xerj-org/xerj

XERJ is a local AI search engine that indexes folders automatically to enable agents and developers to retrieve exact code passages without token waste from sequential file reading.

The problem XERJ solves is the inefficiency of agents reading files one by one through grep or similar tools, which consumes excessive tokens and slows down coding tasks. Its approach uses automatic indexing to create a searchable index of code, documentation, logs, and PDFs in a single command, then serves results through an Elasticsearch-compatible API. This allows agents to query for relevant passages directly rather than loading entire files into context. The tool is built in Rust and designed to run locally on a developer's machine.

Reference coding is the primary use case: pointing an agent at a coding task, having it download relevant open-source repositories, index them, and reuse existing solutions before writing new code. Developers should choose this tool if they work with coding agents and want to reduce token consumption and API costs while improving solution quality. The project is suitable for individual developers and smaller teams running agents locally. The Elasticsearch compatibility means existing search clients can work without modification.

The project maintains a community-driven development model centered on field reports: anyone running XERJ is asked to contribute one short report describing what they indexed, what they used it for, and their honest assessment. Field-report pull requests are exempt from contributor license agreements and merge without signatures. The tool includes a built-in feedback command that drafts reports with version and system information pre-filled. The security model is deliberately simple and documented: the tool binds to localhost by default and refuses cleartext binds on other interfaces unless explicitly enabled, treating the insecure flag as a trust boundary rather than a casual option.