grenade

grenade opened issue helexa/helexa#73

2026-06-17 15:16:45 +00:00

helexa-router: capacity-aware dispatch to a cortex with headroom (+ region affinity, failover)

grenade opened issue helexa/helexa#72

2026-06-17 15:16:38 +00:00

helexa-router: router↔cortex capacity & catalogue poller (multi-operator topology)

grenade opened issue helexa/helexa#71

2026-06-17 15:16:32 +00:00

extract cortex proxy.rs SSE passthrough into a shared streaming-proxy module

grenade opened issue helexa/helexa#70

2026-06-17 15:16:21 +00:00

helexa-router: binary skeleton — axum plaintext server reusing cortex-core

grenade commented on issue helexa/helexa#47

2026-06-17 14:08:57 +00:00

[epic] Multi-tenant governance: auth, federated token budgeting, concurrency control

Scope note: the traffic-ingress tier (router.helexa.ai → a cortex with capacity) is tracked separately in epic #69 (helexa-router), sibling to this one. This epic's Stage 4 (#57/#58/#5…

grenade commented on issue helexa/helexa#61

2026-06-17 14:08:48 +00:00

Epic: first-class opencode integration (dynamic helexa provider)

The two "(later)" router items here — Router/discovery contract and Router-default endpoint resolution — are now owned by the dedicated ingress epic #69 (helexa-router: public…

grenade opened issue helexa/helexa#69

2026-06-17 14:08:40 +00:00

[epic] helexa-router: public multi-operator ingress proxy (router.helexa.ai)

grenade commented on issue helexa/helexa#22

2026-06-17 13:34:12 +00:00

bench: reproducible benchmark harness + published numbers

Demoted p1-now → p3-later.

The unblocking core of this issue has shipped: the reproducible harness (script/bench.py) plus the continuous, version-aware helexa-bench (CLAUDE.md 2026-06-13…

grenade closed issue helexa/helexa#27

2026-06-17 13:19:25 +00:00

tracking: prioritised path to closing every open issue

grenade commented on issue helexa/helexa#27

2026-06-17 13:19:22 +00:00

tracking: prioritised path to closing every open issue

Closing as superseded.

This standalone ordering doc has served its purpose and is now stale: its P1 and P2 sections are entirely closed, and it predates the two epics and the issue range that now…

grenade closed issue helexa/helexa#31

2026-06-17 13:18:39 +00:00

feat(neuron): emit final usage frame on SSE streams (stream_options.include_usage)

grenade commented on issue helexa/helexa#31

2026-06-17 13:18:05 +00:00

feat(neuron): emit final usage frame on SSE streams (stream_options.include_usage)

Closing as superseded by #48 (closed 2026-06-16).

neuron now emits the OpenAI usage frame on streaming responses for both endpoints:

Chat completions: trailing usage_chunk (empty…

grenade pushed to main at gongfoo/gongfoo

2026-06-17 12:45:42 +00:00

0dc088a2f0 fix(runner): widen reaper limits for cuda-13.0 to fit full mistral.rs build

grenade closed issue helexa/helexa#67

2026-06-17 12:06:38 +00:00

neuron: self-derive and advertise optimal limit{context,input,output} from model config + live VRAM; catalogue limit becomes an override

grenade commented on issue helexa/helexa#67

2026-06-17 12:06:33 +00:00

neuron: self-derive and advertise optimal limit{context,input,output} from model config + live VRAM; catalogue limit becomes an override

Done — live-validated on beast

Landed on main in six phases (4f05a87 → 4b28a64). The neuron now self-derives limit{context,input,output} per loaded model from model architecture +…

grenade deleted branch feat/computed-context-limits from helexa/helexa

2026-06-17 11:34:39 +00:00

grenade pushed to main at helexa/helexa

2026-06-17 11:34:20 +00:00

4b28a64b34 feat(#67 phase 5b): enforce the derived input as the prompt cap

dd65eedb24 feat(#67 phase 5a): NEURON_MAX_PROMPT_TOKENS becomes a clamp-only backstop; docs

8b2e01a072 feat(#67 phase 4): advertise neuron-computed limit on /models; drop catalogue override

464b6b0db9 feat(neuron): self-measured prefill tok/s EMA on streaming paths (#67 phase 3)

f2e05d96ec feat(neuron): capture ContextProfile at load + per-rank VRAM fan-out (#67 phase 2)

Compare 6 commits »

grenade pushed to feat/computed-context-limits at helexa/helexa

2026-06-17 11:26:41 +00:00

4b28a64b34 feat(#67 phase 5b): enforce the derived input as the prompt cap

grenade pushed to feat/computed-context-limits at helexa/helexa

2026-06-17 11:14:37 +00:00

dd65eedb24 feat(#67 phase 5a): NEURON_MAX_PROMPT_TOKENS becomes a clamp-only backstop; docs

grenade pushed to feat/computed-context-limits at helexa/helexa

2026-06-17 11:10:24 +00:00

8b2e01a072 feat(#67 phase 4): advertise neuron-computed limit on /models; drop catalogue override