Repolode
Submit
Marker-Inc-Korea

Marker-Inc-Korea/AutoRAG

A personal librarian agent that searches your files, mail, chats and notes in place and returns cited, numbered answers.

View on GitHub
Stars
5,399
Forks
486
Open issues
81
Last commit
yesterday

What is it?

AutoRAG Agent is a personal librarian agent for office work. A single configured model searches local documents, mail, chat archives and notes where they already live, instead of copying them into a central vector database. It opens source files directly to verify facts, then returns numbered answers that point back to the passages they came from.

Indexing is handled automatically. Documents are parsed locally (PDF, HWP/HWPX, DOCX, XLSX, Markdown and more), then searched with BM25 keyword ranking, semantic vector search, or a hybrid of both. Connectors cover Slack, Discord, Telegram, WhatsApp, KakaoTalk, Notion, Gmail/IMAP mail, Obsidian and GitHub. The agent can also organize, rename, move or edit files on request, with changes recorded and deletions sent to the OS trash.

The project ships as the @autorag/librarian package, with a CLI, a terminal UI, an MCP server for coding agents, and a TypeScript API. A desktop app built on the same engine is maintained in a separate repository. The original Python AutoRAG pipeline optimizer now lives in a legacy directory and is in maintenance mode.

Key features

  • 01Searches local files, mail, chat archives and notes in place without copying them into a central database.
  • 02Returns numbered answers with citations pointing to source files and passages.
  • 03Combines BM25 keyword search, semantic vector search and hybrid ranking.
  • 04Parses PDF, HWP/HWPX, DOCX, XLSX and Markdown locally, with optional OCR for images.
  • 05Offers a CLI, a terminal UI, an MCP server and a TypeScript API.

Use cases

  • Finding clauses or figures across a folder of contracts, reports and spreadsheets.
  • Asking questions over months of Slack, Discord or Telegram conversations.
  • Searching mail archives and Obsidian vaults from one interface.
  • Giving a coding agent such as Claude Code access to local document retrieval through MCP.

Things to know

  • Requires Node 24 or newer and Bun 1.2 or newer.
  • The desktop app's source is under Fair Source (FSL-1.1-MIT), not Apache-2.0 like the engine.
  • Semantic search relies on the local MinSync embedder, and file-name search differs by platform (Everything on Windows, fsearch-cli on macOS/Linux).

In collections

MCP serversOpen-source personal AI agents

Topics

#rag#personal-agent#retrieval-augmented-generation#llm#document-parser#embeddings#bm25#mcp#agent#librarian

FAQ

What is AutoRAG?
A personal librarian agent that searches your files, mail, chats and notes in place and returns cited, numbered answers. AutoRAG Agent is a personal librarian agent for office work. A single configured model searches local documents, mail, chat archives and notes where they already live, instead of copying them into a central vector database. It opens source files directly to verify facts, then returns numbered answers that point back to the passages they came from.
Is AutoRAG free to use? What license does it have?
AutoRAG is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is AutoRAG still maintained?
Actively maintained. The last commit to AutoRAG was on Nov 6, 2026.
What can I use AutoRAG for?
Finding clauses or figures across a folder of contracts, reports and spreadsheets. Asking questions over months of Slack, Discord or Telegram conversations. Searching mail archives and Obsidian vaults from one interface. Giving a coding agent such as Claude Code access to local document retrieval through MCP.
Where can I get AutoRAG?
The source code and installation instructions are on GitHub: https://github.com/Marker-Inc-Korea/AutoRAG

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects