chunkhound

ChunkHound describes itself as local-first codebase intelligence: your assistant can search code but does not understand it, so ChunkHound researches the codebase and extracts architecture, patterns and institutional knowledge. It plugs into AI clients through MCP and the README lists Claude, VS Code, Cursor, Windsurf and Zed. Features include the cAST algorithm for semantic code chunking (linked to a research paper), multi-hop semantic search that follows related code beyond direct matches, natural-language semantic search, regex search that needs no API key, and real-time indexing with file watching and branch switching.

It parses 30 languages: programming languages through Tree-sitter (Python, JavaScript, TypeScript, Java, Go, Rust, C/C++ and more), configuration formats such as JSON, YAML, TOML and HCL, and text and PDF through custom parsers. Requirements are Python 3.10+ and uv. Embeddings can come from VoyageAI (recommended), OpenAI or local Ollama, and the code-research LLM can be the Claude Code or Codex CLI (no API key) or an Anthropic or OpenAI key. Install with `uv tool install chunkhound`, put a `.chunkhound.json` in the project root and run `chunkhound index`.

The README positions it for large monorepos, security-sensitive codebases and air-gapped setups, since the code stays on your machine. Note that with a cloud embedding provider, chunks of code are sent to that provider. It is MIT licensed and was last pushed on 2026-10-12. This review is based on the README and repository data; we have not run it ourselves.