Knowledge Systems

Vectorize

Self-hosted retrieval layer for AI coding agents. Indexes a repository on symbol boundaries, intercepts the agent’s own search calls, and returns the few chunks that matter - cutting context waste and keeping every line of code inside your network.

Role in stack
Agent infrastructure -> grounded context
Current state
Working agent support
Best demo use
The strongest built-and-running proof-point that RC builds for agents, not just about them
Honest boundary
Token savings are real but modest, 29% per retrieval; do not lead with the token-cost figure

Seven stages: repository scan with change detection, structure-aware chunking on symbol boundaries, local embedding into a self-hosted vector store plus a keyword index, interception of the agent's file-search call, hybrid retrieval with a reranking pass, context injection with token accounting, and an indexing-scope decision gate.

Completing the run opens the retrieval inspector: 31 scored candidates showing vector score, keyword score, and rerank score side by side, filterable by whether they were injected or dropped, with the reasoning for each.

The demo is a static mock. It runs entirely in your browser on synthetic data — nothing is gated, nothing is collected.