AutoRAG Agent
Local AI search agent for macOS, Windows and Linux that checks answers against source files. Its model-free Lite engine supports offline search.
Alternative to:
AutoRAG Agent searches documents and connected archives where they already live, then reads the relevant sources to produce answers with file, page or line references. It runs on macOS, Windows and Linux and suits people researching information spread across local files, messages and notes.
The full agent uses a configured model through the Pi agent framework to search, verify evidence and organize findings. It combines keyword, semantic and hybrid search, with filename discovery through Everything on Windows or optional FSearch on macOS and Linux. Feedback about useful results informs which search methods it prioritizes for similar questions.
AutoRAG Lite provides local retrieval without an LLM and can search offline. Its MCP server supplies context to an external AI agent. Embeddings run locally, and the tool reports no external telemetry. Remote reranking through OpenRouter is optional; that component uses a cloud service. Local indexes and parsed document mirrors stay in the workspace, while connected archives retain their native storage rather than moving into a central database.
Supported sources include Obsidian vaults, Slack and Discord archives, Notion, email, and cloud drives through rclone. It reads PDF, Word, Excel, PowerPoint and Korean HWP documents, with optional OCR for images and scanned PDFs. Access checks restrict connected-source searches to authorized data. A terminal interface supports ongoing research sessions, and a TypeScript API lets developers add the librarian to their own applications. Install the MIT-licensed @autorag/librarian package with Bun; the project documents Node 24 or later. The full agent can use a configured local model endpoint, including Ollama, or an external provider.
