Reference is a local semantic search tool designed specifically for your files and code, built to empower AI agents. Operating entirely on your machine with no cloud dependency, it ensures that your data never leaves your local environment. Instead of receiving generic advice, asking questions like 'how did I implement rate limiting here' returns your actual code, cited precisely down to the exact function.

Key engineering features include a live index that updates automatically as you save, code-aware chunking utilizing tree-sitter, and a built-in MCP server (featuring commands like /search, /explain, /find_similar, and /check_doc_drift). This allows tools like Claude Code to access precise, cited results directly, preventing them from wasting valuable tokens on repetitive grep loops.

Pros of Reference

Runs 100% locally with no cloud storage, keeping your code and files secure on your machine.
Provides exact code snippets and citations down to the specific function instead of generic advice.
Features a live index that automatically updates as you make changes and save files.
Built-in MCP server helps tools like Claude Code retrieve precise results and avoid token-heavy grep loops.

Cons of Reference

Platform availability is currently focused on Mac (based on available topics/metadata).
Requires integration with compatible AI coding setups/agents to maximize its MCP server capabilities.
Limited to local indexing environments.

Frequently Asked Questions

Does Reference send my code or files to the cloud?

No. Reference is entirely local, meaning nothing leaves your machine and no cloud processing is used.

How does Reference update its search index?

Reference features a live index that updates dynamically as you save your files.

What is the built-in MCP server used for?

The built-in MCP server provides commands like /search, /explain, /find_similar, and /check_doc_drift, allowing AI tools like Claude Code to get precise, cited results without burning tokens on grep loops.