Language: 日本語English

lumen Plugin

Author
Ory Corp
Category
Development
Topics
Code Search & Code Intelligence · AI Agents & AI App Development
First seen
2026-07-09
Last confirmed
2026-07-13
Explanation last updated
2026-07-12

The explanation below is AI-generated. Please verify it against the sources.

Lumen is a locally-run semantic code search tool for AI coding agents such as Claude Code, Cursor, Codex, and OpenCode. According to the README, it indexes a codebase using Go's native AST parser (for Go) and tree-sitter grammars (for other languages), embeds the resulting chunks with a local embedding backend (Ollama or LM Studio), and stores vectors in SQLite with sqlite-vec for cosine-distance KNN search. It exposes a semantic_search MCP tool so agents can find relevant code by meaning instead of reading whole files. The README reports benchmark results across 9 languages showing cost, time, and output-token reductions with maintained patch quality when Lumen is used, based on the project's own bench-swe harness. No cloud services or API keys are required; all data and embeddings stay on the local machine.

Overview

Lumen is not a SaaS. Per the GitHub README, it is an open-source, locally-run tool distributed as a static binary and installed as a plugin (Claude Code plugin marketplace, Cursor plugin bundle, Codex MCP/skill setup, or an OpenCode npm plugin). It runs as a local MCP server and does not send code or embeddings to any external cloud. The homepage listed for this item (ory.sh) describes Ory's separate identity-and-access-management (IAM) product line and does not contain information specific to Lumen.

What you can do with lumen

  • Run semantic_search via MCP so agents like Claude find relevant functions/types by meaning rather than reading entire files (README)
  • Auto-index a project on session start, with Merkle-tree diffing so only changed files are re-processed (README)
  • Use incremental updates so large codebases re-index in seconds after the first run (README)
  • Get semantic chunking for 12 language families, including Go, Python, TypeScript, JavaScript, Rust, Ruby, Java, PHP, C/C++, C#, Dart, and Svelte (README)
  • Share index data automatically across Git worktrees, seeding a new worktree's index from a sibling's (README)
  • Choose a local embedding backend, either Ollama or LM Studio, with no cloud dependency (README)
  • Use shared doctor (health check) and reindex (forced re-index) skills across Claude Code, Cursor, Codex, and OpenCode (README)
  • Control what gets indexed with .lumenignore, .gitignore, .gitattributes, and built-in directory/lock-file exclusions (README)

Sources

Original description (English)

Precise local semantic code search via MCP. Indexes your codebase with Go AST parsing, embeds with Ollama or LM Studio, and exposes vector search to Claude through an MCP server — no cloud, no npm.

History of lumen

Back to list