v0.10.4

0.10.4 (2026-10-16)

Features

• agent: cap the bytes a single Model download may write (#2088) (3852411)
• api: endpoint API key and externalTrafficPolicy for InferenceService (#2043) (27a9214)
• controller: detect a stalled watch path with a lease canary, bounded reconciles and stall alerts (#2072) (d00e0cf)
• controller: detect keyed pool siblings and vLLM env-var keys in ModelRouter (#2068) (78889f0)
• controller: detect ModelRouter backends keyed through --api-key on the command line (#2063) (3674acc)
• controller: stage HF mirror repos with spec.files and expand globs from the repo listing (#2071) (9e5d7d3)

Bug Fixes

• agent: health restarts stop the unhealthy engine first, wait for repeated failures, and back off (#2090) (407e02e)
• agent: keep a Metal service running when its Model lookup fails transiently (#2089) (e22eeab)
• controller: merge DRA pod scheduling fields and reject multiNode with resourceClaims (#2062) (9319745)
• no device-plugin GPU limit when a Model uses DRA resourceClaims (#2060) (8c3e434)

Documentation

• APISIX in front of LLMKube (#2053) (16286c6)
• metal: fix install commands and update the quickstart for relay mode (#2087) (8bc7c6f)
• use an extension-less link in the GPU fault guide (#2056) (780e710)