xerj-org/xerj is a local-first search engine for agents and an MCP server. MCP is the standard interface agents use when calling external tools. The latest release tag is v1.0.0-rc.97, where rc stands for release candidate. No changes are listed for this version.
A single Rust binary indexes code, documents, logs, JSONL, CSV, and PDF locally, providing BM25 full-text search, vector search, hybrid search, code search that works without grep, plus agent memory and RAG all in one. BM25 is a full-text search method that ranks documents by term frequency and document length, while vector search finds documents that are semantically close. RAG is the approach of answering questions based on the documents gathered through retrieval.
The search API is compatible with Elasticsearch, and Corpus Hub provides hundreds of reference memory corpora. The repository is written in Rust, is licensed under Apache-2.0, and was created on 2026-06-30. It has 10,975 GitHub stars.
However, the README states that the insecure execution option grants index read and delete permissions to any caller that can reach it, and warns that it should be used only in local development environments. The provided information contains no figures comparing search quality or performance, and in the 10-point scale rating only the momentum item remains, at 8.0. So it is close to a viable option for managing agent memory and search together locally, but it is hard to judge from this material alone what level of accuracy and performance it reaches.