AutoRAG 2.0: A self-evolving librarian agent that reads your files instead of listing them

What AutoRAG 2.0 actually is

The repository marker-inc-korea/AutoRAG has changed identity. The original Python AutoRAG, a RAG AutoML tool for finding an optimal pipeline, now lives in the legacy/ directory and is maintained in maintenance mode. The new AutoRAG 2.0 is a TypeScript agent built on the Pi agent loop, configured as a librarian. It is not a coordinator for multiple model roles. You configure one model, and that model handles retrieval, reading source files, judging evidence, and curating a structured answer. This is a meaningful shift from the original project's focus on pipeline optimization. The README describes it as searching your files, mail, chats, and notes in place.

The librarian loop: retrieve, read, judge, curate

The agent's workflow is visible in the README's example: a user asks what changes were made to the cloud infrastructure contract in Q3, and AutoRAG returns numbered knowledge units, each citing its source, such as contracts/cloud-2026.pdf:p.4. The mechanism is a loop where retrieval tools supply candidate paths, then the agent opens the source material directly through a built-in bash tool. It reads the underlying documents before curating. This design means the agent is not a search tool that points you to files; it does the reading and synthesizing. The README contrasts this with search tools that dump file paths and matching lines. This is a point of view that trades transparency for convenience, and it only works if the agent's judgment is trustworthy.

Retrieval methods and MinSync's shared chunk lifecycle

AutoRAG combines keyword and meaning search: the README lists hybrid keyword + meaning search with a reranker, plus scheduled re-indexing that runs even while the app is closed. Semantic indexing runs on MinSync's local Qwen3-Embedding-0.6B engine (1024 dimensions), which MinSync provisions and runs; the README states that AutoRAG owns no embedding model. File contents are never sent to a remote embedder unless you explicitly configure a remote minSync.embedder.

Getting it running: configuration and commands

The README's surface table lists the ways in: the desktop app (Dazzi, whose source lives in NomaDamas/AutoRAG-app), a terminal UI started with autorag tui, messenger connections for Telegram and Discord, a CLI for scripts, indexing, and health checks, an MCP server for feeding retrieval to Claude Code, Codex, and other agents, a TypeScript API for embedding the librarian in your own app, and agent skills that let a coding agent install and drive AutoRAG. The npm package is @autorag/librarian; its package.json requires Node 24 or newer and exposes two binaries, autorag and autorag-mcp. From a checkout, the Makefile provides make install (dependencies from bun.lock), make build (library, CLI, and declarations), and make test. Sign-in tokens and API keys live in the system's encrypted store rather than in a config file, and you can use the hosted Dazzi Models plan or connect your own provider key.

The thin PDF extraction retry mechanism

The desktop app previews Markdown, code, PDF, video, audio, Office, and HWP/HWPX files instantly in Quick Look. HWP pages are drawn locally by rhwp, so Hancom Office is not required. The app also stacks near-duplicate files such as proposal-final, proposal-v3 and proposal (1) into one row tagged Latest, Similar, or Identical, using dupey, the same scan as autorag duplicates.

Limitations and wrong-tool cases

AutoRAG 2.0 is not a general-purpose search engine. It is a single-agent librarian that reads configured source directories directly through bash. This design has a clear failure mode: it requires the agent to have execution access to those directories, which is a security consideration for sensitive or multi-tenant environments. The project is also a complete reimagining of the original AutoRAG, so users expecting the Python pipeline optimization tool will find a different product. The legacy tool is maintained in maintenance mode, which means no new features will come to it. If your need is to optimize a RAG pipeline through AutoML, AutoRAG 2.0 is the wrong tool; you should use the legacy version.

Alternatives and the competitive landscape

A real alternative is the original AutoRAG (legacy), which takes a fundamentally different approach: the README describes it as the RAG AutoML tool for automatically finding an optimal RAG pipeline for your data. AutoRAG 2.0 instead is a single-model librarian agent that searches your files, mail, chats, and notes in place and reads source documents directly via bash. It also reaches beyond local files: the README lists Slack, KakaoTalk, Discord, Telegram, WhatsApp (read-only), Mail, GitHub and Obsidian as datasources searched alongside your files. Another alternative would be a traditional search tool, which the README says dumps file paths and matching lines and leaves you to open files, read context, and synthesize answers yourself.

Maintenance, licensing, and upgrade cost

The repository shows recent activity: the latest releases are v2.13.4 and v2.13.3 (both 2026-10-31) and v2.13.2 (2026-10-29), and the last push was on 2026-11-02. The licence is Apache-2.0, declared in the repository metadata and in the package.json of @autorag/librarian. The desktop app is licensed differently: its source lives in NomaDamas/AutoRAG-app under the Fair Source FSL-1.1-MIT licence, and it depends on the Apache-2.0 engine from this repository. The legacy AutoRAG is maintained in maintenance mode with bug fixes and dependency updates, and it continues to be released on PyPI via pip install AutoRAG. The rapid release cadence suggests that users need to track changes between versions.

Editorial conclusion

Adopt AutoRAG 2.0 if you want one agent that searches your files, mail, chats, and notes in place and answers with every claim linked to the passage behind it. Do not adopt it if you need the original RAG pipeline optimizer, which now lives in legacy/ and is installed with pip install AutoRAG, or if you cannot allow the agent to open source documents via bash. Before using it, check that your environment meets the Node 24 requirement in package.json and that running MinSync's local Qwen3-Embedding-0.6B engine is acceptable on your machine. The project is under active development with frequent releases (v2.13.4 on 2026-10-31), so check release notes between versions.

Frequently asked questions

Official sources

1. Issues
2. Marker-Inc-Korea/AutoRAG on GitHub
3. Project website
4. README
5. Releases