Guides
Field guides for grounding AI coding agents in real codebases — the concepts, the tradeoffs, and the commands. All built around one engine you can run locally.
Every guide is grounded in the same engine — hybrid, AST-aware, measured retrieval you can run on your own machine in three commands. Start wherever your question lives.
agentsWhat is an agentic workflow?
How LLM agents plan, act, and loop — why code retrieval is the bottleneck that decides whether they help or hallucinate, and how to ground any agent in your real codebase.
integrationThe MCP server for code search
Connect Claude Code, Cursor, or any MCP client to your codebase: hybrid search, symbol lookups, repo maps, and file tools — one command each, all local.
retrievalCode RAG: RAG for your codebase
What retrieval-augmented generation over source code is, why naive line-chunked vector search fails on real repos, and how to run measured code RAG locally.
retrievalSemantic code search
Find code by meaning, not keywords — where grep and pure vector search each fail, and how hybrid fusion covers the queries both miss.
decisionsRAG vs fine-tuning for code
Retrieve from your codebase or train on it? What each approach actually does, where each fails, and why code questions almost always want RAG.
agentsAgent memory for coding agents
What harnesses remember across sessions, how MCP memory layers work, and why memory should be local, measured, and visible.
setupGive Claude Code your whole codebase
Three commands to ground Claude Code in your real source: install, index, register the MCP server. Five minutes, fully local.
decisionsOCE vs Cursor, Copilot & grep
An honest comparison of codebase-context approaches — what each indexes, the tradeoffs, and when a dedicated engine earns its keep.
practiceContext engineering for coding agents
The context window is the agent's entire world. The four levers — retrieval, structure, budget, verification — and the evidence rail that makes them visible.
Task-based setup lives in the docs; retrieval quality is published on the benchmarks page; licensing on pricing.