AutoRAG Agent
A self-evolving 'librarian' agent that searches across multiple retrieval methods and opens source documents directly to answer questions grounded in a user's own files, superseding the original AutoRAG pipeline-optimization tool.
AutoRAG Agent is built on the Pi agent framework: a single configured model searches multiple retrieval methods, opens source documents directly through bash to verify ground truth, and curates answers into numbered knowledge units, instead of leaving a user to open files and synthesize answers from raw search hits.
It federates data in place by querying CLI-native stores such as lazykatok, discrawl, slacrawl, mailcrawl, rclone and qmd, instead of duplicating files into a centralized vector database, preserving source-native identities and local access control. It needs no pipeline tuning, vector-database maintenance or remote embedding API keys, managing incremental change-data-capture chunking and a local embedding gateway itself, and uses local BM25, vector and hybrid chunk modes for low-latency results.
The original Python-based AutoRAG, a RAG AutoML tool for finding an optimal retrieval pipeline, continues to be maintained in this repository's legacy/ directory and via pip install AutoRAG. This repository is written in TypeScript.