Expand description
Engine inference rankers (SCC-071): EmbeddingScorer fusing stored
entity embeddings via cosine similarity, EngineReranker calling a
separate /rerank model, the remote-model policy, and the rankers
constructor task-pack plumbing uses when inference.enabled is set.
Single implementation: transports resolve scorer + reranker through ONE engine constructor — no transport reimplements embedding auth policy or fallback semantics.
Structs§
- Embedding
Scorer - Fuses stored entity embeddings with the embedded goal. Vectors are preloaded once per pack generation.
- Engine
Reranker - Second-stage reranker calling the configured
/rerankmodel on the top candidates. Any failure is a no-op (graceful degradation).
Functions§
- rankers
- Build the scorer/reranker when inference is enabled; any provider failure degrades to (None, None) so the lexical ranker is always the fallback.
- remote_
inference_ allowed - Remote-model policy (P0, docs/SECURITY.md): repository-derived content
may leave the machine only when
inference.enabledANDsecurity.allow_remote_modelsare both true. Loopback providers need onlyinference.enabled. Fails closed.