Lumen
A fully local semantic code search engine for AI coding agents that indexes a codebase with open-source embedding models and lets tools like Claude Code find relevant code without reading whole files, measurably cutting token cost and time.
Lumen builds a semantic index of a codebase using local embedding models served by Ollama or LM Studio, stored in SQLite with sqlite-vec, so an AI coding agent can call a semantic_search tool to find relevant code instead of reading entire files to locate it. It runs as a single static binary with no API keys, cloud service or external database.
Across nine benchmark runs spanning nine languages on real GitHub bug-fix tasks, the project reports that using Lumen cut cost in every language tested, by up to 39 percent, reduced output tokens by up to 66 percent, and completed sessions up to 53 percent faster, while maintaining patch quality, using an open-source benchmark framework that can be rerun independently. It installs as a plugin for Claude Code and Codex from a plugin marketplace, from npm for OpenCode, or as packaged support for Cursor.