Fused retrieval stack
Retrieval is fused and reranked — not a single brittle vector store.
The unified pipeline handles incremental updates and predictive pre-indexing so context is fresh without full re-scans.
Instant grep paths give a sub-second interactive feel on large repos.
Optional content proofs filter shared indexes so teammate reuse does not leak unproven paths.
@-mention resolvers and codebase search share the same fused retrieval substrate as agent tools.