ory/lumen

Save 30% token costs when using Claude Code, Codex, OpenCode for free - with open source, local semantic search. Works for small and large codebases and monorepos! Enterprise-ready and fully compliant via Ollama and SQLite-vec.

What it solves

AI coding agents often struggle with large codebases because they attempt to read entire files, which is slow and expensive. Lumen provides a local, semantic search engine that acts as a map for these agents, allowing them to retrieve only the specific functions, types, or modules relevant to a task.

How it works

Lumen functions as an MCP (Model Context Protocol) server that sits between your codebase and your AI agent (such as Claude Code, Cursor, or Codex). It uses a Merkle tree to detect file changes and only re-indexes what has been modified. It parses code into semantic chunks using native ASTs or tree-sitter, converts these into vector embeddings using local models (via Ollama or LM Studio), and stores them in a local SQLite database with sqlite-vec. When an agent needs information, it queries this index to find relevant code snippets instead of reading entire files.

Who it’s for

Developers who use AI coding agents and want to reduce API costs, improve response times, and maintain high-quality code generation without sending their codebase to the cloud.

Highlights

  • Zero-Cloud Privacy: All indexing and embedding happen locally on your machine.
  • Significant Efficiency Gains: Benchmarks show up to 39% cost reduction and 53% faster session completion across various programming languages.
  • Incremental Indexing: Uses Merkle tree diffing to ensure only changed files are re-processed, making updates nearly instantaneous for large projects.
  • Broad Language Support: Supports 12 language families, including Go, Python, TypeScript, Rust, and Java.
  • Git Worktree Aware: Automatically shares index data across Git worktrees to avoid redundant processing.

Related

  • Project
    lee-to/aif-handoff

    AIF Handoff is an open‑source autonomous Kanban board that uses Claude, Codex, OpenRouter or OpenCode LLMs to automatically plan, implement, review and close software tasks. Built on the AI Factory framework, it offers a drag‑and‑drop UI, modular runtime adapters, auto‑review loops with convergence detection, and optional human‑in‑the‑loop modes. The repo includes both a Node‑only dev setup and a Docker compose stack, with detailed runtime configuration and fault‑tolerant agent orchestration.

  • Project
    azorng/goose.nvim

    A Neovim plugin that integrates the Goose AI agent, providing a chat interface with editor context awareness and agentic tool-use modes.

  • Project
    Brain0-ai/brain0

    brain0 is an offline-first provenance engine that passively links AI coding-agent transcripts to git history, building an explorable decision graph of your repo. It reveals which prompt wrote each change, what agents read, drift between declared and actual edits, and evidence-driven risk scores.

  • Project
    fennaraOfficial/fennara-godot-ai

    A tool that connects AI assistants to the Godot game engine via MCP, allowing agents to inspect scenes, read logs, and validate changes in real-time.