http://127.0.0.1:1".into(),
+        }],
+        models_config: cat_path.to_string_lossy().into_owned(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
+    };
+
+    let fleet = Arc::new(CortexState::from_config(&config));
+    let app = cortex_gateway::build_app(Arc::clone(&fleet));
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+
+    let body: serde_json::Value = reqwest::Client::new()
+        .get(format!("http://{addr}/v1/models"))
+        .send()
+        .await
+        .unwrap()
+        .json()
+        .await
+        .unwrap();
+
+    let data = body["data"].as_array().expect("data is an array");
+    let entry = |id: &str| {
+        data.iter()
+            .find(|m| m["id"] == id)
+            .unwrap_or_else(|| panic!("{id} present in /v1/models"))
+            .clone()
+    };
+
+    // Priced model: exact values flow through as JSON numbers (USD per 1M
+    // tokens). If anything rescaled by 10⁶ or stringified, these fail.
+    let priced = entry("priced-model");
+    assert_eq!(priced["cost"]["input"], 0.5);
+    assert_eq!(priced["cost"]["output"], 1.5);
+    assert_eq!(priced["cost"]["cache_read"], 0.05);
+    assert_eq!(priced["cost"]["cache_write"], 0.6);
+    assert!(
+        priced["cost"]["input"].is_number(),
+        "cost.input must be a JSON number, not a string"
+    );
+
+    // Intentionally free: cost present, rates explicitly 0.0. Unset cache
+    // tiers are omitted (skip_serializing_if), not emitted as null/0.
+    let free = entry("free-model");
+    assert_eq!(free["cost"]["input"], 0.0);
+    assert_eq!(free["cost"]["output"], 0.0);
+    assert!(
+        free["cost"].get("cache_read").is_none(),
+        "absent cache tiers must be omitted, not null"
+    );
+    assert!(free["cost"].get("cache_write").is_none());
+
+    // Unpriced: the whole `cost` object is omitted — "price unknown",
+    // distinct from the free model's explicit 0.0. This is the #68
+    // distinction opencode needs to avoid showing $0 for a model whose
+    // price simply hasn't been declared.
+    let unpriced = entry("unpriced-model");
+    assert!(
+        unpriced.get("cost").is_none(),
+        "a model with no catalogue cost must omit `cost` entirely, got {:?}",
+        unpriced.get("cost")
+    );
+
+    let _ = std::fs::remove_file(&cat_path);
+}

crates/cortex-gateway/tests/model_limits.rs

@@ -0,0 +1,166 @@
+//! Issue #62 / #67: `GET /v1/models` advertises a per-model serving budget so
+//! an OpenAI-compatible client (opencode's helexa provider) can size and
+//! compact its context without hand-configuration.
+//!
+//! Asserts the composition sources land on the response:
+//!   - `limit` from the neuron's self-derived value (#67) — NOT the catalogue;
+//!     an operator-declared catalogue `limit` is deliberately ignored.
+//!   - `cost` from the catalogue profile (operator-set pricing).
+//!   - `tool_call` / `reasoning` from the neuron's runtime detection (OR-ed in)
+//!
+//! Also asserts the flat, vLLM-convention duplicates (`max_model_len`,
+//! `max_input_tokens`, `max_output_tokens`) mirror `limit` (#78): the
+//! earlier removal of `max_model_len` as "unconsumed" was wrong — Hermes
+//! Agent (and the wider OpenAI client ecosystem) probes those flat keys
+//! and cannot see `limit.context`.
+
+use cortex_core::config::{
+    EvictionSettings, EvictionStrategy, GatewayConfig, GatewaySettings, NeuronEndpoint,
+};
+use cortex_core::harness::ModelLimit;
+use cortex_core::node::{ModelEntry, ModelStatus};
+use cortex_gateway::state::CortexState;
+use std::sync::Arc;
+use tokio::net::TcpListener;
+
+#[tokio::test]
+async fn v1_models_surfaces_limit_cost_and_capability_flags() {
+    // Catalogue declares pricing + an operator `limit` that must be IGNORED
+    // (#67): the neuron's self-derived limit is authoritative.
+    let models_toml = r#"
+[[models]]
+id = "test-model"
+harness = "candle"
+limit.context = 999999
+limit.input = 999999
+limit.output = 999999
+cost.input = 0.0
+cost.output = 0.0
+capabilities = ["text"]
+"#;
+    let cat_path = std::env::temp_dir().join("cortex_test_issue62_models.toml");
+    std::fs::write(&cat_path, models_toml).unwrap();
+
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        neurons: vec![NeuronEndpoint {
+            name: "mock-node".into(),
+            // Never contacted: build_app does not spawn the poller, so the
+            // seeded state below is authoritative for /v1/models.
+            endpoint: "http://127.0.0.1:1".into(),
+        }],
+        models_config: cat_path.to_string_lossy().into_owned(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
+    };
+
+    let fleet = Arc::new(CortexState::from_config(&config));
+
+    // Seed the model as loaded on the node with runtime-detected flags set —
+    // these must OR into the catalogue entry, not be lost.
+    {
+        let mut nodes = fleet.nodes.write().await;
+        let node = nodes.get_mut("mock-node").expect("node exists");
+        node.healthy = true;
+        node.models.insert(
+            "test-model".into(),
+            ModelEntry {
+                id: "test-model".into(),
+                status: ModelStatus::Loaded,
+                last_accessed: None,
+                vram_estimate_mb: Some(8000),
+                capabilities: vec!["text".into()],
+                tool_call: true,
+                reasoning: true,
+                // Neuron's self-derived limit (#67) — the authoritative
+                // source. Distinct from the catalogue's (ignored) values.
+                limit: Some(ModelLimit {
+                    context: 49152,
+                    input: Some(40960),
+                    output: 8192,
+                }),
+            },
+        );
+        // A model with no derivable limit: the flat #78 fields must be
+        // OMITTED (absent-vs-zero is load-bearing), never 0 or a guess.
+        node.models.insert(
+            "no-limit-model".into(),
+            ModelEntry {
+                id: "no-limit-model".into(),
+                status: ModelStatus::Loaded,
+                last_accessed: None,
+                vram_estimate_mb: None,
+                capabilities: vec!["text".into()],
+                tool_call: false,
+                reasoning: false,
+                limit: None,
+            },
+        );
+    }
+
+    let app = cortex_gateway::build_app(Arc::clone(&fleet));
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+
+    let body: serde_json::Value = reqwest::Client::new()
+        .get(format!("http://{addr}/v1/models"))
+        .send()
+        .await
+        .unwrap()
+        .json()
+        .await
+        .unwrap();
+
+    let entry = body["data"]
+        .as_array()
+        .expect("data is an array")
+        .iter()
+        .find(|m| m["id"] == "test-model")
+        .expect("test-model present in /v1/models");
+
+    // `limit` is the neuron's self-derived value (#67), NOT the catalogue's
+    // (which declared 999999 and must be ignored). `cost` still flows from
+    // the catalogue.
+    assert_eq!(entry["limit"]["context"], 49152);
+    assert_eq!(entry["limit"]["input"], 40960);
+    assert_eq!(entry["limit"]["output"], 8192);
+    assert_eq!(entry["cost"]["input"], 0.0);
+    assert_eq!(entry["cost"]["output"], 0.0);
+
+    // Runtime-detected capability flags OR-ed in from the neuron's ModelEntry.
+    assert_eq!(entry["tool_call"], true);
+    assert_eq!(entry["reasoning"], true);
+
+    // Flat ecosystem duplicates (#78) mirror the advertised `limit` so
+    // vLLM-convention probes (Hermes Agent) auto-detect the window.
+    assert_eq!(entry["max_model_len"], 49152);
+    assert_eq!(entry["max_input_tokens"], 40960);
+    assert_eq!(entry["max_output_tokens"], 8192);
+
+    // No limit → flat fields omitted entirely, never 0 or a guess.
+    let unknown = body["data"]
+        .as_array()
+        .unwrap()
+        .iter()
+        .find(|m| m["id"] == "no-limit-model")
+        .expect("no-limit-model present in /v1/models");
+    assert!(unknown.get("limit").is_none());
+    for key in ["max_model_len", "max_input_tokens", "max_output_tokens"] {
+        assert!(
+            unknown.get(key).is_none(),
+            "{key} must be omitted when the window is unknown"
+        );
+    }
+
+    let _ = std::fs::remove_file(&cat_path);
+}

crates/cortex-gateway/tests/poller.rs

@@ -31,6 +31,8 @@ async fn test_poller_discovers_models() {
             endpoint: mock_url,
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -82,6 +84,8 @@ async fn test_poller_updates_gateway_models_endpoint() {
             endpoint: mock_url,
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -153,6 +157,8 @@ async fn test_models_endpoint_unions_capabilities_across_nodes() {
             },
         ],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -215,6 +221,8 @@ async fn test_poller_marks_unreachable_node_unhealthy() {
             endpoint: "http://127.0.0.1:1".into(),
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -224,10 +232,26 @@ async fn test_poller_marks_unreachable_node_unhealthy() {
         nodes.get_mut("dead-node").unwrap().healthy = true;
     }
 
+    // Debounce (#53 follow-up): a single missed poll must NOT evict a
+    // previously-healthy node — a busy neuron briefly slow to answer
+    // shouldn't yank its models out of routing.
     cortex_gateway::poller::poll_once(&fleet).await;
+    assert!(
+        fleet.nodes.read().await.get("dead-node").unwrap().healthy,
+        "one failed poll should not mark a healthy node unhealthy"
+    );
 
-    let nodes = fleet.nodes.read().await;
-    assert!(!nodes.get("dead-node").unwrap().healthy);
+    // It flips unhealthy only after POLL_FAILURE_THRESHOLD (3) consecutive
+    // failures.
+    cortex_gateway::poller::poll_once(&fleet).await;
+    cortex_gateway::poller::poll_once(&fleet).await;
+    assert!(
+        !fleet.nodes.read().await.get("dead-node").unwrap().healthy,
+        "three consecutive failed polls should mark the node unhealthy"
+    );
+
+    // A subsequent successful poll would reset the counter and restore
+    // health; covered implicitly by the discovery tests above.
 }
 
 #[tokio::test]
@@ -252,6 +276,8 @@ async fn test_poller_removes_stale_models() {
             endpoint: mock_url,
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -282,6 +308,8 @@ async fn test_poller_removes_stale_models() {
             endpoint: new_mock_url,
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet2 = Arc::new(CortexState::from_config(&config2));
@@ -298,6 +326,9 @@ async fn test_poller_removes_stale_models() {
                 last_accessed: None,
                 vram_estimate_mb: None,
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
         node.models.insert(
@@ -308,6 +339,9 @@ async fn test_poller_removes_stale_models() {
                 last_accessed: None,
                 vram_estimate_mb: None,
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
     }
@@ -357,6 +391,8 @@ async fn test_poller_captures_activation_from_health() {
             endpoint: mock_url,
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));
@@ -401,6 +437,8 @@ async fn test_poller_parses_recovering_status() {
             endpoint: mock_url,
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
 
     let fleet = Arc::new(CortexState::from_config(&config));

crates/cortex-gateway/tests/prompt_prevalidation.rs

@@ -0,0 +1,175 @@
+//! Fail-fast prompt pre-validation + advisory client hints (#56).
+//!
+//! cortex refuses a prompt that already exceeds the model's advertised
+//! context window before dispatching to neuron — the same #60
+//! `context_length_exceeded` envelope neuron would emit, just earlier — and
+//! attaches an advisory `X-Helexa-Advice` header for fingerprinted clients.
+
+use axum::Json;
+use axum::extract::Path;
+use axum::routing::{get, post};
+use cortex_core::config::{
+    EvictionSettings, EvictionStrategy, GatewayConfig, GatewaySettings, NeuronEndpoint,
+};
+use cortex_core::harness::ModelLimit;
+use cortex_core::node::{ModelEntry, ModelStatus};
+use cortex_gateway::state::CortexState;
+use serde_json::{Value, json};
+use std::sync::Arc;
+use std::sync::atomic::{AtomicU64, Ordering};
+use tokio::net::TcpListener;
+
+/// Mock neuron with a hit counter, so a test can prove a request was (or
+/// wasn't) dispatched past the gateway's pre-validation.
+async fn spawn_counting_neuron() -> (String, Arc<AtomicU64>) {
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    let base_url = format!("http://{addr}");
+    let inference_url = base_url.clone();
+    let hits = Arc::new(AtomicU64::new(0));
+    let sink = Arc::clone(&hits);
+    let app = axum::Router::new()
+        .route(
+            "/models/{model_id}/endpoint",
+            get(move |Path(_): Path<String>| {
+                let url = inference_url.clone();
+                async move { Json(json!({ "url": url })) }
+            }),
+        )
+        .route(
+            "/v1/chat/completions",
+            post(move || {
+                let sink = Arc::clone(&sink);
+                async move {
+                    sink.fetch_add(1, Ordering::SeqCst);
+                    Json(json!({
+                        "id": "c", "object": "chat.completion", "created": 1_700_000_000_u64,
+                        "model": "test-model",
+                        "choices": [{"index": 0, "message": {"role": "assistant", "content": "ok"}, "finish_reason": "stop"}],
+                        "usage": {"prompt_tokens": 3, "completion_tokens": 1, "total_tokens": 4}
+                    }))
+                }
+            }),
+        );
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+    (base_url, hits)
+}
+
+/// Gateway over one neuron with `test-model` loaded and a tiny advertised
+/// context window (so a modest prompt overflows it).
+async fn spawn_gateway(neuron: &str, context: usize) -> String {
+    let config = GatewayConfig {
+        gateway: GatewaySettings {
+            listen: "127.0.0.1:0".into(),
+            metrics_listen: "127.0.0.1:0".into(),
+        },
+        eviction: EvictionSettings {
+            strategy: EvictionStrategy::Lru,
+            defrag_after_cycles: 0,
+        },
+        neurons: vec![NeuronEndpoint {
+            name: "mock-node".into(),
+            endpoint: neuron.to_string(),
+        }],
+        models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
+    };
+    let fleet = Arc::new(CortexState::from_config(&config));
+    {
+        let mut nodes = fleet.nodes.write().await;
+        let n = nodes.get_mut("mock-node").unwrap();
+        n.healthy = true;
+        n.models.insert(
+            "test-model".into(),
+            ModelEntry {
+                id: "test-model".into(),
+                status: ModelStatus::Loaded,
+                last_accessed: None,
+                vram_estimate_mb: Some(8000),
+                capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: Some(ModelLimit {
+                    context,
+                    input: None,
+                    output: 16,
+                }),
+            },
+        );
+    }
+    let app = cortex_gateway::build_app(Arc::clone(&fleet));
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+    format!("http://{addr}")
+}
+
+#[tokio::test]
+async fn over_long_prompt_is_rejected_before_dispatch() {
+    let (neuron, hits) = spawn_counting_neuron().await;
+    let gateway = spawn_gateway(&neuron, 50).await; // tiny 50-token window
+
+    // ~1200 chars → ~300 est tokens, well over 50.
+    let big = "word ".repeat(240);
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .header("user-agent", "litellm/1.0")
+        .json(&json!({"model": "test-model", "messages": [{"role": "user", "content": big}]}))
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::BAD_REQUEST);
+    // Advisory hint for the fingerprinted client (header only, never body).
+    assert!(
+        resp.headers().get("x-helexa-advice").is_some(),
+        "litellm should get advice"
+    );
+    let body: Value = resp.json().await.unwrap();
+    assert_eq!(body["error"]["code"], "context_length_exceeded");
+    assert_eq!(body["error"]["max"], 50);
+    // Refused at the edge — neuron never saw it.
+    assert_eq!(hits.load(Ordering::SeqCst), 0);
+}
+
+#[tokio::test]
+async fn within_context_passes_through() {
+    let (neuron, hits) = spawn_counting_neuron().await;
+    let gateway = spawn_gateway(&neuron, 4096).await;
+
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        .json(&json!({"model": "test-model", "messages": [{"role": "user", "content": "hi"}]}))
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::OK);
+    let _ = resp.bytes().await.unwrap();
+    assert_eq!(hits.load(Ordering::SeqCst), 1, "served by neuron");
+}
+
+#[tokio::test]
+async fn unknown_client_gets_no_advice_header() {
+    let (neuron, _hits) = spawn_counting_neuron().await;
+    let gateway = spawn_gateway(&neuron, 50).await;
+
+    let big = "word ".repeat(240);
+    let resp = reqwest::Client::new()
+        .post(format!("{gateway}/v1/chat/completions"))
+        // no/unknown User-Agent → no advice, but still a clean 400
+        .json(&json!({"model": "test-model", "messages": [{"role": "user", "content": big}]}))
+        .send()
+        .await
+        .unwrap();
+
+    assert_eq!(resp.status(), reqwest::StatusCode::BAD_REQUEST);
+    assert!(resp.headers().get("x-helexa-advice").is_none());
+    let body: Value = resp.json().await.unwrap();
+    assert_eq!(body["error"]["code"], "context_length_exceeded");
+}

crates/cortex-gateway/tests/proxy_basic.rs

@@ -117,6 +117,8 @@ async fn test_no_healthy_nodes() {
             endpoint: "http://127.0.0.1:1".into(),
         }],
         models_config: "/dev/null".into(),
+        entitlements: Default::default(),
+        upstream: Default::default(),
     };
     let fleet = std::sync::Arc::new(cortex_gateway::state::CortexState::from_config(&config));
 
@@ -139,7 +141,7 @@ async fn test_no_healthy_nodes() {
         .await
         .expect("request should succeed");
 
-    assert_eq!(resp.status(), 404);
+    assert_eq!(resp.status(), 503);
 
     let body: serde_json::Value = resp.json().await.unwrap();
     assert!(
@@ -192,6 +194,9 @@ async fn test_recovering_model_returns_503_and_stays_listed() {
                 last_accessed: None,
                 vram_estimate_mb: Some(8000),
                 capabilities: Vec::new(),
+                tool_call: false,
+                reasoning: false,
+                limit: None,
             },
         );
     }

crates/cortex-gateway/tests/upstream_chain.rs

@@ -0,0 +1,106 @@
+//! B3: the chained entitlement provider (local → upstream) and fail-closed
+//! semantics, exercised against a mock helexa-upstream `/authz/v1`.
+
+use axum::{Json, Router, routing::post};
+use cortex_core::config::{ApiKeyConfig, EntitlementsConfig, UpstreamClientConfig};
+use cortex_core::entitlements::{AuthError, EntitlementProvider};
+use cortex_gateway::entitlements_chain::ChainedEntitlementProvider;
+use cortex_gateway::entitlements_local::LocalEntitlementProvider;
+use cortex_gateway::entitlements_upstream::UpstreamEntitlementProvider;
+use serde_json::{Value, json};
+use tokio::net::TcpListener;
+
+/// Mock upstream: `mesh-key` resolves to a mesh account; anything else 401.
+/// reserve always grants reservation 1.
+async fn spawn_mock_upstream() -> String {
+    async fn resolve(Json(body): Json<Value>) -> axum::response::Response {
+        use axum::response::IntoResponse;
+        if body["api_key"] == "mesh-key" {
+            Json(json!({"principal": {"account_id": "mesh-acct", "key_id": "mesh-key-1"}}))
+                .into_response()
+        } else {
+            (
+                axum::http::StatusCode::UNAUTHORIZED,
+                Json(json!({"error": {"code": "invalid_api_key"}})),
+            )
+                .into_response()
+        }
+    }
+    async fn reserve() -> Json<Value> {
+        Json(json!({ "reservation_id": 1 }))
+    }
+    let app = Router::new()
+        .route("/authz/v1/resolve", post(resolve))
+        .route("/authz/v1/reserve", post(reserve));
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+    format!("http://{addr}")
+}
+
+fn local_with_key() -> LocalEntitlementProvider {
+    let cfg = EntitlementsConfig {
+        require_auth: false,
+        keys: vec![ApiKeyConfig {
+            key: "local-key".into(),
+            account_id: "op".into(),
+            key_id: None,
+            hard_cap: None,
+            window: Default::default(),
+        }],
+    };
+    LocalEntitlementProvider::from_config(&cfg)
+}
+
+fn chain(local: LocalEntitlementProvider, url: &str) -> ChainedEntitlementProvider {
+    let upstream = UpstreamEntitlementProvider::new(&UpstreamClientConfig {
+        enabled: true,
+        url: url.to_string(),
+        bearer: "client-secret".into(),
+        timeout_secs: 5,
+        served_usage_report_interval_secs: 60,
+    });
+    ChainedEntitlementProvider::new(local, upstream)
+}
+
+#[tokio::test]
+async fn local_key_resolves_locally() {
+    let url = spawn_mock_upstream().await;
+    let c = chain(local_with_key(), &url);
+    let p = c.resolve("local-key").await.expect("local resolves");
+    assert_eq!(p.account_id, "op");
+}
+
+#[tokio::test]
+async fn unknown_key_falls_through_to_upstream() {
+    let url = spawn_mock_upstream().await;
+    let c = chain(local_with_key(), &url);
+    let p = c.resolve("mesh-key").await.expect("upstream resolves");
+    assert_eq!(p.account_id, "mesh-acct");
+    assert_eq!(p.key_id, "mesh-key-1");
+}
+
+#[tokio::test]
+async fn unknown_everywhere_is_invalid_key() {
+    let url = spawn_mock_upstream().await;
+    let c = chain(local_with_key(), &url);
+    match c.resolve("nope").await {
+        Err(AuthError::InvalidKey) => {}
+        other => panic!("expected InvalidKey, got {other:?}"),
+    }
+}
+
+#[tokio::test]
+async fn upstream_unreachable_fails_closed_as_unavailable() {
+    // No mock — point at a dead port. A locally-unknown key must surface
+    // Unavailable (→ 503), never InvalidKey (→ 401).
+    let c = chain(local_with_key(), "http://127.0.0.1:1");
+    match c.resolve("some-mesh-key").await {
+        Err(AuthError::Unavailable { retry_after_secs }) => assert!(retry_after_secs > 0),
+        other => panic!("expected Unavailable, got {other:?}"),
+    }
+    // A local key still resolves without touching upstream.
+    assert_eq!(c.resolve("local-key").await.unwrap().account_id, "op");
+}

crates/helexa-bench/Cargo.toml

@@ -27,12 +27,15 @@ futures = { workspace = true }
 tokio-stream = { workspace = true }
 eventsource-stream = { workspace = true }
 
+# read-only JSON API (api.rs)
+axum = { workspace = true }
+tower-http = { workspace = true }
+
 # SQLite system-of-record. `bundled` compiles SQLite from source so the
 # binary has no libsqlite3 runtime dependency — matches the project's
 # single-static-binary packaging.
 rusqlite = { version = "0.32", features = ["bundled"] }
 
 [dev-dependencies]
-axum = { workspace = true }
 # Jail (isolated cwd + env) for config tests.
 figment = { workspace = true, features = ["test"] }

crates/helexa-bench/src/api.rs

@@ -0,0 +1,146 @@
+//! Read-only JSON API over the bench SQLite store.
+//!
+//! Consumed by the `bench/` visualisation app and for programmatic
+//! access. Served by the `run` daemon (alongside the sweep loop) and by
+//! the standalone `serve` subcommand. CORS is permissive because the UI
+//! is hosted separately (different origin); the API is internal-only
+//! (WireGuard + firewalld) and read-only, so this predates the auth epic.
+
+use crate::store::{RunFilter, Store};
+use anyhow::Result;
+use axum::Router;
+use axum::extract::{Query, State};
+use axum::http::StatusCode;
+use axum::response::Json;
+use axum::routing::get;
+use serde::Deserialize;
+use serde_json::json;
+use std::sync::Arc;
+use tokio::sync::Mutex;
+use tower_http::cors::CorsLayer;
+
+/// Shared API state: a dedicated read connection to the store, guarded
+/// (rusqlite `Connection` isn't `Sync`). Separate from the sweep's
+/// writer connection — WAL lets them run concurrently.
+pub type ApiState = Arc<Mutex<Store>>;
+
+/// Open an API state over the store at `db_path`.
+pub fn open_state(db_path: &str) -> Result<ApiState> {
+    Ok(Arc::new(Mutex::new(Store::open(db_path)?)))
+}
+
+/// Build the API router.
+pub fn api_routes(state: ApiState) -> Router {
+    Router::new()
+        .route("/api/health", get(health))
+        .route("/api/dimensions", get(dimensions))
+        .route("/api/summary", get(summary))
+        .route("/api/scaling", get(scaling))
+        .route("/api/swap", get(swap))
+        .route("/api/capability", get(capability))
+        .route("/api/series", get(series))
+        .route("/api/runs", get(runs))
+        .layer(CorsLayer::permissive())
+        .with_state(state)
+}
+
+/// Bind `listen` and serve the API until the process exits.
+pub async fn serve(listen: &str, state: ApiState) -> Result<()> {
+    let listener = tokio::net::TcpListener::bind(listen).await?;
+    tracing::info!(%listen, "bench API listening");
+    axum::serve(listener, api_routes(state)).await?;
+    Ok(())
+}
+
+type ApiError = (StatusCode, String);
+
+fn err500(e: anyhow::Error) -> ApiError {
+    (StatusCode::INTERNAL_SERVER_ERROR, format!("{e:#}"))
+}
+
+async fn health(State(s): State<ApiState>) -> Result<Json<serde_json::Value>, ApiError> {
+    let store = s.lock().await;
+    let count = store.run_count().map_err(err500)?;
+    Ok(Json(json!({ "status": "ok", "run_count": count })))
+}
+
+async fn dimensions(State(s): State<ApiState>) -> Result<Json<crate::store::Dimensions>, ApiError> {
+    let store = s.lock().await;
+    store.dimensions().map(Json).map_err(err500)
+}
+
+async fn summary(
+    State(s): State<ApiState>,
+) -> Result<Json<Vec<crate::store::ReportRow>>, ApiError> {
+    let store = s.lock().await;
+    store.summary().map(Json).map_err(err500)
+}
+
+/// Context-length scaling curves per (target, model) — prefill & decode
+/// tok/s vs context, with decode-flatness (#88).
+async fn scaling(
+    State(s): State<ApiState>,
+) -> Result<Json<Vec<crate::store::ScalingCurve>>, ApiError> {
+    let store = s.lock().await;
+    store.scaling().map(Json).map_err(err500)
+}
+
+/// Cold-load / model-swap costs per (target, model) — reload latency + cold
+/// first-request (#90).
+async fn swap(State(s): State<ApiState>) -> Result<Json<Vec<crate::store::SwapCost>>, ApiError> {
+    let store = s.lock().await;
+    store.swap_costs().map(Json).map_err(err500)
+}
+
+/// Capability-probe runs — stored artifacts + quality scores (#91).
+async fn capability(
+    State(s): State<ApiState>,
+) -> Result<Json<Vec<crate::store::CapabilityRun>>, ApiError> {
+    let store = s.lock().await;
+    store.capability_runs(false).map(Json).map_err(err500)
+}
+
+#[derive(Debug, Deserialize)]
+struct SeriesQuery {
+    /// Optional — when omitted the store resolves the host serving this model.
+    host: Option<String>,
+    model: String,
+    scenario: String,
+}
+
+async fn series(
+    State(s): State<ApiState>,
+    Query(q): Query<SeriesQuery>,
+) -> Result<Json<Vec<crate::store::SeriesPoint>>, ApiError> {
+    let store = s.lock().await;
+    store
+        .series(q.host.as_deref(), &q.model, &q.scenario)
+        .map(Json)
+        .map_err(err500)
+}
+
+#[derive(Debug, Deserialize)]
+struct RunsQuery {
+    host: Option<String>,
+    model: Option<String>,
+    scenario: Option<String>,
+    sha: Option<String>,
+    ok: Option<bool>,
+    limit: Option<u32>,
+}
+
+async fn runs(
+    State(s): State<ApiState>,
+    Query(q): Query<RunsQuery>,
+) -> Result<Json<Vec<crate::store::RunRow>>, ApiError> {
+    let filter = RunFilter {
+        host: q.host,
+        model: q.model,
+        scenario: q.scenario,
+        sha: q.sha,
+        ok: q.ok,
+        limit: q.limit,
+    };
+    let store = s.lock().await;
+    store.runs(&filter).map(Json).map_err(err500)
+}

crates/helexa-bench/src/client.rs

@@ -3,10 +3,10 @@
 //! `openai` targets use the OpenAI-compatible surface (preliminary).
 
 use crate::config::{TargetConfig, TargetKind};
-use anyhow::{Context, Result};
+use anyhow::{Context, Result, anyhow};
 use cortex_core::build_info::BuildInfo;
-use cortex_core::discovery::DiscoveryResponse;
-use cortex_core::harness::ModelInfo;
+use cortex_core::discovery::{DiscoveryResponse, HealthResponse};
+use cortex_core::harness::{ModelInfo, ModelSpec};
 use cortex_core::openai::ModelsResponse;
 use std::time::Duration;
 
@@ -94,6 +94,84 @@ impl TargetClient {
         Ok(Some(disco))
     }
 
+    /// Runtime device health (neuron only): per-GPU VRAM used/free,
+    /// utilization, and temperature from `GET /health`. Bench samples this
+    /// around each measured run to record VRAM high-water + GPU telemetry
+    /// (#87). Returns `Ok(None)` for non-neuron targets; a soft `Ok(None)`
+    /// (not an error) on transport failure so a flaky `/health` never fails
+    /// a measurement.
+    pub async fn fetch_health(&self, target: &TargetConfig) -> Result<Option<HealthResponse>> {
+        if target.kind != TargetKind::Neuron {
+            return Ok(None);
+        }
+        let base = target.endpoint.trim_end_matches('/');
+        let health = self
+            .http
+            .get(format!("{base}/health"))
+            .timeout(META_TIMEOUT)
+            .send()
+            .await
+            .context("GET /health")?
+            .error_for_status()
+            .context("GET /health status")?
+            .json::<HealthResponse>()
+            .await
+            .context("decoding /health")?;
+        Ok(Some(health))
+    }
+
+    /// Unload a model (neuron only): `POST /models/unload {model_id}`.
+    /// Used by the deliberate swap-cost measurement (#90), never the sweep.
+    pub async fn unload_model(&self, target: &TargetConfig, model_id: &str) -> Result<()> {
+        let base = target.endpoint.trim_end_matches('/');
+        self.http
+            .post(format!("{base}/models/unload"))
+            .json(&serde_json::json!({ "model_id": model_id }))
+            .send()
+            .await
+            .context("POST /models/unload")?
+            .error_for_status()
+            .context("POST /models/unload status")?;
+        Ok(())
+    }
+
+    /// Load a model from a spec (neuron only): `POST /models/load`. neuron
+    /// returns synchronously once loaded, so the call duration is the reload
+    /// cost the swap-cost measurement records (#90).
+    pub async fn load_model(&self, target: &TargetConfig, spec: &ModelSpec) -> Result<()> {
+        let base = target.endpoint.trim_end_matches('/');
+        self.http
+            .post(format!("{base}/models/load"))
+            .json(spec)
+            // A cold load can take tens of seconds; use the full request
+            // timeout rather than the short metadata one.
+            .send()
+            .await
+            .context("POST /models/load")?
+            .error_for_status()
+            .context("POST /models/load status")?;
+        Ok(())
+    }
+
+    /// Reconstruct a reload [`ModelSpec`] from a model's `/models` entry.
+    /// Tensor-parallel is inferred from the device count; `quant` is left
+    /// `None` for neuron to resolve from the catalogue / its prior load.
+    pub fn spec_from_info(info: &ModelInfo) -> Result<ModelSpec> {
+        if info.devices.is_empty() {
+            return Err(anyhow!(
+                "model '{}' reports no devices; cannot reconstruct a load spec",
+                info.id
+            ));
+        }
+        Ok(ModelSpec {
+            model_id: info.id.clone(),
+            harness: info.harness.clone(),
+            quant: None,
+            tensor_parallel: (info.devices.len() > 1).then_some(info.devices.len() as u32),
+            devices: Some(info.devices.clone()),
+        })
+    }
+
     /// Warm models — those ready to serve without a cold load.
     ///
     /// Neuron: `GET /models` filtered to `status == "loaded"` (skips
@@ -151,6 +229,10 @@ impl TargetClient {
                         devices: Vec::new(),
                         vram_used_mb: None,
                         capabilities: Vec::new(),
+                        limit: None,
+                        cost: None,
+                        tool_call: false,
+                        reasoning: false,
                     })
                     .collect())
             }

crates/helexa-bench/src/config.rs

@@ -16,11 +16,35 @@ pub struct BenchConfig {
     pub bench: BenchSettings,
     #[serde(default)]
     pub scenarios: ScenarioConfig,
+    /// Read-only JSON API (consumed by the bench UI + programmatic access).
+    #[serde(default)]
+    pub api: ApiSettings,
     /// Endpoints to benchmark. At least one is required for `run`/`once`.
     #[serde(default)]
     pub targets: Vec<TargetConfig>,
 }
 
+/// The read-only HTTP API the `run` daemon (and the `serve` subcommand)
+/// exposes over the SQLite store.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ApiSettings {
+    /// Whether to bind the API at all.
+    #[serde(default = "default_api_enabled")]
+    pub enabled: bool,
+    /// Listen address for the API.
+    #[serde(default = "default_api_listen")]
+    pub listen: String,
+}
+
+impl Default for ApiSettings {
+    fn default() -> Self {
+        ApiSettings {
+            enabled: default_api_enabled(),
+            listen: default_api_listen(),
+        }
+    }
+}
+
 /// Loop/timing knobs.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct BenchSettings {
@@ -80,6 +104,37 @@ pub struct ScenarioConfig {
     /// Max generated tokens per request.
     #[serde(default = "default_max_tokens")]
     pub max_tokens: u64,
+    /// Concurrency levels (#89): one `concurrency:<n>` scenario per entry,
+    /// each firing N simultaneous streams. Defaults to empty (opt-in) — a
+    /// burst puts real load on a serving fleet, so operators enable it
+    /// deliberately, e.g. `concurrency_levels = [2, 4, 8]`.
+    #[serde(default)]
+    pub concurrency_levels: Vec<u32>,
+    /// Approximate prompt size (tokens) used by the concurrency scenarios.
+    #[serde(default = "default_concurrency_prompt_tokens")]
+    pub concurrency_prompt_tokens: u32,
+    /// Capability probes (#91): one `capability:<name>` scenario per entry,
+    /// each running a fixed prompt and storing the full output artifact for
+    /// quality scoring (manual now, LLM-judge later). Defaults to empty
+    /// (opt-in) — these generate long outputs and exist to compare reasoning
+    /// quality across models, not to run on every sweep by default.
+    #[serde(default)]
+    pub capability_probes: Vec<CapabilityProbe>,
+}
+
+/// One capability probe: a named prompt whose output is stored and scored
+/// for quality (#91). The probe is deterministic (temperature 0) so the
+/// same model+build produces a stable artifact to score.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct CapabilityProbe {
+    /// Stable id fragment — the scenario becomes `capability:<name>`.
+    pub name: String,
+    /// The prompt sent verbatim as the user message.
+    pub prompt: String,
+    /// Generation budget for the probe (planning answers are long; the
+    /// default 256 is too small). Defaults to 2048.
+    #[serde(default = "default_capability_max_tokens")]
+    pub max_tokens: u64,
 }
 
 impl Default for ScenarioConfig {
@@ -87,6 +142,9 @@ impl Default for ScenarioConfig {
         ScenarioConfig {
             prompt_sizes: default_prompt_sizes(),
             max_tokens: default_max_tokens(),
+            concurrency_levels: Vec::new(),
+            concurrency_prompt_tokens: default_concurrency_prompt_tokens(),
+            capability_probes: Vec::new(),
         }
     }
 }
@@ -151,12 +209,24 @@ fn default_timeout() -> u64 {
 fn default_db_path() -> String {
     "/var/lib/helexa-bench/bench.sqlite".to_string()
 }
+fn default_api_enabled() -> bool {
+    true
+}
+fn default_api_listen() -> String {
+    "0.0.0.0:13132".to_string()
+}
 fn default_prompt_sizes() -> Vec<u32> {
     vec![128, 4096]
 }
 fn default_max_tokens() -> u64 {
     256
 }
+fn default_concurrency_prompt_tokens() -> u32 {
+    512
+}
+fn default_capability_max_tokens() -> u64 {
+    2048
+}
 
 #[cfg(test)]
 // Jail's closure must return figment::Result; the large-Err type is

crates/helexa-bench/src/lib.rs

@@ -4,6 +4,7 @@
 //! full build/version provenance into SQLite so improvements can be
 //! tracked automatically across neuron implementation updates.
 
+pub mod api;
 pub mod client;
 pub mod config;
 pub mod report;

crates/helexa-bench/src/main.rs

@@ -10,6 +10,7 @@
 
 use anyhow::{Context, Result};
 use clap::{Parser, Subcommand};
+use helexa_bench::api;
 use helexa_bench::config::BenchConfig;
 use helexa_bench::report;
 use helexa_bench::store::Store;
@@ -37,6 +38,38 @@ enum Command {
         #[arg(short, long, default_value = "helexa-bench.toml")]
         config: String,
     },
+    /// Serve the read-only JSON API only (no sweeping).
+    Serve {
+        #[arg(short, long, default_value = "helexa-bench.toml")]
+        config: String,
+    },
+    /// Measure cold-load / model-swap cost (#90): for each neuron target's
+    /// warm models, unload → time reload → time a cold first request, recorded
+    /// under scenario "swap". DELIBERATE — takes each model offline for its
+    /// reload, so run it in a maintenance window, not against live traffic.
+    SwapCost {
+        #[arg(short, long, default_value = "helexa-bench.toml")]
+        config: String,
+    },
+    /// Attach a quality score to a capability-probe run (#91). Find run ids
+    /// with `report --capability`. `--scorer` records who scored it
+    /// (defaults to "manual"); a future LLM-judge would set e.g. "llm:…".
+    Score {
+        #[arg(short, long, default_value = "helexa-bench.toml")]
+        config: String,
+        /// Override the SQLite path (skips reading the config file).
+        #[arg(long)]
+        db: Option<String>,
+        /// The run id to score.
+        #[arg(long)]
+        id: i64,
+        /// The quality score to attach (scale is the operator's rubric).
+        #[arg(long)]
+        score: f64,
+        /// Who/what produced the score.
+        #[arg(long, default_value = "manual")]
+        scorer: String,
+    },
     /// Render recorded results. Uses `--db` if given, else the db_path
     /// from `--config`.
     Report {
@@ -48,6 +81,19 @@ enum Command {
         /// Output format.
         #[arg(long, default_value = "md")]
         format: Format,
+        /// Render the context-length scaling view (prefill & decode tok/s
+        /// vs context per model, with decode-flatness) instead of the flat
+        /// results table (#88).
+        #[arg(long)]
+        scaling: bool,
+        /// Render the cold-load / model-swap cost view (#90) instead of the
+        /// flat results table.
+        #[arg(long)]
+        swap: bool,
+        /// Render the capability-probe view (#91): stored artifacts + quality
+        /// scores, with per-model median.
+        #[arg(long)]
+        capability: bool,
     },
 }
 
@@ -77,10 +123,31 @@ async fn run(cli: Cli) -> Result<()> {
         Command::Run { config } => {
             let cfg = load_config(&config)?;
             require_targets(&cfg)?;
+            // Bind the read API alongside the sweep loop (one bob service
+            // does both). Its own store connection; WAL keeps the sweep
+            // writer and the API readers from blocking each other.
+            if cfg.api.enabled {
+                let state = api::open_state(&cfg.bench.db_path)?;
+                let listen = cfg.api.listen.clone();
+                tokio::spawn(async move {
+                    if let Err(e) = api::serve(&listen, state).await {
+                        tracing::error!(error = %format!("{e:#}"), "bench API server exited");
+                    }
+                });
+            }
             let sweeper = Sweeper::new(cfg)?;
             tracing::info!("helexa-bench started; entering continuous sweep loop");
             sweeper.run_forever().await
         }
+        Command::Serve { config } => {
+            let cfg = load_config(&config)?;
+            if !cfg.api.enabled {
+                anyhow::bail!("[api] enabled = false — nothing to serve");
+            }
+            let state = api::open_state(&cfg.bench.db_path)?;
+            tracing::info!("helexa-bench serving API only");
+            api::serve(&cfg.api.listen, state).await
+        }
         Command::Once { config } => {
             let cfg = load_config(&config)?;
             require_targets(&cfg)?;
@@ -95,16 +162,79 @@ async fn run(cli: Cli) -> Result<()> {
             );
             Ok(())
         }
-        Command::Report { config, db, format } => {
+        Command::SwapCost { config } => {
+            let cfg = load_config(&config)?;
+            require_targets(&cfg)?;
+            let sweeper = Sweeper::new(cfg)?;
+            tracing::warn!(
+                "swap-cost: cycling each warm model (unload → reload → cold request); models go offline during reload"
+            );
+            let summary = sweeper.swap_cost_once().await?;
+            tracing::info!(
+                measured = summary.measured,
+                failed = summary.failed,
+                unreachable = summary.targets_unreachable,
+                "swap-cost measurement complete"
+            );
+            Ok(())
+        }
+        Command::Score {
+            config,
+            db,
+            id,
+            score,
+            scorer,
+        } => {
             let db_path = match db {
                 Some(p) => p,
                 None => load_config(&config)?.bench.db_path,
             };
             let store = Store::open(&db_path)?;
+            match store.set_score(id, score, &scorer)? {
+                0 => anyhow::bail!("no run with id {id}"),
+                _ => {
+                    println!("scored run {id}: {score} ({scorer})");
+                    Ok(())
+                }
+            }
+        }
+        Command::Report {
+            config,
+            db,
+            format,
+            scaling,
+            swap,
+            capability,
+        } => {
+            let db_path = match db {
+                Some(p) => p,
+                None => load_config(&config)?.bench.db_path,
+            };
+            let store = Store::open(&db_path)?;
+            let rendered = if capability {
+                let runs = store.capability_runs(false)?;
+                match format {
+                    Format::Md => report::render_capability_markdown(&runs),
+                    Format::Json => report::render_capability_json(&runs)?,
+                }
+            } else if swap {
+                let costs = store.swap_costs()?;
+                match format {
+                    Format::Md => report::render_swap_markdown(&costs),
+                    Format::Json => report::render_swap_json(&costs)?,
+                }
+            } else if scaling {
+                let curves = store.scaling()?;
+                match format {
+                    Format::Md => report::render_scaling_markdown(&curves),
+                    Format::Json => report::render_scaling_json(&curves)?,
+                }
+            } else {
                 let rows = store.report_rows()?;
-            let rendered = match format {
+                match format {
                     Format::Md => report::render_markdown(&rows),
                     Format::Json => report::render_json(&rows)?,
+                }
             };
             println!("{rendered}");
             Ok(())

crates/helexa-bench/src/report.rs

@@ -3,28 +3,36 @@
 //! doc: engine, model, prompt tok, TTFT (s), decode tok/s, total (s),
 //! plus the build SHA each cell was measured against.
 
-use crate::store::ReportRow;
+use crate::store::{CapabilityRun, ReportRow, ScalingCurve, SwapCost};
 use anyhow::Result;
 
 pub fn render_markdown(rows: &[ReportRow]) -> String {
     let mut out = String::new();
     out.push_str(
-        "| engine | model | prompt tok | TTFT (s) | decode tok/s | total (s) | build | n |\n",
+        "| engine | model | prompt tok | prefill tok/s | TTFT (s) | TTFT p95 | \
+         decode tok/s | total (s) | total p95 | VRAM (GB) | conc | queue ms | rej | build | n |\n",
     );
-    out.push_str("|---|---|---:|---:|---:|---:|---|---:|\n");
+    out.push_str("|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---|---:|\n");
     for r in rows {
         let ptok = r
             .prompt_tokens
             .map(|t| t.to_string())
             .unwrap_or_else(|| format!("~{}", r.prompt_size_approx));
         out.push_str(&format!(
-            "| {} | {} | {} | {} | {} | {} | `{}` | {} |\n",
+            "| {} | {} | {} | {} | {} | {} | {} | {} | {} | {} | {} | {} | {} | `{}` | {} |\n",
             r.target_name,
             r.model_id,
             ptok,
+            fmt_opt(r.prefill_tps_median, 1),
             fmt_opt(r.ttft_s_median, 3),
+            fmt_opt(r.ttft_s_p95, 3),
             fmt_opt(r.decode_tps_median, 1),
             fmt_opt(r.total_s_median, 3),
+            fmt_opt(r.total_s_p95, 3),
+            fmt_vram(r.vram_used_mb_median, r.vram_total_mb),
+            fmt_u64(r.concurrency),
+            fmt_opt(r.queue_wait_ms_median, 0),
+            fmt_opt(r.rejected_median, 0),
             r.git_sha,
             r.samples,
         ));
@@ -43,16 +51,205 @@ pub fn render_json(rows: &[ReportRow]) -> Result<String> {
                 "prompt_size_approx": r.prompt_size_approx,
                 "prompt_tokens": r.prompt_tokens,
                 "ttft_s_median": r.ttft_s_median,
+                "ttft_s_p95": r.ttft_s_p95,
+                "ttft_s_p99": r.ttft_s_p99,
                 "decode_tps_median": r.decode_tps_median,
                 "total_s_median": r.total_s_median,
+                "total_s_p95": r.total_s_p95,
+                "total_s_p99": r.total_s_p99,
+                "prefill_ms_median": r.prefill_ms_median,
+                "decode_ms_median": r.decode_ms_median,
+                "prefill_tps_median": r.prefill_tps_median,
+                "vram_used_mb_median": r.vram_used_mb_median,
+                "vram_total_mb": r.vram_total_mb,
+                "gpu_util_pct_median": r.gpu_util_pct_median,
+                "gpu_temp_c_median": r.gpu_temp_c_median,
+                "concurrency": r.concurrency,
+                "ttft_p95_load_s": r.ttft_p95_load_s,
+                "queue_wait_ms_median": r.queue_wait_ms_median,
+                "rejected_median": r.rejected_median,
                 "git_sha": r.git_sha,
                 "samples": r.samples,
+                "gpu": r.gpu,
             })
         })
         .collect();
     Ok(serde_json::to_string_pretty(&arr)?)
 }
 
+/// Context-length scaling view (#88): one block per (target, model) with
+/// prefill & decode tok/s vs context, then the decode-flatness verdict.
+pub fn render_scaling_markdown(curves: &[ScalingCurve]) -> String {
+    let mut out = String::new();
+    for c in curves {
+        let gpu = c.gpu.as_deref().unwrap_or("");
+        out.push_str(&format!(
+            "### {} · {}  (`{}`{})\n\n",
+            c.target_name,
+            c.model_id,
+            c.git_sha,
+            if gpu.is_empty() {
+                String::new()
+            } else {
+                format!(", {gpu}")
+            },
+        ));
+        out.push_str("| ctx tok | prefill tok/s | decode tok/s | n |\n");
+        out.push_str("|---:|---:|---:|---:|\n");
+        for p in &c.points {
+            let ctx = p
+                .prompt_tokens
+                .map(|t| t.to_string())
+                .unwrap_or_else(|| format!("~{}", p.prompt_size));
+            out.push_str(&format!(
+                "| {} | {} | {} | {} |\n",
+                ctx,
+                fmt_opt(p.prefill_tps, 1),
+                fmt_opt(p.decode_tps, 1),
+                p.samples,
+            ));
+        }
+        match c.decode_flatness {
+            Some(f) => out.push_str(&format!(
+                "\ndecode flatness: {f:.2} — decode tok/s {} across the context range \
+                 ({})\n\n",
+                if f >= 0.9 {
+                    "holds"
+                } else if f >= 0.7 {
+                    "softens"
+                } else {
+                    "drops sharply"
+                },
+                if f >= 0.9 {
+                    "Gated-DeltaNet O(1) decode confirmed"
+                } else {
+                    "investigate where it breaks"
+                },
+            )),
+            None => out.push_str("\ndecode flatness: — (need ≥2 context points)\n\n"),
+        }
+    }
+    out
+}
+
+pub fn render_scaling_json(curves: &[ScalingCurve]) -> Result<String> {
+    Ok(serde_json::to_string_pretty(curves)?)
+}
+
+/// Cold-load / model-swap cost view (#90): reload latency + cold
+/// first-request per model.
+pub fn render_swap_markdown(costs: &[SwapCost]) -> String {
+    let mut out = String::new();
+    out.push_str(
+        "| engine | model | unload (s) | reload (s) | cold TTFT (s) | cold total (s) | build | n |\n",
+    );
+    out.push_str("|---|---|---:|---:|---:|---:|---|---:|\n");
+    for c in costs {
+        out.push_str(&format!(
+            "| {} | {} | {} | {} | {} | {} | `{}` | {} |\n",
+            c.target_name,
+            c.model_id,
+            fmt_ms_as_s(c.unload_ms_median),
+            fmt_ms_as_s(c.load_ms_median),
+            fmt_opt(c.cold_ttft_s_median, 3),
+            fmt_opt(c.cold_total_s_median, 3),
+            c.git_sha,
+            c.samples,
+        ));
+    }
+    out
+}
+
+pub fn render_swap_json(costs: &[SwapCost]) -> Result<String> {
+    Ok(serde_json::to_string_pretty(costs)?)
+}
+
+/// Capability-probe view (#91): per (model, probe) the median quality score
+/// (the A/B number), then each run's id, score, and an artifact snippet so
+/// unscored runs can be located and scored (`helexa-bench score --id …`).
+pub fn render_capability_markdown(runs: &[CapabilityRun]) -> String {
+    use std::collections::BTreeMap;
+    let mut groups: BTreeMap<(String, String, String), Vec<&CapabilityRun>> = BTreeMap::new();
+    for r in runs {
+        groups
+            .entry((
+                r.target_name.clone(),
+                r.model_id.clone(),
+                r.scenario_id.clone(),
+            ))
+            .or_default()
+            .push(r);
+    }
+    let mut out = String::new();
+    for ((target, model, scenario), rs) in groups {
+        let scores: Vec<f64> = rs.iter().filter_map(|r| r.quality_score).collect();
+        let median = median_slice(&scores);
+        out.push_str(&format!(
+            "### {target} · {model} · {scenario} — median score {} ({}/{} scored)\n\n",
+            median
+                .map(|m| format!("{m:.1}"))
+                .unwrap_or_else(|| "—".into()),
+            scores.len(),
+            rs.len(),
+        ));
+        out.push_str("| run | score | scorer | build | artifact (snippet) |\n");
+        out.push_str("|---:|---:|---|---|---|\n");
+        for r in rs {
+            out.push_str(&format!(
+                "| {} | {} | {} | `{}` | {} |\n",
+                r.id,
+                r.quality_score
+                    .map(|s| format!("{s:.1}"))
+                    .unwrap_or_else(|| "—".into()),
+                r.scorer.as_deref().unwrap_or("—"),
+                r.git_sha,
+                snippet(r.artifact.as_deref()),
+            ));
+        }
+        out.push('\n');
+    }
+    out
+}
+
+pub fn render_capability_json(runs: &[CapabilityRun]) -> Result<String> {
+    Ok(serde_json::to_string_pretty(runs)?)
+}
+
+/// First ~80 chars of an artifact on one line, for the table cell.
+fn snippet(artifact: Option<&str>) -> String {
+    match artifact {
+        Some(a) => {
+            let one_line: String = a.split_whitespace().collect::<Vec<_>>().join(" ");
+            let trimmed: String = one_line.chars().take(80).collect();
+            if one_line.chars().count() > 80 {
+                format!("{trimmed}…")
+            } else {
+                trimmed
+            }
+        }
+        None => "—".to_string(),
+    }
+}
+
+fn median_slice(v: &[f64]) -> Option<f64> {
+    if v.is_empty() {
+        return None;
+    }
+    let mut s = v.to_vec();
+    s.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
+    let lo = (s.len() - 1) / 2;
+    let hi = s.len() / 2;
+    Some((s[lo] + s[hi]) / 2.0)
+}
+
+/// Milliseconds rendered as seconds (reload costs read naturally in s).
+fn fmt_ms_as_s(ms: Option<f64>) -> String {
+    match ms {
+        Some(x) => format!("{:.2}", x / 1000.0),
+        None => "—".to_string(),
+    }
+}
+
 fn fmt_opt(v: Option<f64>, places: usize) -> String {
     match v {
         Some(x) => format!("{x:.places$}"),
@@ -60,9 +257,113 @@ fn fmt_opt(v: Option<f64>, places: usize) -> String {
     }
 }
 
+/// Integer cell (concurrency width); `—` when unset (non-concurrency rows).
+fn fmt_u64(v: Option<u64>) -> String {
+    match v {
+        Some(x) => x.to_string(),
+        None => "—".to_string(),
+    }
+}
+
+/// `used/total` in GB (e.g. `42.0/64.0`) — the headroom-at-a-glance cell.
+/// `used` alone if the node total is unknown; `—` if no telemetry.
+fn fmt_vram(used_mb: Option<f64>, total_mb: Option<u64>) -> String {
+    match (used_mb, total_mb) {
+        (Some(u), Some(t)) => format!("{:.1}/{:.1}", u / 1024.0, t as f64 / 1024.0),
+        (Some(u), None) => format!("{:.1}", u / 1024.0),
+        _ => "—".to_string(),
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::store::{ScalingCurve, ScalingPoint};
+
+    #[test]
+    fn capability_markdown_groups_with_median_and_snippet() {
+        let runs = vec![
+            CapabilityRun {
+                id: 7,
+                ts: "t".into(),
+                target_name: "beast".into(),
+                model_id: "m".into(),
+                scenario_id: "capability:plan".into(),
+                git_sha: "abc".into(),
+                quality_score: Some(8.0),
+                scorer: Some("manual".into()),
+                artifact: Some("A detailed plan with trade-offs and sequencing.".into()),
+            },
+            CapabilityRun {
+                id: 8,
+                ts: "t".into(),
+                target_name: "beast".into(),
+                model_id: "m".into(),
+                scenario_id: "capability:plan".into(),
+                git_sha: "abc".into(),
+                quality_score: Some(6.0),
+                scorer: Some("manual".into()),
+                artifact: Some("Shorter plan.".into()),
+            },
+        ];
+        let md = render_capability_markdown(&runs);
+        assert!(md.contains("capability:plan"));
+        assert!(md.contains("median score 7.0")); // median(8,6)
+        assert!(md.contains("trade-offs"));
+        assert!(md.contains("| 7 |"));
+    }
+
+    #[test]
+    fn swap_markdown_renders_reload_and_cold_costs() {
+        let costs = vec![SwapCost {
+            target_name: "beast".into(),
+            model_id: "Qwen/Qwen3.6-27B".into(),
+            git_sha: "abc1234".into(),
+            gpu: Some("2× RTX 5090".into()),
+            unload_ms_median: Some(320.0),
+            load_ms_median: Some(25000.0),
+            cold_ttft_s_median: Some(2.5),
+            cold_total_s_median: Some(5.0),
+            samples: 3,
+        }];
+        let md = render_swap_markdown(&costs);
+        assert!(md.contains("reload (s)"));
+        assert!(md.contains("beast"));
+        assert!(md.contains("25.00")); // 25000 ms → 25.00 s
+        assert!(md.contains("2.500"));
+    }
+
+    #[test]
+    fn scaling_markdown_renders_curve_and_flatness() {
+        let curves = vec![ScalingCurve {
+            target_name: "beast".into(),
+            model_id: "Qwen/Qwen3.6-27B".into(),
+            git_sha: "abc1234".into(),
+            gpu: Some("2× RTX 5090".into()),
+            points: vec![
+                ScalingPoint {
+                    prompt_size: 128,
+                    prompt_tokens: Some(130),
+                    prefill_tps: Some(900.0),
+                    decode_tps: Some(50.0),
+                    samples: 5,
+                },
+                ScalingPoint {
+                    prompt_size: 4096,
+                    prompt_tokens: Some(4100),
+                    prefill_tps: Some(2800.0),
+                    decode_tps: Some(48.0),
+                    samples: 5,
+                },
+            ],
+            decode_flatness: Some(0.96),
+        }];
+        let md = render_scaling_markdown(&curves);
+        assert!(md.contains("### beast · Qwen/Qwen3.6-27B"));
+        assert!(md.contains("ctx tok"));
+        assert!(md.contains("decode flatness: 0.96"));
+        assert!(md.contains("holds"));
+    }
 
     #[test]
     fn markdown_has_header_and_row() {
@@ -76,13 +377,36 @@ mod tests {
             ttft_s_median: Some(0.123),
             decode_tps_median: Some(45.6),
             total_s_median: Some(1.234),
+            ttft_s_p95: Some(0.222),
+            ttft_s_p99: Some(0.250),
+            total_s_p95: Some(1.5),
+            total_s_p99: Some(1.6),
+            prefill_ms_median: Some(120.0),
+            decode_ms_median: Some(1100.0),
+            prefill_tps_median: Some(1066.7),
+            vram_used_mb_median: Some(43008.0),
+            vram_total_mb: Some(65536),
+            gpu_util_pct_median: Some(89.0),
+            gpu_temp_c_median: Some(64.0),
+            concurrency: None,
+            ttft_p95_load_s: None,
+            queue_wait_ms_median: None,
+            rejected_median: None,
             samples: 5,
+            gpu: Some("2× RTX 5090".into()),
         }];
         let md = render_markdown(&rows);
         assert!(md.contains("| engine |"));
+        assert!(md.contains("prefill tok/s"));
+        assert!(md.contains("VRAM (GB)"));
+        assert!(md.contains("conc"));
         assert!(md.contains("beast"));
         assert!(md.contains("`30d50d6`"));
         assert!(md.contains("0.123"));
+        // p95 column rendered.
+        assert!(md.contains("0.222"));
+        // VRAM used/total in GB (43008/65536 MiB → 42.0/64.0).
+        assert!(md.contains("42.0/64.0"));
     }
 
     #[test]
@@ -97,7 +421,23 @@ mod tests {
             ttft_s_median: Some(0.1),
             decode_tps_median: None,
             total_s_median: Some(0.5),
+            ttft_s_p95: Some(0.1),
+            ttft_s_p99: Some(0.1),
+            total_s_p95: Some(0.5),
+            total_s_p99: Some(0.5),
+            prefill_ms_median: None,
+            decode_ms_median: None,
+            prefill_tps_median: None,
+            vram_used_mb_median: None,
+            vram_total_mb: None,
+            gpu_util_pct_median: None,
+            gpu_temp_c_median: None,
+            concurrency: None,
+            ttft_p95_load_s: None,
+            queue_wait_ms_median: None,
+            rejected_median: None,
             samples: 1,
+            gpu: None,
         }];
         let md = render_markdown(&rows);
         assert!(md.contains("~128"));

crates/helexa-bench/src/scenario.rs

@@ -62,6 +62,41 @@ pub struct ScenarioMetrics {
     pub prompt_tokens: Option<u64>,
     /// Completion tokens: from `usage` when present, else content-chunk count.
     pub completion_tokens: u64,
+    /// Server-measured prefill duration (ms), from the `usage.helexa_timing`
+    /// extension (#85). `None` when the server didn't emit it (external
+    /// engines, non-instrumented paths). The honest prefill-phase number,
+    /// distinct from client-observed `ttft_s` which also includes request
+    /// setup + first-byte network latency.
+    pub prefill_ms: Option<u64>,
+    /// Server-measured decode duration (ms), from `usage.helexa_timing`.
+    pub decode_ms: Option<u64>,
+    /// Tokens submitted to prefill — the denominator for prefill tok/s.
+    pub prefill_tokens: Option<u64>,
+    // ── Concurrency / agentic-load fields (#89) ──────────────────────────
+    // Set only by the concurrency scenario, which fans out N simultaneous
+    // streams to characterize the real a0/hermes/opencode workload that
+    // batch-1 single-request measurement can't see. `None` for single
+    // requests. For a concurrency burst, the inherited fields carry the
+    // aggregate: `ttft_s` = median TTFT across streams, `decode_tps` = node
+    // throughput (total tokens / burst window), `total_s` = burst wall-clock,
+    // `completion_tokens` = total across streams.
+    /// Number of simultaneous streams in the burst (the cell dimension).
+    pub concurrency: Option<u32>,
+    /// p95 of per-stream TTFT within the burst — the tail under simultaneous
+    /// load, where batch-1 serialization actually hurts.
+    pub ttft_p95_s: Option<f64>,
+    /// Median per-stream admission queue-wait (ms), approximated as
+    /// `ttft − prefill_ms` (#85): on a batch-1 server, later streams wait for
+    /// earlier ones, so TTFT inflates while server prefill stays constant —
+    /// the gap is the wait. `None` if streams didn't report `helexa_timing`.
+    pub queue_wait_ms_median: Option<f64>,
+    /// Streams shed by admission control (HTTP 429/503) during the burst —
+    /// honest backpressure, not silent failures.
+    pub rejected: Option<u32>,
+    /// Full generated text, captured only by the capability probe (#91) so
+    /// the output can be quality-scored later (manual or LLM-judge). `None`
+    /// for latency/throughput scenarios, which discard the text.
+    pub artifact: Option<String>,
 }
 
 #[async_trait]
@@ -85,10 +120,13 @@ pub trait Scenario: Send + Sync {
     async fn run(&self, ctx: &RunCtx) -> Result<ScenarioMetrics>;
 }
 
-/// Build the active scenario set from config. One chat-latency scenario
-/// per configured prompt size.
+/// Build the active scenario set from config: one chat-latency scenario per
+/// prompt size, plus one concurrency scenario per configured level (#89).
+/// Concurrency levels default to empty (opt-in), since a burst puts real
+/// simultaneous load on a serving fleet — operators enable it deliberately.
 pub fn build_scenarios(cfg: &ScenarioConfig) -> Vec<Box<dyn Scenario>> {
-    cfg.prompt_sizes
+    let mut scenarios: Vec<Box<dyn Scenario>> = cfg
+        .prompt_sizes
         .iter()
         .map(|&size| {
             Box::new(ChatLatencyScenario {
@@ -96,7 +134,46 @@ pub fn build_scenarios(cfg: &ScenarioConfig) -> Vec<Box<dyn Scenario>> {
                 approx_prompt_tokens: size,
             }) as Box<dyn Scenario>
         })
-        .collect()
+        .collect();
+    for &n in &cfg.concurrency_levels {
+        scenarios.push(Box::new(ConcurrencyScenario {
+            id: format!("concurrency:{n}"),
+            concurrency: n,
+            approx_prompt_tokens: cfg.concurrency_prompt_tokens,
+        }) as Box<dyn Scenario>);
+    }
+    for probe in &cfg.capability_probes {
+        scenarios.push(Box::new(CapabilityScenario {
+            id: format!("capability:{}", probe.name),
+            prompt: probe.prompt.clone(),
+            max_tokens: probe.max_tokens,
+        }) as Box<dyn Scenario>);
+    }
+    scenarios
+}
+
+/// A single small streamed request, timed like a chat-latency run. Used by
+/// the swap-cost measurement (#90) to capture the cold first-request latency
+/// straight after a reload. Reuses the shared SSE-timing core.
+pub async fn cold_probe(ctx: &RunCtx<'_>) -> Result<ScenarioMetrics> {
+    let prompt = build_prompt(128);
+    let payload = chat_payload(ctx, &prompt);
+    tokio::time::timeout(ctx.timeout, stream_and_measure(ctx, &payload))
+        .await
+        .map_err(|_| anyhow!("cold probe timed out after {:?}", ctx.timeout))?
+}
+
+/// The chat-completions request body shared by the latency and concurrency
+/// scenarios — streamed, deterministic (temperature 0), usage included.
+fn chat_payload(ctx: &RunCtx, prompt: &str) -> serde_json::Value {
+    json!({
+        "model": ctx.model_id,
+        "messages": [{"role": "user", "content": prompt}],
+        "max_tokens": ctx.max_tokens,
+        "temperature": 0,
+        "stream": true,
+        "stream_options": {"include_usage": true},
+    })
 }
 
 /// Streamed single-request chat-completions latency probe — the batch-1
@@ -118,15 +195,7 @@ impl Scenario for ChatLatencyScenario {
 
     async fn run(&self, ctx: &RunCtx) -> Result<ScenarioMetrics> {
         let prompt = build_prompt(self.approx_prompt_tokens);
-        let payload = json!({
-            "model": ctx.model_id,
-            "messages": [{"role": "user", "content": prompt}],
-            "max_tokens": ctx.max_tokens,
-            "temperature": 0,
-            "stream": true,
-            "stream_options": {"include_usage": true},
-        });
-
+        let payload = chat_payload(ctx, &prompt);
         let fut = stream_and_measure(ctx, &payload);
         tokio::time::timeout(ctx.timeout, fut)
             .await
@@ -134,11 +203,182 @@ impl Scenario for ChatLatencyScenario {
     }
 }
 
+/// Fan-out load probe: fire `concurrency` identical streams at once and
+/// measure how the fleet behaves under simultaneous pressure (#89). This is
+/// the only scenario that exercises the real a0/hermes/opencode pattern —
+/// many agentic requests per user turn — which batch-1 single-request
+/// timing cannot characterize. On a batch-1 serialized server, aggregate
+/// throughput stays ~flat while TTFT/queue-wait inflate with `concurrency`;
+/// that gap is the evidence for/against continuous batching.
+pub struct ConcurrencyScenario {
+    id: String,
+    concurrency: u32,
+    approx_prompt_tokens: u32,
+}
+
+#[async_trait]
+impl Scenario for ConcurrencyScenario {
+    fn id(&self) -> &str {
+        &self.id
+    }
+
+    fn prompt_size(&self) -> u32 {
+        self.approx_prompt_tokens
+    }
+
+    async fn run(&self, ctx: &RunCtx) -> Result<ScenarioMetrics> {
+        let prompt = build_prompt(self.approx_prompt_tokens);
+        let payload = chat_payload(ctx, &prompt);
+
+        // Fire all streams at once; each is independently timed and capped by
+        // the per-request timeout so one hung stream can't stall the burst.
+        let burst_start = Instant::now();
+        let futs = (0..self.concurrency).map(|_| async {
+            tokio::time::timeout(ctx.timeout, stream_and_measure(ctx, &payload)).await
+        });
+        let results = futures::future::join_all(futs).await;
+        let burst_window = burst_start.elapsed().as_secs_f64();
+
+        let mut streams: Vec<ScenarioMetrics> = Vec::new();
+        let mut rejected: u32 = 0;
+        for r in results {
+            match r {
+                Ok(Ok(m)) => streams.push(m),
+                // Admission backpressure (429/503) is shed load, counted
+                // separately from genuine failures/timeouts.
+                Ok(Err(e)) if is_admission_reject(&e) => rejected += 1,
+                Ok(Err(_)) | Err(_) => {}
+            }
+        }
+        if streams.is_empty() {
+            return Err(anyhow!(
+                "all {} concurrent streams failed ({rejected} shed by admission)",
+                self.concurrency
+            ));
+        }
+
+        let total_tokens: u64 = streams.iter().map(|m| m.completion_tokens).sum();
+        let ttfts: Vec<f64> = streams.iter().map(|m| m.ttft_s).collect();
+        // queue-wait ≈ TTFT − server prefill (#85); only for streams that
+        // reported helexa_timing.
+        let queue_waits: Vec<f64> = streams
+            .iter()
+            .filter_map(|m| {
+                m.prefill_ms
+                    .map(|p| (m.ttft_s * 1000.0 - p as f64).max(0.0))
+            })
+            .collect();
+        // Aggregate decode throughput across the whole node for the burst.
+        let aggregate_tps = if burst_window > 0.0 {
+            Some(total_tokens as f64 / burst_window)
+        } else {
+            None
+        };
+
+        Ok(ScenarioMetrics {
+            ttft_s: median(&ttfts).unwrap_or(0.0),
+            decode_tps: aggregate_tps,
+            total_s: burst_window,
+            prompt_tokens: streams.iter().find_map(|m| m.prompt_tokens),
+            completion_tokens: total_tokens,
+            prefill_ms: None,
+            decode_ms: None,
+            prefill_tokens: None,
+            concurrency: Some(self.concurrency),
+            ttft_p95_s: percentile(&ttfts, 95.0),
+            queue_wait_ms_median: median(&queue_waits),
+            rejected: Some(rejected),
+            artifact: None,
+        })
+    }
+}
+
+/// Quality probe (#91): runs a fixed prompt and stores the full generated
+/// text as an artifact for later scoring (manual now, LLM-judge later). The
+/// point is to compare reasoning/planning quality across models — the axis
+/// speed-only scenarios miss — so the frontier A/B (F3) picks on capability,
+/// not just throughput.
+pub struct CapabilityScenario {
+    id: String,
+    prompt: String,
+    max_tokens: u64,
+}
+
+#[async_trait]
+impl Scenario for CapabilityScenario {
+    fn id(&self) -> &str {
+        &self.id
+    }
+
+    /// Capability probes have no synthetic prompt-token target; the cell is
+    /// keyed by the scenario id alone.
+    fn prompt_size(&self) -> u32 {
+        0
+    }
+
+    async fn run(&self, ctx: &RunCtx) -> Result<ScenarioMetrics> {
+        let payload = json!({
+            "model": ctx.model_id,
+            "messages": [{"role": "user", "content": self.prompt}],
+            "max_tokens": self.max_tokens,
+            "temperature": 0,
+            "stream": true,
+            "stream_options": {"include_usage": true},
+        });
+        let fut = stream_and_measure_inner(ctx, &payload, true);
+        tokio::time::timeout(ctx.timeout, fut)
+            .await
+            .map_err(|_| anyhow!("capability probe timed out after {:?}", ctx.timeout))?
+    }
+}
+
+/// Whether a stream error was admission backpressure (HTTP 429/503) rather
+/// than a genuine failure. `stream_and_measure` renders the upstream status
+/// into the error string, so a substring check is sufficient.
+fn is_admission_reject(e: &anyhow::Error) -> bool {
+    let s = e.to_string();
+    s.contains("429") || s.contains("503")
+}
+
+/// Median of a slice (sorted copy). `None` if empty.
+fn median(values: &[f64]) -> Option<f64> {
+    if values.is_empty() {
+        return None;
+    }
+    let mut v = values.to_vec();
+    v.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
+    let lo = (v.len() - 1) / 2;
+    let hi = v.len() / 2;
+    Some((v[lo] + v[hi]) / 2.0)
+}
+
+/// Nearest-rank percentile of a slice (`p` in 0..=100). `None` if empty.
+fn percentile(values: &[f64], p: f64) -> Option<f64> {
+    if values.is_empty() {
+        return None;
+    }
+    let mut v = values.to_vec();
+    v.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
+    let rank = (p / 100.0 * v.len() as f64).ceil() as usize;
+    Some(v[rank.clamp(1, v.len()) - 1])
+}
+
 /// The SSE-timing core, ported from `bench.py::one_run`. Kept free of the
 /// `Scenario` trait so it's unit-testable against a mock byte stream.
 async fn stream_and_measure(
     ctx: &RunCtx<'_>,
     payload: &serde_json::Value,
+) -> Result<ScenarioMetrics> {
+    stream_and_measure_inner(ctx, payload, false).await
+}
+
+/// As [`stream_and_measure`] but accumulates the full visible text when
+/// `capture_text` is set — used by the capability probe (#91) to store the
+/// generated artifact for later quality scoring.
+async fn stream_and_measure_inner(
+    ctx: &RunCtx<'_>,
+    payload: &serde_json::Value,
+    capture_text: bool,
 ) -> Result<ScenarioMetrics> {
     let start = Instant::now();
     let resp = ctx
@@ -160,6 +400,10 @@ async fn stream_and_measure(
     let mut chunk_count: u64 = 0;
     let mut prompt_tokens: Option<u64> = None;
     let mut completion_tokens: Option<u64> = None;
+    let mut prefill_ms: Option<u64> = None;
+    let mut decode_ms: Option<u64> = None;
+    let mut prefill_tokens: Option<u64> = None;
+    let mut captured = String::new();
 
     while let Some(event) = stream.next().await {
         let event = event.context("reading SSE stream")?;
@@ -172,46 +416,86 @@ async fn stream_and_measure(
             Ok(c) => c,
             Err(_) => continue, // tolerate non-JSON keepalive frames
         };
-        if let Some(choice) = chunk.choices.first()
-            && choice
+        if let Some(choice) = chunk.choices.first() {
+            // Liveness counts ANY generated delta (#117). Thinking
+            // models (Qwen3-Next-Thinking, Qwen3 with thinking on)
+            // stream `reasoning_content` first — sometimes for their
+            // entire budget — and a content-only view misread that as
+            // a dead stream ("no content chunks received") while also
+            // producing impossible client-side rates (reasoning-
+            // inclusive token counts over a visible-content-only
+            // window; observed: "244 tok/s" on a 3060). For
+            // non-thinking models the first delta IS content, so
+            // `ttft_s` semantics are unchanged for them.
+            let content = choice
                 .delta
                 .get("content")
                 .and_then(|c| c.as_str())
-                .is_some_and(|s| !s.is_empty())
-        {
+                .filter(|c| !c.is_empty());
+            let reasoning = choice
+                .delta
+                .get("reasoning_content")
+                .and_then(|c| c.as_str())
+                .filter(|c| !c.is_empty());
+            if content.is_some() || reasoning.is_some() {
                 if first.is_none() {
                     first = Some(now);
                 }
                 last = Some(now);
                 chunk_count += 1;
             }
+            if capture_text && let Some(text) = content {
+                captured.push_str(text);
+            }
+        }
         if let Some(usage) = chunk.usage {
             prompt_tokens = Some(usage.prompt_tokens);
             completion_tokens = Some(usage.completion_tokens);
+            if let Some(t) = usage.helexa_timing {
+                prefill_ms = Some(t.prefill_ms);
+                decode_ms = Some(t.decode_ms);
+                prefill_tokens = Some(t.prefill_tokens);
+            }
         }
     }
     let end = Instant::now();
 
-    let first = first.ok_or_else(|| anyhow!("no content chunks received"))?;
+    let first = first.ok_or_else(|| anyhow!("no generated chunks received"))?;
 
-    // neuron emits one SSE chunk per visible token, so chunk_count is an
-    // engine-truth count when no usage frame is sent.
+    // neuron emits one SSE chunk per generated token, so chunk_count is
+    // an engine-truth count when no usage frame is sent.
     let tokens = completion_tokens.filter(|&t| t > 0).unwrap_or(chunk_count);
-    // decode rate is only meaningful over a real inter-chunk window.
+    // Decode rate: prefer the server-measured split (#85) — it counts
+    // every generated token over the actual decode window, immune to
+    // reasoning-suppression frame mismatches. Fall back to the client
+    // inter-chunk window with the CHUNK count (same frame) — never
+    // usage.completion_tokens over the chunk window, which mixes a
+    // reasoning-inclusive numerator with a visible-only denominator.
     let window = last
         .filter(|&l| l > first)
         .map(|l| (l - first).as_secs_f64())
         .unwrap_or(0.0);
+    let decode_tps = match decode_ms {
+        Some(ms) if ms > 200 && tokens > 0 => Some(tokens as f64 / (ms as f64 / 1000.0)),
+        _ if window > 0.2 => Some(chunk_count as f64 / window),
+        _ => None,
+    };
     Ok(ScenarioMetrics {
         ttft_s: (first - start).as_secs_f64(),
-        decode_tps: if window > 0.2 {
-            Some(tokens as f64 / window)
-        } else {
-            None
-        },
+        decode_tps,
         total_s: (end - start).as_secs_f64(),
         prompt_tokens,
         completion_tokens: tokens,
+        prefill_ms,
+        decode_ms,
+        prefill_tokens,
+        // Concurrency fields unset on the single-request path; the
+        // concurrency scenario builds its own aggregate (#89).
+        concurrency: None,
+        ttft_p95_s: None,
+        queue_wait_ms_median: None,
+        rejected: None,
+        artifact: if capture_text { Some(captured) } else { None },
     })
 }
 
@@ -229,6 +513,54 @@ mod tests {
         assert!(small.ends_with("/no_think"));
     }
 
+    #[test]
+    fn median_and_percentile_basics() {
+        assert_eq!(median(&[3.0, 1.0, 2.0]), Some(2.0));
+        assert_eq!(median(&[]), None);
+        let v = [1.0, 2.0, 3.0, 4.0, 5.0];
+        assert_eq!(percentile(&v, 50.0), Some(3.0));
+        assert_eq!(percentile(&v, 95.0), Some(5.0)); // nearest-rank → max with n=5
+        assert_eq!(percentile(&[], 95.0), None);
+    }
+
+    #[test]
+    fn admission_rejects_detected_by_status() {
+        assert!(is_admission_reject(&anyhow!(
+            "upstream returned 429 Too Many Requests"
+        )));
+        assert!(is_admission_reject(&anyhow!(
+            "upstream returned 503 Service Unavailable"
+        )));
+        assert!(!is_admission_reject(&anyhow!(
+            "upstream returned 500 Internal"
+        )));
+        assert!(!is_admission_reject(&anyhow!("connection refused")));
+    }
+
+    #[test]
+    fn concurrency_scenarios_built_from_config() {
+        use crate::config::{CapabilityProbe, ScenarioConfig};
+        let cfg = ScenarioConfig {
+            prompt_sizes: vec![128],
+            max_tokens: 64,
+            concurrency_levels: vec![2, 8],
+            concurrency_prompt_tokens: 512,
+            capability_probes: vec![CapabilityProbe {
+                name: "plan".into(),
+                prompt: "Write a plan.".into(),
+                max_tokens: 2048,
+            }],
+        };
+        let ids: Vec<String> = build_scenarios(&cfg)
+            .iter()
+            .map(|s| s.id().to_string())
+            .collect();
+        assert!(ids.contains(&"chat:128".to_string()));
+        assert!(ids.contains(&"concurrency:2".to_string()));
+        assert!(ids.contains(&"concurrency:8".to_string()));
+        assert!(ids.contains(&"capability:plan".to_string()));
+    }
+
     #[test]
     fn prompt_floor_for_tiny_targets() {
         // max(approx,16) floor means even 0 yields a non-trivial prompt.

crates/helexa-bench/src/sweep.rs

@@ -9,11 +9,11 @@
 
 use crate::client::TargetClient;
 use crate::config::{BenchConfig, TargetConfig, TargetKind};
-use crate::scenario::{RunCtx, build_scenarios};
+use crate::scenario::{RunCtx, ScenarioMetrics, build_scenarios};
 use crate::store::{RunRecord, Store};
 use anyhow::Result;
 use cortex_core::build_info::BuildInfo;
-use cortex_core::discovery::DiscoveryResponse;
+use cortex_core::discovery::{DiscoveryResponse, HealthResponse};
 use cortex_core::harness::ModelInfo;
 
 /// helexa-bench's own build version.
@@ -38,6 +38,38 @@ pub struct SweepSummary {
     pub targets_unreachable: usize,
 }
 
+/// Node-level GPU telemetry folded from one `/health` snapshot (#87):
+/// VRAM used summed across the node's devices, and the hottest/busiest
+/// single device for utilization and temperature.
+#[derive(Debug, Clone, Copy)]
+struct HealthAgg {
+    vram_used_mb: u64,
+    gpu_util_pct: u32,
+    gpu_temp_c: u32,
+}
+
+/// Cold-load / model-swap timing for one measure_swap cycle (#90).
+#[derive(Debug, Clone, Copy)]
+struct SwapTiming {
+    unload_ms: u64,
+    load_ms: u64,
+}
+
+impl HealthAgg {
+    fn from_health(h: &HealthResponse) -> Self {
+        HealthAgg {
+            vram_used_mb: h.devices.iter().map(|d| d.vram_used_mb).sum(),
+            gpu_util_pct: h
+                .devices
+                .iter()
+                .map(|d| d.utilization_pct)
+                .max()
+                .unwrap_or(0),
+            gpu_temp_c: h.devices.iter().map(|d| d.temp_c).max().unwrap_or(0),
+        }
+    }
+}
+
 pub struct Sweeper {
     cfg: BenchConfig,
     client: TargetClient,
@@ -72,6 +104,105 @@ impl Sweeper {
         }
     }
 
+    /// Deliberate cold-load / model-swap cost measurement (#90), invoked by
+    /// the `swap-cost` subcommand — **never** the continuous sweep. For each
+    /// neuron target and each currently-warm model: unload it, time the
+    /// reload, then time a cold first request. This takes the model offline
+    /// for the reload, so it is an explicit operator action (maintenance
+    /// window), recorded under `scenario_id = "swap"`.
+    pub async fn swap_cost_once(&self) -> Result<SweepSummary> {
+        let mut summary = SweepSummary::default();
+        for target in &self.cfg.targets {
+            if target.kind != TargetKind::Neuron {
+                continue; // load/unload is a neuron-native operation
+            }
+            let build = match self.client.fetch_version(target).await {
+                Ok(b) => b,
+                Err(e) => {
+                    summary.targets_unreachable += 1;
+                    tracing::warn!(target = %target.name, error = %format!("{e:#}"), "swap: target unreachable");
+                    continue;
+                }
+            };
+            let discovery = self.client.fetch_discovery(target).await.unwrap_or(None);
+            let models = self.client.warm_models(target).await.unwrap_or_default();
+            for model in &models {
+                match self
+                    .measure_swap(target, &build, discovery.as_ref(), model)
+                    .await
+                {
+                    Ok(()) => summary.measured += 1,
+                    Err(e) => {
+                        summary.failed += 1;
+                        tracing::warn!(target = %target.name, model = %model.id, error = %format!("{e:#}"), "swap: measurement failed");
+                    }
+                }
+            }
+        }
+        Ok(summary)
+    }
+
+    /// Unload → timed reload → timed cold first request for one model.
+    async fn measure_swap(
+        &self,
+        target: &TargetConfig,
+        build: &BuildInfo,
+        discovery: Option<&DiscoveryResponse>,
+        model: &ModelInfo,
+    ) -> Result<()> {
+        let spec = TargetClient::spec_from_info(model)?;
+        tracing::warn!(target = %target.name, model = %model.id, "swap: unloading (model goes offline until reload)");
+
+        let t0 = std::time::Instant::now();
+        self.client.unload_model(target, &model.id).await?;
+        let unload_ms = t0.elapsed().as_millis() as u64;
+
+        let t1 = std::time::Instant::now();
+        self.client.load_model(target, &spec).await?;
+        let load_ms = t1.elapsed().as_millis() as u64;
+        tracing::info!(target = %target.name, model = %model.id, unload_ms, load_ms, "swap: reloaded; measuring cold first request");
+
+        // Cold first request — caches empty straight after the load.
+        let ctx = RunCtx {
+            client: self.client.http(),
+            chat_url: self.client.chat_url(target),
+            model_id: model.id.clone(),
+            max_tokens: self.cfg.scenarios.max_tokens,
+            timeout: self.cfg.bench.request_timeout(),
+        };
+        let cold = crate::scenario::cold_probe(&ctx).await;
+        let swap = SwapTiming { unload_ms, load_ms };
+        let rec = match &cold {
+            Ok(m) => self.build_record(
+                target,
+                build,
+                discovery,
+                model,
+                "swap",
+                0,
+                Ok(m),
+                None,
+                Some(swap),
+            ),
+            Err(e) => {
+                let msg = format!("{e:#}");
+                self.build_record(
+                    target,
+                    build,
+                    discovery,
+                    model,
+                    "swap",
+                    0,
+                    Err(&msg),
+                    None,
+                    Some(swap),
+                )
+            }
+        };
+        self.store.insert_run(&rec)?;
+        Ok(())
+    }
+
     /// One full pass over all targets.
     pub async fn run_once(&self) -> Result<SweepSummary> {
         let mut summary = SweepSummary::default();
@@ -131,7 +262,21 @@ impl Sweeper {
                 }
 
                 for i in 0..need {
-                    match scenario.run(&ctx).await {
+                    let result = scenario.run(&ctx).await;
+                    // Sample GPU telemetry right after the run, while the
+                    // model is loaded and decode VRAM is at its recent peak
+                    // (#87). neuron's /health is ~5s-cached, so this is a
+                    // coarse high-water proxy, not an instantaneous peak — but
+                    // it's the headroom signal we can read over the wire. A
+                    // flaky /health degrades to None, never a failed run.
+                    let health = self
+                        .client
+                        .fetch_health(target)
+                        .await
+                        .ok()
+                        .flatten()
+                        .map(|h| HealthAgg::from_health(&h));
+                    match result {
                         Ok(m) => {
                             let rec = self.build_record(
                                 target,
@@ -141,6 +286,8 @@ impl Sweeper {
                                 scenario.id(),
                                 scenario.prompt_size(),
                                 Ok(&m),
+                                health,
+                                None,
                             );
                             self.store.insert_run(&rec)?;
                             summary.measured += 1;
@@ -160,6 +307,8 @@ impl Sweeper {
                                 scenario.id(),
                                 scenario.prompt_size(),
                                 Err(&msg),
+                                health,
+                                None,
                             );
                             self.store.insert_run(&rec)?;
                             summary.failed += 1;
@@ -186,19 +335,14 @@ impl Sweeper {
         scenario_id: &str,
         prompt_size: u32,
         result: Result<&crate::scenario::ScenarioMetrics, &str>,
+        health: Option<HealthAgg>,
+        swap: Option<SwapTiming>,
     ) -> RunRecord {
-        let (ok, error, ttft, decode, total, prompt_tokens, completion) = match result {
-            Ok(m) => (
-                true,
-                None,
-                Some(m.ttft_s),
-                m.decode_tps,
-                Some(m.total_s),
-                m.prompt_tokens,
-                Some(m.completion_tokens),
-            ),
-            Err(e) => (false, Some(e.to_string()), None, None, None, None, None),
+        let (m, error): (Option<&ScenarioMetrics>, Option<String>) = match result {
+            Ok(m) => (Some(m), None),
+            Err(e) => (None, Some(e.to_string())),
         };
+        let ok = m.is_some();
 
         RunRecord {
             ts: chrono::Utc::now().to_rfc3339(),
@@ -230,12 +374,29 @@ impl Sweeper {
                 .unwrap_or_else(|_| "[]".to_string()),
             scenario_id: scenario_id.to_string(),
             prompt_size_approx: prompt_size,
-            prompt_tokens_actual: prompt_tokens,
+            prompt_tokens_actual: m.and_then(|m| m.prompt_tokens),
             max_tokens: self.cfg.scenarios.max_tokens,
-            ttft_s: ttft,
-            decode_tps: decode,
-            total_s: total,
-            completion_tokens: completion,
+            ttft_s: m.map(|m| m.ttft_s),
+            decode_tps: m.and_then(|m| m.decode_tps),
+            total_s: m.map(|m| m.total_s),
+            completion_tokens: m.map(|m| m.completion_tokens),
+            prefill_ms: m.and_then(|m| m.prefill_ms),
+            decode_ms: m.and_then(|m| m.decode_ms),
+            prefill_tokens: m.and_then(|m| m.prefill_tokens),
+            vram_used_mb: health.map(|h| h.vram_used_mb),
+            gpu_util_pct: health.map(|h| h.gpu_util_pct),
+            gpu_temp_c: health.map(|h| h.gpu_temp_c),
+            concurrency: m.and_then(|m| m.concurrency),
+            ttft_p95_s: m.and_then(|m| m.ttft_p95_s),
+            queue_wait_ms: m.and_then(|m| m.queue_wait_ms_median),
+            rejected: m.and_then(|m| m.rejected),
+            swap_unload_ms: swap.map(|s| s.unload_ms),
+            swap_load_ms: swap.map(|s| s.load_ms),
+            // Capability artifact (#91); score/scorer are attached later by
+            // the `score` subcommand or a future LLM-judge.
+            artifact: m.and_then(|m| m.artifact.clone()),
+            quality_score: None,
+            scorer: None,
             ok,
             error,
         }

crates/helexa-bench/tests/api.rs

@@ -0,0 +1,234 @@
+//! Read-API tests: seed a temp store, serve the router, assert JSON.
+
+use helexa_bench::api;
+use helexa_bench::store::{RunRecord, Store};
+use serde_json::Value;
+
+#[allow(clippy::too_many_arguments)]
+fn rec(
+    host: &str,
+    sha: &str,
+    build_ts: Option<&str>,
+    model: &str,
+    scenario: &str,
+    ttft: f64,
+    ok: bool,
+) -> RunRecord {
+    RunRecord {
+        ts: "2026-06-13T00:00:00Z".into(),
+        target_name: host.into(),
+        target_kind: "neuron".into(),
+        endpoint: format!("http://{host}:13131"),
+        hostname: Some(host.into()),
+        driver_version: Some("580.159".into()),
+        cuda_version: Some("13.0".into()),
+        gpus_json: Some("[]".into()),
+        git_sha: sha.into(),
+        git_sha_long: None,
+        package_version: "0.1.16".into(),
+        git_dirty: false,
+        build_timestamp: build_ts.map(|s| s.to_string()),
+        rustc_version: None,
+        profile: Some("release".into()),
+        features_json: "[\"cuda\"]".into(),
+        candle_version: Some("0.10.2".into()),
+        bench_version: "0.1.16".into(),
+        bench_sha: "deadbee".into(),
+        model_id: model.into(),
+        harness: "candle".into(),
+        capabilities_json: "[\"text\"]".into(),
+        devices_json: "[0]".into(),
+        scenario_id: scenario.into(),
+        prompt_size_approx: 128,
+        prompt_tokens_actual: Some(130),
+        max_tokens: 64,
+        ttft_s: if ok { Some(ttft) } else { None },
+        decode_tps: if ok { Some(30.0) } else { None },
+        total_s: if ok { Some(2.0) } else { None },
+        completion_tokens: if ok { Some(60) } else { None },
+        prefill_ms: if ok { Some(150) } else { None },
+        decode_ms: if ok { Some(1800) } else { None },
+        prefill_tokens: if ok { Some(130) } else { None },
+        vram_used_mb: if ok { Some(42000) } else { None },
+        gpu_util_pct: if ok { Some(85) } else { None },
+        gpu_temp_c: if ok { Some(63) } else { None },
+        concurrency: None,
+        ttft_p95_s: None,
+        queue_wait_ms: None,
+        rejected: None,
+        swap_unload_ms: None,
+        swap_load_ms: None,
+        artifact: None,
+        quality_score: None,
+        scorer: None,
+        ok,
+        error: if ok { None } else { Some("boom".into()) },
+    }
+}
+
+/// Seed a temp db, return its path.
+fn seed(tag: &str) -> String {
+    let path = std::env::temp_dir().join(format!("hb-api-{}-{tag}.sqlite", std::process::id()));
+    let _ = std::fs::remove_file(&path);
+    let p = path.to_string_lossy().to_string();
+    let store = Store::open(&p).unwrap();
+    // beast / m / chat:128 across two builds (old then new).
+    store
+        .insert_run(&rec(
+            "beast",
+            "old",
+            Some("2026-06-01T00:00:00Z"),
+            "m",
+            "chat:128",
+            0.20,
+            true,
+        ))
+        .unwrap();
+    store
+        .insert_run(&rec(
+            "beast",
+            "new",
+            Some("2026-06-10T00:00:00Z"),
+            "m",
+            "chat:128",
+            0.10,
+            true,
+        ))
+        .unwrap();
+    store
+        .insert_run(&rec(
+            "beast",
+            "new",
+            Some("2026-06-10T00:00:00Z"),
+            "m",
+            "chat:128",
+            0.12,
+            true,
+        ))
+        .unwrap();
+    // a failed row (must not count in series/summary medians)
+    store
+        .insert_run(&rec(
+            "beast",
+            "new",
+            Some("2026-06-10T00:00:00Z"),
+            "m",
+            "chat:128",
+            0.0,
+            false,
+        ))
+        .unwrap();
+    // a different host for the runs filter
+    store
+        .insert_run(&rec(
+            "benjy",
+            "new",
+            Some("2026-06-10T00:00:00Z"),
+            "n",
+            "chat:128",
+            0.15,
+            true,
+        ))
+        .unwrap();
+    p
+}
+
+async fn spawn(db: &str) -> String {
+    let state = api::open_state(db).unwrap();
+    let app = api::api_routes(state);
+    let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    tokio::spawn(async move {
+        axum::serve(listener, app).await.unwrap();
+    });
+    format!("http://{addr}")
+}
+
+async fn get(base: &str, path: &str) -> Value {
+    reqwest::get(format!("{base}{path}"))
+        .await
+        .unwrap()
+        .json()
+        .await
+        .unwrap()
+}
+
+#[tokio::test]
+async fn health_reports_run_count() {
+    let base = spawn(&seed("health")).await;
+    let v = get(&base, "/api/health").await;
+    assert_eq!(v["status"], "ok");
+    assert_eq!(v["run_count"], 5);
+}
+
+#[tokio::test]
+async fn dimensions_lists_distinct_values_and_builds_chronologically() {
+    let base = spawn(&seed("dims")).await;
+    let v = get(&base, "/api/dimensions").await;
+    let hosts: Vec<&str> = v["hosts"]
+        .as_array()
+        .unwrap()
+        .iter()
+        .map(|x| x.as_str().unwrap())
+        .collect();
+    assert_eq!(hosts, vec!["beast", "benjy"]);
+    assert_eq!(v["models"].as_array().unwrap().len(), 2);
+    // builds ordered by earliest build_timestamp: old before new
+    let builds = v["builds"].as_array().unwrap();
+    assert_eq!(builds[0]["git_sha"], "old");
+    assert_eq!(builds[1]["git_sha"], "new");
+}
+
+#[tokio::test]
+async fn summary_uses_latest_sha_and_ignores_failures() {
+    let base = spawn(&seed("summary")).await;
+    let v = get(&base, "/api/summary").await;
+    let rows = v.as_array().unwrap();
+    let beast = rows
+        .iter()
+        .find(|r| r["target_name"] == "beast" && r["scenario_id"] == "chat:128")
+        .unwrap();
+    assert_eq!(beast["git_sha"], "new");
+    assert_eq!(beast["samples"], 2); // two ok rows on "new"; failure excluded
+    // median of 0.10 and 0.12
+    assert!((beast["ttft_s_median"].as_f64().unwrap() - 0.11).abs() < 1e-9);
+}
+
+#[tokio::test]
+async fn series_is_chronological_per_build() {
+    let base = spawn(&seed("series")).await;
+    let v = get(&base, "/api/series?host=beast&model=m&scenario=chat:128").await;
+    let pts = v.as_array().unwrap();
+    assert_eq!(pts.len(), 2);
+    assert_eq!(pts[0]["git_sha"], "old");
+    assert_eq!(pts[1]["git_sha"], "new");
+    assert_eq!(pts[0]["samples"], 1);
+    assert_eq!(pts[1]["samples"], 2);
+}
+
+#[tokio::test]
+async fn series_resolves_host_when_omitted() {
+    // The public UI selects by model alone; the store resolves the host.
+    let base = spawn(&seed("series-nohost")).await;
+    let v = get(&base, "/api/series?model=m&scenario=chat:128").await;
+    let pts = v.as_array().unwrap();
+    assert_eq!(pts.len(), 2);
+    assert_eq!(pts[0]["git_sha"], "old");
+    assert_eq!(pts[1]["git_sha"], "new");
+}
+
+#[tokio::test]
+async fn runs_filters_by_host() {
+    let base = spawn(&seed("runs")).await;
+    let all = get(&base, "/api/runs").await;
+    assert_eq!(all.as_array().unwrap().len(), 5);
+    let beast = get(&base, "/api/runs?host=beast").await;
+    let rows = beast.as_array().unwrap();
+    assert_eq!(rows.len(), 4);
+    assert!(rows.iter().all(|r| r["host"] == "beast"));
+    // failed row carries its error + ok=false
+    assert!(
+        rows.iter()
+            .any(|r| r["ok"] == false && r["error"] == "boom")
+    );
+}

crates/helexa-bench/tests/sweep_integration.rs

@@ -89,7 +89,11 @@ fn config_for(endpoint: String, db_path: String) -> BenchConfig {
         scenarios: ScenarioConfig {
             prompt_sizes: vec![128], // single scenario keeps assertions simple
             max_tokens: 16,
+            concurrency_levels: Vec::new(),
+            concurrency_prompt_tokens: 512,
+            capability_probes: Vec::new(),
         },
+        api: Default::default(),
         targets: vec![TargetConfig {
             name: "mock".into(),
             kind: TargetKind::Neuron,

crates/helexa-router/Cargo.toml

@@ -0,0 +1,41 @@
+[package]
+name = "helexa-router"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+repository.workspace = true
+
+[[bin]]
+name = "helexa-router"
+path = "src/main.rs"
+
+[lib]
+name = "helexa_router"
+path = "src/lib.rs"
+
+[dependencies]
+cortex-core = { workspace = true }
+helexa-stream = { path = "../helexa-stream" }
+
+tokio = { workspace = true }
+axum = { workspace = true }
+tower-http = { workspace = true }
+reqwest = { workspace = true }
+serde = { workspace = true }
+serde_json = { workspace = true }
+figment = { workspace = true }
+anyhow = { workspace = true }
+thiserror = { workspace = true }
+clap = { workspace = true }
+tracing = { workspace = true }
+tracing-subscriber = { workspace = true }
+chrono = { workspace = true }
+
+[dev-dependencies]
+# Jail (isolated cwd + env) for config tests.
+figment = { workspace = true, features = ["test"] }
+# Self-signed cert generation + a minimal HTTPS server for the outbound
+# TLS-pinning tests (#74).
+rcgen = "0.13"
+rustls = "0.23"
+tokio-rustls = "0.26"

crates/helexa-router/src/catalogue.rs

@@ -0,0 +1,260 @@
+//! Federation catalogue (#75) — the router's aggregate `/v1/models`.
+//!
+//! Presents the **deduped union** of every reachable cortex's `/v1/models`
+//! as the router's own catalogue, so an opencode client doing discovery
+//! against the router resolves the whole federation without knowing about
+//! operators or cortexes (resolves #61's "Router/discovery contract").
+//!
+//! Re-tiering: the fractal design is neuron ← cortex ← router. At the
+//! router tier the "nodes" are **cortexes**, so the merged entry's
+//! `feasible_on` / `locations` are rewritten to **operator names**, not the
+//! neuron names a cortex reports. That keeps the federation view honest
+//! ("served by these operators") without leaking each operator's internal
+//! topology (neuron names, per-device VRAM) to end users.
+//!
+//! Conflict resolution when operators advertise the same model with
+//! different enrichment:
+//! - **`limit`** → the *tightest* (smallest `context`), so a client never
+//!   overflows the most-constrained operator that might serve it (same rule
+//!   cortex uses across its neurons).
+//! - **`cost`** → the *cheapest* (lowest input, then output), the
+//!   federation "from" price. Richer policy (a range, region/price-aware
+//!   selection) couples to #68 and is left as a follow-up.
+
+use crate::state::{CortexTopology, entry_feasible};
+use cortex_core::harness::{ModelCost, ModelLimit};
+use cortex_core::node::{CortexModelEntry, ModelLocation, ModelStatus};
+use std::collections::HashMap;
+
+/// Build the federation catalogue: the deduped union of every reachable
+/// cortex's serveable models, merged across operators and sorted by id.
+pub fn aggregate_models(topology: &HashMap<String, CortexTopology>) -> Vec<CortexModelEntry> {
+    // Iterate cortexes in name order so `feasible_on` / `locations` and the
+    // limit/cost tie-breaks are deterministic regardless of map ordering.
+    let mut cortexes: Vec<(&String, &CortexTopology)> = topology.iter().collect();
+    cortexes.sort_by(|a, b| a.0.cmp(b.0));
+
+    let mut merged: HashMap<String, CortexModelEntry> = HashMap::new();
+    for (cortex_name, t) in cortexes {
+        if !t.reachable {
+            continue;
+        }
+        for entry in t.models.values() {
+            // Only surface models the cortex can actually serve — a
+            // catalogue-only entry no neuron can host shouldn't appear in
+            // the federation view.
+            if !entry_feasible(entry) {
+                continue;
+            }
+            merged
+                .entry(entry.id.clone())
+                .and_modify(|acc| merge_into(acc, cortex_name, entry))
+                .or_insert_with(|| router_entry(cortex_name, entry));
+        }
+    }
+
+    let mut out: Vec<CortexModelEntry> = merged.into_values().collect();
+    out.sort_by(|a, b| a.id.cmp(&b.id));
+    // Re-derive the flat ecosystem fields (#78) from the merged (tightest)
+    // limit — the values deserialized from each cortex are per-operator and
+    // may not match the federation-wide merge.
+    for e in &mut out {
+        e.sync_flat_limit();
+    }
+    out
+}
+
+/// Seed a federation entry from the first cortex that serves the model,
+/// re-tiering `feasible_on` / `locations` to the operator name.
+fn router_entry(cortex: &str, e: &CortexModelEntry) -> CortexModelEntry {
+    CortexModelEntry {
+        id: e.id.clone(),
+        object: "model".into(),
+        created: e.created,
+        owned_by: e.owned_by.clone(),
+        loaded: e.loaded,
+        feasible_on: vec![cortex.to_string()],
+        locations: loaded_location(cortex, e),
+        capabilities: e.capabilities.clone(),
+        limit: e.limit.clone(),
+        cost: e.cost.clone(),
+        tool_call: e.tool_call,
+        reasoning: e.reasoning,
+        // Derived from `limit` by the final sync pass in aggregate_models.
+        max_model_len: None,
+        max_input_tokens: None,
+        max_output_tokens: None,
+    }
+}
+
+/// Fold another cortex's view of the same model into the merged entry.
+fn merge_into(acc: &mut CortexModelEntry, cortex: &str, e: &CortexModelEntry) {
+    acc.loaded |= e.loaded;
+    acc.feasible_on.push(cortex.to_string());
+    acc.locations.extend(loaded_location(cortex, e));
+    for cap in &e.capabilities {
+        if !acc.capabilities.contains(cap) {
+            acc.capabilities.push(cap.clone());
+        }
+    }
+    acc.tool_call |= e.tool_call;
+    acc.reasoning |= e.reasoning;
+    acc.limit = tightest_limit(acc.limit.take(), e.limit.clone());
+    acc.cost = cheapest_cost(acc.cost.take(), e.cost.clone());
+}
+
+/// A single cortex-tier location when the model is loaded at that operator;
+/// empty when only cold-loadable. Neuron-level VRAM is deliberately dropped.
+fn loaded_location(cortex: &str, e: &CortexModelEntry) -> Vec<ModelLocation> {
+    if e.loaded {
+        vec![ModelLocation {
+            node: cortex.to_string(),
+            status: ModelStatus::Loaded,
+            vram_estimate_mb: None,
+        }]
+    } else {
+        Vec::new()
+    }
+}
+
+/// Smaller `context` wins — never advertise more headroom than the
+/// most-constrained operator can honour.
+fn tightest_limit(a: Option<ModelLimit>, b: Option<ModelLimit>) -> Option<ModelLimit> {
+    match (a, b) {
+        (None, x) | (x, None) => x,
+        (Some(a), Some(b)) => Some(if b.context < a.context { b } else { a }),
+    }
+}
+
+/// Cheapest by (input, output) price — the federation "from" price.
+fn cheapest_cost(a: Option<ModelCost>, b: Option<ModelCost>) -> Option<ModelCost> {
+    match (a, b) {
+        (None, x) | (x, None) => x,
+        (Some(a), Some(b)) => Some(if (b.input, b.output) < (a.input, a.output) {
+            b
+        } else {
+            a
+        }),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::state::CortexTopology;
+
+    fn entry(id: &str, loaded: bool, feasible: bool) -> CortexModelEntry {
+        CortexModelEntry {
+            id: id.into(),
+            object: "model".into(),
+            created: 0,
+            owned_by: "helexa".into(),
+            loaded,
+            feasible_on: if feasible || loaded {
+                vec!["some-neuron".into()]
+            } else {
+                vec![]
+            },
+            locations: vec![],
+            capabilities: vec![],
+            limit: None,
+            cost: None,
+            tool_call: false,
+            reasoning: false,
+            max_model_len: None,
+            max_input_tokens: None,
+            max_output_tokens: None,
+        }
+    }
+
+    fn cortex(reachable: bool, entries: Vec<CortexModelEntry>) -> CortexTopology {
+        CortexTopology {
+            reachable,
+            consecutive_failures: 0,
+            last_poll: None,
+            healthy_nodes: 1,
+            total_nodes: 1,
+            models: entries.into_iter().map(|e| (e.id.clone(), e)).collect(),
+        }
+    }
+
+    #[test]
+    fn dedupes_and_merges_availability_across_cortexes() {
+        let mut topo = HashMap::new();
+        // c-a: model loaded. c-b: same model only cold-loadable.
+        topo.insert("c-a".into(), cortex(true, vec![entry("m", true, true)]));
+        topo.insert("c-b".into(), cortex(true, vec![entry("m", false, true)]));
+
+        let out = aggregate_models(&topo);
+        assert_eq!(out.len(), 1, "duplicate model id collapses to one");
+        let m = &out[0];
+        assert!(m.loaded, "loaded somewhere → loaded");
+        // feasible_on re-tiered to operator names, both present, sorted.
+        assert_eq!(m.feasible_on, vec!["c-a".to_string(), "c-b".to_string()]);
+        // Only the loaded operator contributes a location, named by operator.
+        assert_eq!(m.locations.len(), 1);
+        assert_eq!(m.locations[0].node, "c-a");
+        assert_eq!(m.locations[0].vram_estimate_mb, None);
+    }
+
+    #[test]
+    fn unreachable_cortex_is_excluded() {
+        let mut topo = HashMap::new();
+        topo.insert("up".into(), cortex(true, vec![entry("m", true, true)]));
+        topo.insert(
+            "down".into(),
+            cortex(false, vec![entry("other", true, true)]),
+        );
+        let out = aggregate_models(&topo);
+        assert_eq!(out.len(), 1);
+        assert_eq!(out[0].id, "m");
+    }
+
+    #[test]
+    fn catalogue_only_infeasible_entries_are_hidden() {
+        let mut topo = HashMap::new();
+        topo.insert("c".into(), cortex(true, vec![entry("ghost", false, false)]));
+        assert!(aggregate_models(&topo).is_empty());
+    }
+
+    #[test]
+    fn preserves_tightest_limit_and_cheapest_cost() {
+        let mut a = entry("m", true, true);
+        a.limit = Some(ModelLimit {
+            context: 32_768,
+            input: None,
+            output: 4096,
+        });
+        a.cost = Some(ModelCost {
+            input: 0.50,
+            output: 1.50,
+            cache_read: None,
+            cache_write: None,
+        });
+        let mut b = entry("m", true, true);
+        b.limit = Some(ModelLimit {
+            context: 16_384, // tighter
+            input: None,
+            output: 4096,
+        });
+        b.cost = Some(ModelCost {
+            input: 0.20, // cheaper
+            output: 0.80,
+            cache_read: None,
+            cache_write: None,
+        });
+
+        let mut topo = HashMap::new();
+        topo.insert("c-a".into(), cortex(true, vec![a]));
+        topo.insert("c-b".into(), cortex(true, vec![b]));
+
+        let out = aggregate_models(&topo);
+        assert_eq!(out.len(), 1);
+        assert_eq!(out[0].limit.as_ref().unwrap().context, 16_384);
+        assert_eq!(out[0].cost.as_ref().unwrap().input, 0.20);
+        // Flat #78 fields re-derived from the merged (tightest) limit.
+        assert_eq!(out[0].max_model_len, Some(16_384));
+        assert_eq!(out[0].max_input_tokens, None);
+        assert_eq!(out[0].max_output_tokens, Some(4096));
+    }
+}

crates/helexa-router/src/config.rs

@@ -0,0 +1,100 @@
+use figment::{
+    Figment,
+    providers::{Env, Format, Toml},
+};
+use serde::{Deserialize, Serialize};
+use std::path::Path;
+
+/// Top-level `helexa-router` configuration.
+///
+/// Loaded from TOML with `HELEXA_ROUTER_`-prefixed env overrides (using
+/// `__` as the nesting separator), matching the cortex/neuron convention.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct RouterConfig {
+    pub router: RouterSettings,
+    /// Downstream cortex endpoints the router can dispatch to. The skeleton
+    /// (#70) only loads these; capacity/catalogue polling (#72) and
+    /// capacity-aware dispatch (#73) consume them later.
+    #[serde(default)]
+    pub cortexes: Vec<CortexEndpoint>,
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct RouterSettings {
+    /// Address to listen on for the inbound API (e.g. "0.0.0.0:8088").
+    ///
+    /// Plaintext only — operator/edge nginx terminates client TLS in front
+    /// of the router (see #69's TLS posture). The router never owns an
+    /// inbound TLS listener.
+    pub listen: String,
+    /// How often (seconds) the background poller refreshes each cortex's
+    /// health + `/v1/models` topology (#72). Defaults to 10s, matching the
+    /// cortex↔neuron poll cadence one tier down.
+    #[serde(default = "default_poll_interval_secs")]
+    pub poll_interval_secs: u64,
+    /// This router instance's region (e.g. "eu-west"). When set, dispatch
+    /// (#73) prefers cortexes whose `region` matches, before falling back to
+    /// any feasible cortex. `None` → no geo affinity.
+    #[serde(default)]
+    pub region: Option<String>,
+}
+
+fn default_poll_interval_secs() -> u64 {
+    10
+}
+
+/// One downstream cortex the router may proxy to. The router verifies the
+/// cortex's outbound TLS cert (#74) and routes on capacity (#73); it holds
+/// no entitlement logic of its own and forwards the client bearer verbatim.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct CortexEndpoint {
+    /// Human-readable label (e.g. "lair-cafe").
+    pub name: String,
+    /// Base URL of the cortex gateway (e.g. "https://cortex.example.com").
+    pub endpoint: String,
+    /// Optional region tag (e.g. "eu-west") for geo affinity in dispatch
+    /// (#73). `None` → no region preference applies to this cortex.
+    #[serde(default)]
+    pub region: Option<String>,
+    /// Path to a PEM trust anchor that **enrols** this cortex (#74): the
+    /// expected CA (or self-signed cert) the cortex's TLS cert must chain
+    /// to. When set on an `https://` endpoint, the router builds a client
+    /// that trusts **only** this anchor (platform roots disabled), so the
+    /// outbound router→cortex hop — which carries the client's bearer —
+    /// reaches a cert the router was told to expect, and a rogue endpoint
+    /// presenting any other (even publicly-valid) cert is rejected at the
+    /// TLS handshake. A rejected handshake surfaces as a connection error,
+    /// which the poller (#72) already treats as unreachable → excluded.
+    ///
+    /// `None` → standard platform-root validation (use for cortexes behind
+    /// a publicly-trusted cert, or plaintext `http://` on a private network
+    /// where the WireGuard mesh is the trust boundary).
+    #[serde(default)]
+    pub tls_ca: Option<String>,
+}
+
+impl RouterConfig {
+    /// Load configuration from a TOML file, with environment variable
+    /// overrides prefixed with `HELEXA_ROUTER_` and `__` as the separator
+    /// (e.g. `HELEXA_ROUTER_ROUTER__LISTEN=0.0.0.0:8088`).
+    pub fn load(path: impl AsRef<Path>) -> Result<Self, Box<figment::Error>> {
+        Figment::new()
+            .merge(Toml::file(path))
+            .merge(Env::prefixed("HELEXA_ROUTER_").split("__"))
+            .extract()
+            .map_err(Box::new)
+    }
+}
+
+impl Default for RouterConfig {
+    fn default() -> Self {
+        Self {
+            router: RouterSettings {
+                listen: "0.0.0.0:8088".into(),
+                poll_interval_secs: default_poll_interval_secs(),
+                region: None,
+            },
+            cortexes: vec![],
+        }
+    }
+}

crates/helexa-router/src/dispatch.rs

@@ -0,0 +1,221 @@
+//! Capacity-aware dispatch (#73) — the router's data path.
+//!
+//! Given an inbound request's `model`, pick a reachable cortex that can
+//! serve it (preferring warm/loaded, region-affine, higher-headroom),
+//! forward the client's bearer **unchanged** (auth stays at cortex), and
+//! stream the response back verbatim via the shared [`helexa_stream`]
+//! module. Cortex's #63-shaped rejections (`429 rate_limit_exceeded`,
+//! `400 context_length_exceeded`, …) pass through untouched. Transport
+//! failures fail over to the next feasible cortex; a genuine HTTP response —
+//! any status — is returned as-is and never retried away.
+//!
+//! The router holds **no entitlement logic**: it routes on capacity, not
+//! budget.
+
+use crate::config::CortexEndpoint;
+use crate::error::envelope_response;
+use crate::state::RouterState;
+use axum::body::Bytes;
+use axum::http::HeaderMap;
+use axum::response::Response;
+use cortex_core::error_envelope::OpenAiError;
+use helexa_stream::{ChunkObserver, StreamError};
+use std::cmp::Reverse;
+use std::collections::HashMap;
+
+/// Retry-After hint (seconds) on the router's own transient rejections.
+const RETRY_AFTER_SECS: u64 = 5;
+
+/// Outcome of choosing where to send a request.
+#[derive(Debug, PartialEq, Eq)]
+pub enum Selection {
+    /// Feasible reachable cortexes, best-first (failover order).
+    Candidates(Vec<CortexEndpoint>),
+    /// Some cortex knows the model but none are reachable right now → 503.
+    NoReachableCapacity,
+    /// No configured cortex serves the model at all → 404.
+    UnknownModel,
+}
+
+/// Rank the reachable cortexes that can serve `model`, best-first.
+///
+/// Ordering (each a tie-break for the next): loaded/warm before cold-loadable
+/// · region match before not · more healthy nodes before fewer · name for
+/// determinism.
+pub async fn select_cortexes(state: &RouterState, model: &str) -> Selection {
+    let topo = state.topology.read().await;
+    let by_name: HashMap<&str, &CortexEndpoint> = state
+        .cortexes
+        .iter()
+        .map(|c| (c.name.as_str(), c))
+        .collect();
+
+    let mut ranked: Vec<Ranked> = Vec::new();
+    let mut known_anywhere = false;
+
+    for (name, t) in topo.iter() {
+        let Some(entry) = t.models.get(model) else {
+            continue;
+        };
+        if !crate::state::entry_feasible(entry) {
+            continue;
+        }
+        // Known even via an unreachable cortex's last-good poll — lets us
+        // tell "temporarily down" (503) from "nobody serves it" (404).
+        known_anywhere = true;
+        if !t.reachable {
+            continue;
+        }
+        let Some(ep) = by_name.get(name.as_str()) else {
+            continue;
+        };
+        let region_match = match (&state.region, &ep.region) {
+            (Some(r), Some(cr)) => r == cr,
+            _ => false,
+        };
+        ranked.push(Ranked {
+            loaded: entry.loaded,
+            region_match,
+            healthy_nodes: t.healthy_nodes,
+            endpoint: (*ep).clone(),
+        });
+    }
+
+    if ranked.is_empty() {
+        return if known_anywhere {
+            Selection::NoReachableCapacity
+        } else {
+            Selection::UnknownModel
+        };
+    }
+
+    ranked.sort_by(|a, b| {
+        // false < true, so negate the "good" booleans to sort good first.
+        (
+            !a.loaded,
+            !a.region_match,
+            Reverse(a.healthy_nodes),
+            &a.endpoint.name,
+        )
+            .cmp(&(
+                !b.loaded,
+                !b.region_match,
+                Reverse(b.healthy_nodes),
+                &b.endpoint.name,
+            ))
+    });
+
+    Selection::Candidates(ranked.into_iter().map(|r| r.endpoint).collect())
+}
+
+struct Ranked {
+    loaded: bool,
+    region_match: bool,
+    healthy_nodes: u32,
+    endpoint: CortexEndpoint,
+}
+
+/// Proxy an inbound inference request to a capacity-bearing cortex.
+///
+/// `path` is the inference path to forward to (same on the cortex, e.g.
+/// `/v1/chat/completions`). The body is parsed only to extract `model`.
+pub async fn dispatch(
+    state: &RouterState,
+    path: &str,
+    headers: HeaderMap,
+    body: Bytes,
+) -> Response {
+    let Some(model) = extract_model(&body) else {
+        return envelope_response(OpenAiError::new(
+            400,
+            "invalid_request_error",
+            "missing_model_field",
+            "missing 'model' field in request body",
+        ));
+    };
+
+    let candidates = match select_cortexes(state, &model).await {
+        Selection::Candidates(c) => c,
+        Selection::UnknownModel => {
+            return envelope_response(
+                OpenAiError::new(
+                    404,
+                    "invalid_request_error",
+                    "model_not_found",
+                    format!("no operator serves model '{model}'"),
+                )
+                .with_param("model"),
+            );
+        }
+        Selection::NoReachableCapacity => {
+            return envelope_response(OpenAiError::service_unavailable(
+                format!("model '{model}' is temporarily unavailable on all operators"),
+                Some(RETRY_AFTER_SECS),
+            ));
+        }
+    };
+
+    // Try candidates in order, failing over only on transport errors. A
+    // genuine HTTP response (any status — including cortex's #63 429/400)
+    // is returned verbatim and never retried away.
+    for ep in &candidates {
+        // A candidate whose pinned TLS client failed to build (#74) is
+        // disabled — skip it and fail over, same as an unreachable cortex.
+        let Some(client) = state.client_for(&ep.name) else {
+            tracing::warn!(cortex = %ep.name, "no TLS client (disabled); skipping candidate");
+            continue;
+        };
+        let url = format!("{}{}", ep.endpoint, path);
+        tracing::info!(cortex = %ep.name, url = %url, model = %model, "dispatching");
+        match helexa_stream::forward_streaming(
+            client,
+            &url,
+            headers.clone(),
+            body.clone(),
+            NoopObserver,
+        )
+        .await
+        {
+            Ok(resp) => return resp,
+            Err(StreamError::Upstream(e)) => {
+                tracing::warn!(
+                    cortex = %ep.name,
+                    url = %url,
+                    error = %e,
+                    "cortex unreachable; failing over"
+                );
+                continue;
+            }
+            Err(StreamError::ResponseBuild(msg)) => {
+                tracing::error!(cortex = %ep.name, error = %msg, "failed to build proxied response");
+                return envelope_response(OpenAiError::without_code(
+                    500,
+                    "api_error",
+                    "failed to build proxied response",
+                ));
+            }
+        }
+    }
+
+    // Every feasible cortex failed to connect.
+    tracing::warn!(model = %model, tried = candidates.len(), "all feasible operators unreachable");
+    envelope_response(OpenAiError::service_unavailable(
+        format!("all operators able to serve '{model}' are unreachable"),
+        Some(RETRY_AFTER_SECS),
+    ))
+}
+
+/// Pull the `model` field out of a request body without re-serialising it.
+fn extract_model(body: &Bytes) -> Option<String> {
+    let v: serde_json::Value = serde_json::from_slice(body).ok()?;
+    v.get("model")?.as_str().map(str::to_string)
+}
+
+/// The router proxies bytes verbatim and keeps no per-request policy, so it
+/// needs no observation hooks. (Token metrics/metering stay at cortex.)
+struct NoopObserver;
+
+impl ChunkObserver for NoopObserver {
+    fn observe(&mut self, _chunk: &[u8]) {}
+    fn finish(&mut self) {}
+}

crates/helexa-router/src/error.rs

@@ -0,0 +1,27 @@
+//! Router adapter from the shared, axum-agnostic
+//! [`cortex_core::error_envelope::OpenAiError`] (#60/#63) to an axum
+//! [`Response`], setting `Retry-After` when the envelope carries one.
+//!
+//! cortex-core owns the envelope shape; this is the only place the router
+//! crosses from that data into axum. Mirrors cortex-gateway's adapter so
+//! the router's own rejections (no feasible operator, all unreachable) are
+//! the same #63-shaped envelopes clients already understand — distinct from
+//! cortex's rejections, which the router proxies through verbatim.
+
+use axum::http::{HeaderValue, StatusCode, header};
+use axum::response::{IntoResponse, Json, Response};
+use cortex_core::error_envelope::OpenAiError;
+
+/// Render an [`OpenAiError`] as an axum response (status + JSON envelope +
+/// optional `Retry-After`).
+pub fn envelope_response(err: OpenAiError) -> Response {
+    let status = StatusCode::from_u16(err.status).unwrap_or(StatusCode::INTERNAL_SERVER_ERROR);
+    let retry_after = err.retry_after_secs;
+    let mut response = (status, Json(err.body())).into_response();
+    if let Some(secs) = retry_after
+        && let Ok(value) = HeaderValue::from_str(&secs.to_string())
+    {
+        response.headers_mut().insert(header::RETRY_AFTER, value);
+    }
+    response
+}

crates/helexa-router/src/handlers.rs

@@ -0,0 +1,89 @@
+use crate::state::RouterState;
+use crate::{catalogue, dispatch};
+use axum::body::Bytes;
+use axum::http::HeaderMap;
+use axum::response::Response;
+use axum::{Json, Router, extract::State, routing::get, routing::post};
+use serde_json::{Value, json};
+use std::sync::Arc;
+
+/// Routes served by the router. Inference paths are capacity-aware-dispatched
+/// (#73) to a downstream cortex; `/health` and a stub `/v1/models` are local.
+pub fn api_routes() -> Router<Arc<RouterState>> {
+    Router::new()
+        .route("/v1/chat/completions", post(chat_completions))
+        .route("/v1/completions", post(completions))
+        .route("/v1/responses", post(responses))
+        .route("/v1/messages", post(messages))
+        .route("/v1/models", get(list_models))
+        .route("/health", get(health))
+        .route("/", get(health))
+}
+
+// ── Inference paths — forwarded verbatim to a chosen cortex ──────────
+//
+// Each handler dispatches to the same path on a capacity-bearing cortex.
+// The body is parsed only to read `model`; the bearer and bytes are
+// forwarded unchanged, and the SSE response streams back verbatim.
+
+async fn chat_completions(
+    State(state): State<Arc<RouterState>>,
+    headers: HeaderMap,
+    body: Bytes,
+) -> Response {
+    dispatch::dispatch(&state, "/v1/chat/completions", headers, body).await
+}
+
+async fn completions(
+    State(state): State<Arc<RouterState>>,
+    headers: HeaderMap,
+    body: Bytes,
+) -> Response {
+    dispatch::dispatch(&state, "/v1/completions", headers, body).await
+}
+
+async fn responses(
+    State(state): State<Arc<RouterState>>,
+    headers: HeaderMap,
+    body: Bytes,
+) -> Response {
+    dispatch::dispatch(&state, "/v1/responses", headers, body).await
+}
+
+async fn messages(
+    State(state): State<Arc<RouterState>>,
+    headers: HeaderMap,
+    body: Bytes,
+) -> Response {
+    dispatch::dispatch(&state, "/v1/messages", headers, body).await
+}
+
+/// `GET /health` — router liveness plus a summary of downstream cortex
+/// reachability from the topology poller (#72). `status` reflects the
+/// router process itself (always `ok` if it answers); downstream health is
+/// the informational `cortexes` block, so a fully-degraded fleet doesn't
+/// make the router look dead to its own liveness probe.
+async fn health(State(state): State<Arc<RouterState>>) -> Json<Value> {
+    let topo = state.topology.read().await;
+    let reachable = topo.values().filter(|t| t.reachable).count();
+    Json(json!({
+        "status": "ok",
+        "cortexes": {
+            "configured": state.cortexes.len(),
+            "reachable": reachable,
+        }
+    }))
+}
+
+/// `GET /v1/models` — the federation catalogue (#75): the deduped union of
+/// every reachable cortex's `/v1/models`, so a client doing discovery
+/// against the router resolves the whole federation without knowing about
+/// operators or cortexes.
+async fn list_models(State(state): State<Arc<RouterState>>) -> Json<Value> {
+    let topo = state.topology.read().await;
+    let data: Vec<Value> = catalogue::aggregate_models(&topo)
+        .iter()
+        .map(|e| json!(e))
+        .collect();
+    Json(json!({ "object": "list", "data": data }))
+}

crates/helexa-router/src/lib.rs

@@ -0,0 +1,60 @@
+//! helexa-router — public multi-operator ingress proxy (router.helexa.ai).
+//!
+//! The router is the data-plane *ingress* tier: a geo-distributed,
+//! capacity-aware, OpenAI/Anthropic-compatible reverse proxy in front of
+//! many operator-run cortexes ("cortex-of-cortexes"). End users configure
+//! one `baseURL` and the router forwards their request to a cortex with
+//! capacity, proxying #63-shaped rejections back verbatim.
+//!
+//! It holds **zero entitlement logic** — auth/budget stays at cortex
+//! (epic #47); the router forwards the client bearer unchanged and routes
+//! on capacity (epic #69). A background [`poller`] keeps a live
+//! per-cortex topology (#72) that the dispatcher (#73) will route on.
+
+pub mod catalogue;
+pub mod config;
+pub mod dispatch;
+pub mod error;
+pub mod handlers;
+pub mod poller;
+pub mod state;
+
+use anyhow::Result;
+use config::RouterConfig;
+use std::sync::Arc;
+use tower_http::cors::CorsLayer;
+use tower_http::trace::TraceLayer;
+
+/// Build the axum application: handlers + CORS + tracing. No auth layer —
+/// the router asserts no identity of its own and forwards the client bearer
+/// to the downstream cortex, which authenticates it (#69).
+pub fn build_app(state: Arc<state::RouterState>) -> axum::Router {
+    axum::Router::new()
+        .merge(handlers::api_routes())
+        .layer(CorsLayer::permissive())
+        .layer(TraceLayer::new_for_http())
+        .with_state(state)
+}
+
+/// Start the router: build state from config and bind the plaintext HTTP
+/// listener. TLS is terminated by edge nginx ahead of this process.
+pub async fn run(config: RouterConfig) -> Result<()> {
+    let state = Arc::new(state::RouterState::from_config(&config));
+
+    // Background topology poller (#72): refresh each cortex's health +
+    // catalogue so routing decisions see live capacity.
+    let poller_state = Arc::clone(&state);
+    tokio::spawn(async move {
+        poller::poll_loop(poller_state).await;
+    });
+
+    let app = build_app(Arc::clone(&state));
+
+    let listen_addr = config.router.listen.parse::<std::net::SocketAddr>()?;
+    tracing::info!("helexa-router listening on {listen_addr}");
+
+    let listener = tokio::net::TcpListener::bind(listen_addr).await?;
+    axum::serve(listener, app).await?;
+
+    Ok(())
+}

crates/helexa-router/src/main.rs

@@ -0,0 +1,52 @@
+use anyhow::Result;
+use clap::{Parser, Subcommand};
+use helexa_router::config::RouterConfig;
+use tracing_subscriber::EnvFilter;
+
+#[derive(Parser)]
+#[command(name = "helexa-router")]
+#[command(about = "Public multi-operator ingress proxy for helexa")]
+#[command(version)]
+struct Cli {
+    #[command(subcommand)]
+    command: Commands,
+}
+
+#[derive(Subcommand)]
+enum Commands {
+    /// Start the router server.
+    Serve {
+        /// Path to the router config file.
+        #[arg(short, long, default_value = "helexa-router.toml")]
+        config: String,
+    },
+}
+
+#[tokio::main]
+async fn main() -> Result<()> {
+    tracing_subscriber::fmt()
+        .with_env_filter(
+            EnvFilter::try_from_default_env()
+                .unwrap_or_else(|_| EnvFilter::new("info,helexa_router=debug")),
+        )
+        .init();
+
+    let cli = Cli::parse();
+
+    match cli.command {
+        Commands::Serve { config } => {
+            let cfg = RouterConfig::load(&config)
+                .map_err(|e| anyhow::anyhow!("failed to load config from '{config}': {e}"))?;
+
+            tracing::info!(
+                cortexes = cfg.cortexes.len(),
+                listen = %cfg.router.listen,
+                "starting helexa-router"
+            );
+
+            helexa_router::run(cfg).await?;
+        }
+    }
+
+    Ok(())
+}

crates/helexa-router/src/poller.rs

@@ -0,0 +1,150 @@
+//! Background poller that refreshes the multi-operator topology (#72).
+//!
+//! The same pattern as cortex↔neuron, one tier up: periodically poll each
+//! configured cortex's `GET /v1/models` (catalogue × topology feasibility +
+//! loaded state) and `GET /health` (coarse node-health/load), building the
+//! live map the dispatcher (#73) routes on. An unreachable or erroring
+//! cortex is debounced over [`POLL_FAILURE_THRESHOLD`] consecutive misses,
+//! then flipped unhealthy and excluded from routing; it recovers on the
+//! next successful poll.
+
+use crate::state::RouterState;
+use chrono::Utc;
+use cortex_core::node::CortexModelEntry;
+use serde::Deserialize;
+use std::time::Duration;
+
+/// Per-cortex HTTP timeout for each poll request.
+const POLL_TIMEOUT: Duration = Duration::from_secs(5);
+
+/// Consecutive failed polls before a cortex is marked unreachable. Mirrors
+/// cortex's neuron-poll debounce: a single blip (a busy cortex briefly slow
+/// to answer) can't yank it — and all its models — out of routing.
+pub const POLL_FAILURE_THRESHOLD: u32 = 3;
+
+/// cortex's `/v1/models` envelope — `{ "object": "list", "data": [...] }`.
+#[derive(Debug, Deserialize)]
+struct ModelsEnvelope {
+    #[serde(default)]
+    data: Vec<CortexModelEntry>,
+}
+
+/// The subset of cortex's `/health` the router reads.
+#[derive(Debug, Deserialize)]
+struct CortexHealth {
+    nodes: CortexHealthNodes,
+}
+
+#[derive(Debug, Deserialize)]
+struct CortexHealthNodes {
+    healthy: u32,
+    total: u32,
+}
+
+/// Run forever, polling all cortexes on the configured interval.
+pub async fn poll_loop(state: std::sync::Arc<RouterState>) {
+    loop {
+        poll_once(&state).await;
+        tokio::time::sleep(state.poll_interval).await;
+    }
+}
+
+/// Poll every configured cortex once. Public for testing.
+pub async fn poll_once(state: &RouterState) {
+    for cortex in &state.cortexes {
+        poll_cortex(state, &cortex.name, &cortex.endpoint).await;
+    }
+}
+
+/// Poll one cortex: refresh its model map from `/v1/models`, then its node
+/// health from `/health`. A `/v1/models` failure debounces toward
+/// unreachable; the `/health` poll is best-effort and never flips
+/// reachability on its own (a cortex serving `/v1/models` is routable even
+/// if `/health` momentarily isn't).
+async fn poll_cortex(state: &RouterState, name: &str, endpoint: &str) {
+    // A cortex whose pinned TLS client failed to build (#74) is disabled:
+    // there is no client to poll with, so it stays unreachable.
+    let Some(client) = state.client_for(name) else {
+        let mut topo = state.topology.write().await;
+        if let Some(entry) = topo.get_mut(name) {
+            entry.consecutive_failures = entry.consecutive_failures.saturating_add(1);
+            entry.reachable = false;
+        }
+        tracing::warn!(cortex = name, "no TLS client (disabled); skipping poll");
+        return;
+    };
+
+    let models = fetch_models(client, endpoint).await;
+
+    let mut topo = state.topology.write().await;
+    let Some(entry) = topo.get_mut(name) else {
+        return; // not a configured cortex (shouldn't happen)
+    };
+
+    match models {
+        Ok(models) => {
+            entry.models = models.into_iter().map(|m| (m.id.clone(), m)).collect();
+            entry.reachable = true;
+            entry.consecutive_failures = 0;
+            entry.last_poll = Some(Utc::now());
+            tracing::debug!(cortex = name, models = entry.models.len(), "poll ok");
+        }
+        Err(reason) => {
+            entry.consecutive_failures = entry.consecutive_failures.saturating_add(1);
+            if entry.consecutive_failures >= POLL_FAILURE_THRESHOLD {
+                entry.reachable = false;
+            }
+            tracing::warn!(
+                cortex = name,
+                failures = entry.consecutive_failures,
+                reachable = entry.reachable,
+                reason,
+                "cortex poll failed"
+            );
+        }
+    }
+    drop(topo);
+
+    // Best-effort health (node counts). Never flips reachability.
+    if let Some((healthy, total)) = fetch_health(client, endpoint).await {
+        let mut topo = state.topology.write().await;
+        if let Some(entry) = topo.get_mut(name) {
+            entry.healthy_nodes = healthy;
+            entry.total_nodes = total;
+        }
+    }
+}
+
+/// GET `/v1/models`, returning the parsed entries or a short failure reason.
+async fn fetch_models(
+    client: &reqwest::Client,
+    endpoint: &str,
+) -> Result<Vec<CortexModelEntry>, &'static str> {
+    let url = format!("{endpoint}/v1/models");
+    let resp = client
+        .get(&url)
+        .timeout(POLL_TIMEOUT)
+        .send()
+        .await
+        .map_err(|_| "unreachable")?;
+    if !resp.status().is_success() {
+        return Err("non-success status");
+    }
+    let envelope = resp
+        .json::<ModelsEnvelope>()
+        .await
+        .map_err(|_| "bad json")?;
+    Ok(envelope.data)
+}
+
+/// GET `/health`, returning `(healthy, total)` node counts. `None` on any
+/// failure — the caller leaves the previous counts in place.
+async fn fetch_health(client: &reqwest::Client, endpoint: &str) -> Option<(u32, u32)> {
+    let url = format!("{endpoint}/health");
+    let resp = client.get(&url).timeout(POLL_TIMEOUT).send().await.ok()?;
+    if !resp.status().is_success() {
+        return None;
+    }
+    let health = resp.json::<CortexHealth>().await.ok()?;
+    Some((health.nodes.healthy, health.nodes.total))
+}

crates/helexa-router/src/state.rs

@@ -0,0 +1,144 @@
+use crate::config::{CortexEndpoint, RouterConfig};
+use chrono::{DateTime, Utc};
+use cortex_core::node::CortexModelEntry;
+use std::collections::HashMap;
+use std::time::Duration;
+use tokio::sync::RwLock;
+
+/// Shared router state: the configured cortex list plus the live topology
+/// map the poller (#72) maintains and the dispatcher (#73) will route on.
+///
+/// This is the router tier of the fractal neuron ← cortex ← router design:
+/// just as cortex polls each neuron for capacity/catalogue, the router
+/// polls each cortex's `/health` + `/v1/models`.
+#[derive(Debug)]
+pub struct RouterState {
+    /// Downstream cortex endpoints, as configured.
+    pub cortexes: Vec<CortexEndpoint>,
+    /// Per-cortex HTTP client, keyed by cortex name (#74). A cortex enrolled
+    /// with a `tls_ca` gets a client that trusts only that anchor; others
+    /// get a default client. A cortex whose `tls_ca` failed to load is
+    /// **absent** here — `client_for` returns `None` and it is never
+    /// polled or routed to (fail closed: a misconfigured pin must not
+    /// silently fall back to unpinned TLS).
+    clients: HashMap<String, reqwest::Client>,
+    /// This router instance's region, for dispatch geo affinity (#73).
+    pub region: Option<String>,
+    /// How often the poller refreshes the topology.
+    pub poll_interval: Duration,
+    /// Live per-cortex topology, keyed by cortex name. Pre-populated from
+    /// config (every configured cortex present, `reachable = false`) so the
+    /// poller and handlers always find an entry; the poller flips
+    /// reachability and fills the model map.
+    pub topology: RwLock<HashMap<String, CortexTopology>>,
+}
+
+/// Live view of one downstream cortex, refreshed each poll.
+#[derive(Debug, Clone, Default)]
+pub struct CortexTopology {
+    /// Whether the cortex is currently routable. Flipped `false` only after
+    /// [`crate::poller::POLL_FAILURE_THRESHOLD`] consecutive failed polls
+    /// (debounces transient blips); restored on the next successful poll.
+    pub reachable: bool,
+    /// Consecutive failed polls; reset to 0 on success.
+    pub consecutive_failures: u32,
+    /// Timestamp of the last successful poll.
+    pub last_poll: Option<DateTime<Utc>>,
+    /// Healthy / total neuron counts from the cortex's `/health` (coarse
+    /// load signal; #73 refines headroom). 0/0 until first health poll.
+    pub healthy_nodes: u32,
+    pub total_nodes: u32,
+    /// The cortex's full `/v1/models` entries, keyed by model id. Stored
+    /// whole (not distilled to a loaded/feasible bool) so the federation
+    /// catalogue (#75) can preserve per-model `limit`/`cost`/capabilities.
+    pub models: HashMap<String, CortexModelEntry>,
+}
+
+/// Whether a cortex can serve this model — loaded now, or feasible to
+/// cold-load (its catalogue × topology says some neuron can host it).
+pub fn entry_feasible(entry: &CortexModelEntry) -> bool {
+    entry.loaded || !entry.feasible_on.is_empty()
+}
+
+impl RouterState {
+    pub fn from_config(config: &RouterConfig) -> Self {
+        let topology = config
+            .cortexes
+            .iter()
+            .map(|c| (c.name.clone(), CortexTopology::default()))
+            .collect();
+
+        // One client per cortex. A `tls_ca` that fails to load omits the
+        // cortex from the map (fail closed) rather than degrading to an
+        // unpinned client.
+        let mut clients = HashMap::new();
+        for c in &config.cortexes {
+            match build_client(c.tls_ca.as_deref()) {
+                Ok(client) => {
+                    clients.insert(c.name.clone(), client);
+                }
+                Err(e) => {
+                    tracing::error!(
+                        cortex = %c.name,
+                        tls_ca = c.tls_ca.as_deref().unwrap_or(""),
+                        error = %e,
+                        "failed to build pinned TLS client; cortex disabled (fail closed)"
+                    );
+                }
+            }
+        }
+
+        Self {
+            cortexes: config.cortexes.clone(),
+            clients,
+            region: config.router.region.clone(),
+            poll_interval: Duration::from_secs(config.router.poll_interval_secs),
+            topology: RwLock::new(topology),
+        }
+    }
+
+    /// The HTTP client to use for `name`, or `None` if the cortex is
+    /// disabled (its `tls_ca` failed to load). Callers must treat `None` as
+    /// "not routable / not pollable".
+    pub fn client_for(&self, name: &str) -> Option<&reqwest::Client> {
+        self.clients.get(name)
+    }
+
+    /// Names of reachable cortexes that can serve `model_id` (loaded or
+    /// feasible to cold-load). Groundwork for capacity-aware dispatch (#73);
+    /// unreachable cortexes are excluded by construction.
+    pub async fn cortexes_serving(&self, model_id: &str) -> Vec<String> {
+        let topo = self.topology.read().await;
+        topo.iter()
+            .filter(|(_, t)| t.reachable)
+            .filter(|(_, t)| t.models.get(model_id).is_some_and(entry_feasible))
+            .map(|(name, _)| name.clone())
+            .collect()
+    }
+}
+
+/// Build a cortex HTTP client. With `tls_ca` set, the client trusts **only**
+/// that PEM anchor (platform roots disabled) — pinning the router→cortex hop
+/// to an enrolled cert (#74). Without it, standard platform-root validation.
+pub fn build_client(tls_ca: Option<&str>) -> Result<reqwest::Client, BuildClientError> {
+    let mut builder = reqwest::Client::builder();
+    if let Some(path) = tls_ca {
+        let pem = std::fs::read(path).map_err(|e| BuildClientError::Read(path.to_string(), e))?;
+        let cert = reqwest::Certificate::from_pem(&pem).map_err(BuildClientError::Parse)?;
+        builder = builder
+            .tls_built_in_root_certs(false)
+            .add_root_certificate(cert);
+    }
+    builder.build().map_err(BuildClientError::Build)
+}
+
+/// Why a cortex's pinned client could not be built (→ cortex disabled).
+#[derive(Debug, thiserror::Error)]
+pub enum BuildClientError {
+    #[error("reading TLS anchor '{0}'")]
+    Read(String, #[source] std::io::Error),
+    #[error("parsing TLS anchor PEM")]
+    Parse(#[source] reqwest::Error),
+    #[error("building HTTP client")]
+    Build(#[source] reqwest::Error),
+}