XERJ: Let your coding assistant search your files instead of reading them all

What it does

XERJ is a search engine that runs on your own machine as one Rust program. It reads (indexes) folders of code, docs, logs, JSONL or CSV files and PDFs, and supports keyword search (BM25), meaning-based vector search, and a mix of both. It speaks the same API as Elasticsearch, a widely used search server. Through MCP (a way to plug tools into an AI assistant), an assistant asks XERJ for the right passage instead of opening whole files one by one. The creators report a controlled study where this cut a coding agent's output tokens by 2.7x at the same solve rate.

How to use it

Input → action → output: start the XERJ server, connect it to Claude Code or Cursor, and point it at a project folder. When you or your assistant ask a question like where a function is defined, it returns ranked passages with file and line locations.

Common use cases

• Index your project folder and ask your coding assistant how a feature works; it pulls the matching code passages with file and line numbers.
• Have your assistant download and index open source repos similar to your task, then search how they solved the same problem before writing code.
• Store notes in an agent memory index so your assistant can search earlier decisions in later sessions.