Hey Product Hunt! I'm Aeneas, and I built Lumen to solve a frustration I kept running into with AI coding agents. Tools like Claude Code and Codex are powerful, but they often brute force code discovery with grep and find, because they don't have a reliable way to search your codebase semantically. They either re-read files on every query or miss relevant context entirely. The result is slower, less accurate suggestions and a lot of wasted tokens. Lumen fixes this by running a local semantic index on your repo. It parses your code into meaningful chunks (functions, types, methods) using language-aware ASTs, embeds them with a local model via Ollama or LM Studio, and stores everything in a SQLite database with vector search. A Merkle tree tracks changes, so only modified files get re-indexed keeping it fast even on large repos. It plugs into Claude Code as a first-class plugin and exposes search via MCP, so agents can call it like any other tool. Everything stays on your machine — no data leaves, no cloud dependency. Would love to hear what languages or editor integrations you'd want to see next. Happy to answer anything!

Lumen (by Ory)

Give your AI agent eyes on your codebase and save tokens

Give your AI agent eyes on your codebase and save tokens