Skip to content

v1.8.8 — local-model delegation

Latest

Choose a tag to compare

@KbWen KbWen released this 04 Jul 09:46
f53ffb9

Local models get an official on-ramp: the governed flow can now delegate scoped work to an installer's locally-hosted model (PRs #316#317).

  • /ask-local — local-model delegation entry (#316): new optional module driving any OpenAI-compatible local endpoint (Ollama / LM Studio / vLLM) as a delegated junior executor. Explicit opt-in only; silent fallback + zero cost when no endpoint exists; review/code modes with a classification cap. The patch contract keeps Write Isolation intact: the local model never writes files — the primary reviews, applies, and rejects the whole patch on any scope violation. §8.2 reused unchanged — zero new gates.
  • codex --oss local variant (#316): codex-cli §5a documents the native local-Ollama path, same governance + tightened caps.
  • Validator: HANDEDOFF→IMPLEMENTING reverse edge (#317): the documented ship-entry reversal is now representable in strict gate progression (sh+ps1, 3 regression tests incl. an M10 negative control). Found by dogfooding #316's own pre-merge quality loop.

Full details: CHANGELOG.md [1.8.8].