grenade

grenade pushed to main at helexa/helexa

2026-06-18 09:46:31 +00:00

3b9a6e37f6 Merge fix/cortex-poll-debounce-retryable: poll debounce + retryable 503 for feasible-but-unhealthy node

526b662c5e fix(cortex): poll-failure debounce + retryable 503 for feasible-but-unhealthy node

Compare 2 commits »

grenade pushed to main at helexa/helexa

2026-06-18 09:43:04 +00:00

db7e373b90 fix(neuron): decouple GET /models from the inference worker (control-plane starvation)

grenade created branch fix/cortex-poll-debounce-retryable in helexa/helexa

2026-06-18 09:39:23 +00:00

grenade pushed to fix/cortex-poll-debounce-retryable at helexa/helexa

2026-06-18 09:39:23 +00:00

526b662c5e fix(cortex): poll-failure debounce + retryable 503 for feasible-but-unhealthy node

grenade pushed to fix/neuron-models-control-plane-starvation at helexa/helexa

2026-06-18 09:32:20 +00:00

db7e373b90 fix(neuron): decouple GET /models from the inference worker (control-plane starvation)

grenade created branch fix/neuron-models-control-plane-starvation in helexa/helexa

2026-06-18 09:32:19 +00:00

grenade pushed to main at helexa/helexa

2026-06-17 18:47:08 +00:00

5c1623a817 fix(#49): allow-anonymous mode must ignore unrecognized keys, not 401

grenade created branch fix/47-anonymous-tolerates-bad-key in helexa/helexa

2026-06-17 18:40:39 +00:00

grenade pushed to fix/47-anonymous-tolerates-bad-key at helexa/helexa

2026-06-17 18:40:39 +00:00

5c1623a817 fix(#49): allow-anonymous mode must ignore unrecognized keys, not 401

grenade closed issue helexa/helexa#54

2026-06-17 18:27:50 +00:00

neuron: per-principal in-flight cap (fair-share)

grenade closed issue helexa/helexa#49

2026-06-17 18:27:42 +00:00

cortex: API-key authentication + principal resolution

grenade closed issue helexa/helexa#55

2026-06-17 18:27:32 +00:00

cortex: load-aware routing across replicas + backpressure propagation

grenade closed issue helexa/helexa#56

2026-06-17 18:27:32 +00:00

cortex: fail-fast prompt pre-validation + advisory client hints

grenade closed issue helexa/helexa#53

2026-06-17 18:27:29 +00:00

neuron: admission control — bounded queue + backpressure, expose load in /health

grenade closed issue helexa/helexa#52

2026-06-17 18:27:28 +00:00

cortex: budget enforcement — hard per-key caps, reserve→settle, 429 + Retry-After

grenade closed issue helexa/helexa#51

2026-06-17 18:27:27 +00:00

cortex: per-request token metering + per-key/account spend ledger

grenade closed issue helexa/helexa#50

2026-06-17 18:27:26 +00:00

cortex: EntitlementProvider trait + local/static provider

grenade commented on issue helexa/helexa#56

2026-06-17 18:27:20 +00:00

cortex: fail-fast prompt pre-validation + advisory client hints

Done — merged to main (057bc71, merge 3b60dd7).

cortex pre-validates prompt length against the model's advertised limit.context (#62/#67) before dispatch (in proxy_with_metrics,…

grenade commented on issue helexa/helexa#55

2026-06-17 18:27:15 +00:00

cortex: load-aware routing across replicas + backpressure propagation

Done — merged to main (dd31c3c, merge 4feaaf1).

NodeState.model_load holds each neuron's per-model admission load, stashed by the poller from GET /health (#53). router::resolve now…

grenade commented on issue helexa/helexa#54

2026-06-17 18:27:10 +00:00

neuron: per-principal in-flight cap (fair-share)

Done — merged to main (c83f1eb).

AdmissionController now keeps per-principal accounting (Mutex<AdmissionState> holding the overall pending count + a per-principal map). enter(principal)…