xerj

Why we included this project

Teams building AI agents against their own repositories often find the real bottleneck is turning scattered files into something an LLM can query without exhausting the context window. XERJ handles that step directly: run its autoindexer against a folder and it identifies each file type, runs source through tree-sitter so symbols and line numbers are preserved, and creates a search index per dataset it finds. Because the server speaks the Elasticsearch API, existing dashboards and client libraries work without modification, and the same index serves keyword, vector, and hybrid retrieval for RAG pipelines or long-lived agent memory. The whole thing is one static binary with no JVM or runtime dependencies, which keeps a self-hosted deployment simple. For teams that would rather have an agent ask questions about their data than feed it entire files, this is a practical fit.