Repository navigation

proxy: directory capacity, cross-engine JIT rule, /v1/models context_length (C) - #6

Closed

ehrvs wants to merge 10 commits into

feat/slot-aware-scoringfrom

feat/slot-capacity-proxy

Closed

proxy: directory capacity, cross-engine JIT rule, /v1/models context_length (C)#6
ehrvs wants to merge 10 commits into
feat/slot-aware-scoringfrom
feat/slot-capacity-proxy

Commits on Oct 16, 2026
- engine-manager: typed per-model slot capacity extraction
  
  ehrvs
  and
  claude
  committed
  3802cb5
- Stop llama.cpp slots poll from auto-loading models; test slots via bundled manifests
  
  ehrvs
  and
  claude
  committed
  1a0b7b5
- Carry slot capacity and jit_load through the node directory
  
  ehrvs
  and
  claude
  committed
  197491b
- Address review: observed-at heartbeat, settings read failures, key validation
  
  ehrvs
  and
  claude
  committed
  305a73a
- Merge remote-tracking branch 'pms/feat/slot-capacity-directory' into feat/slot-capacity-proxy
  ehrvs
  committed
  ab87ed4
- Wire directory capacity into the proxy; add /v1/models context_length
  
  ehrvs
  and
  claude
  committed
  461034b
- Update observed-at comment for A2 heartbeat semantics
  
  ehrvs
  and
  claude
  committed
  12a7b20
- Fix Codex findings: JIT claim, atomic failover eligibility, scanner observation time, Ollama key normalisation, fleet context_length
  
  ehrvs
  and
  claude
  committed
  9ea5408
- Hold JIT claim past request end until a newer sweep; apply eligibility on selected-node path
  
  ehrvs
  and
  claude
  committed
  77707f6
- Prefer the local node on routing ties before scheduler order
  
  ehrvs
  and
  claude
  committed
  65e5aa4