XERJ
Code search only helps when the model has not already seen the code. XERJ is a local Rust search engine for coding agents: one binary indexes code, docs, logs and PDFs, then serves answers with file and line references over an Elasticsearch compatible API and MCP. The agent stops reading whole files and pulls only the passage it needs. The project reports 2.7 times fewer output tokens than grep, and 26 times fewer than a memory based approach, on its own benchmark. It has grown fast since the first write up, from 4.2k stars to about 19.1k, and is now at release candidate v1.0.0-rc.102, released 6 November 2026. Worth trying on a private codebase, where the gain is largest.
What you get
- One local Rust binary that indexes code, docs, logs and PDFs, with no cloud service to run
- Answers with file and line references, over an Elasticsearch compatible API and MCP
- The agent pulls only the passage it needs instead of reading whole files
- Reference coding: index the repos closest to your task and reuse how they solved it before writing your own code
The win is the token bill, not the index: the agent reads the passage, not the file. Try it on a private codebase first, where the gain is largest, and check the file and line references before you trust the answers.
Repo: https://github.com/xerj-org/xerj
Quick poll
What would you use XERJ for first?
- Searching a private codebase without reading whole files
- Indexing docs, logs and PDFs alongside code
- Wiring it into my agents over MCP
- Watching it for now
Vote in the comments.
If you found XERJ interesting, I write about tools like this every week in The AI-Native Engineering Leader, what actually works for engineering leaders building with AI agents, and what to skip.
Top comments (0)