The living index of how your system actually works: conforming patterns, sealed boundaries, recorded decisions, and retrievable evidence — feeding every review, every spec, and every AI in the pipeline.
Code Guard comments, Spec Forge grounding, VS Code context, and Ask Foixar answers all draw from the same source of truth.
Semantic vectors plus keyword precision — the right chunk surfaces whether you ask like an architect or like a BA.
Merges update the index incrementally — grounding reflects the codebase as of the last PR, not last quarter.
“It's the difference between a linter and a colleague. The comment doesn't just say no — it shows you the fourteen places that do it right.”
For any convention, the index knows the conforming implementations — so guidance arrives as ‘14 examples in payments-api’, not abstract rules.
Sealed paths and slice maps live in the index — the same map that refuses out-of-bounds generation powers architecture questions.
ADRs attach to the code they govern. Retrieve a chunk and its ‘why’ comes with it — the wiring behind Decision Memory.
Spec Forge evidence scores are computed against this index — 619 chunks is not a metaphor, it is the retrieval set.
The VS Code extension injects index context into whatever AI you run — Copilot, Claude, Cursor — so suggestions conform the first time.
Indexing runs in your Azure tenant against your repos. Nothing leaves; nothing trains on your code.
Thirty minutes. Your stack. Bring one real feature request.