RAG & Knowledge
MarkItDown
MarkItDown — Microsoft tool that converts PDFs, Office files, images and more to clean Markdown for LLMs, MIT
★ 193.5k starsMIT
View guide →LibreAI / RAG & Knowledge / QAnything
// rag · netease-youdao/QAnything
QAnything
QAnything is an open-source rag & knowledge project by netease-youdao (netease-youdao/QAnything on GitHub). It has 14,197 GitHub stars and 1,354 forks, is released under the AGPL-3.0 license, and was last updated on GitHub on 23 September 2026.
Local knowledge-base Q&A over PDFs, Office files, images and web pages - offline and CPU-friendly.
★ 14.2k stars⑂ 1.4k forksAGPL-3.0updated 2026-09
▦ visual — this project's README ships no embeddable screenshot,
so here's a placeholder. The guide below is complete regardless.
so here's a placeholder. The guide below is complete regardless.
// project facts
Project facts
- Repository
- netease-youdao/QAnything
- GitHub stars
- 14,197
- Forks
- 1,354
- License
- AGPL-3.0
- Last updated on GitHub
- 23 September 2026
- Category
- RAG & Knowledge
- Listing
- Full install guide
- Official site
- Official site
// how it works
How it works
- Prepare your machine
Install Docker and Docker Compose and make sure you have at least 20GB of RAM; no GPU is required for the default setup.
- Clone the repository
Pull the QAnything repo, which contains the per-platform Docker Compose files for one-click startup.
- Start the stack
Run the Docker Compose file for your operating system and wait for the 'qanything backend service is ready' log line (about 30 seconds).
- Configure a model and ask
Open the web UI at http://localhost:8777/qanything/, set your OpenAI-compatible API base and key (or a local Ollama model) in the front-end settings, upload documents and start asking.
// features
Why people love it
Two-stage retrieval
Embedding retrieval followed by reranking with BCEmbedding models keeps answer quality stable as data volume grows, instead of degrading like single-stage retrieval.
Many file formats
Supports PDF, Word, PPT, Excel, Markdown, email, TXT, images (with OCR), CSV and web links, with strong parsing of tables, columns and cross-page layouts.
Offline and CPU-friendly
Designed for data security: install and use it with the network unplugged, running by default in a pure CPU environment on Windows, Mac or Linux.
Cross-language Q&A
Ask in Chinese or English regardless of the language of the source files.
Flexible modes
Quick-start mode (upload and ask, or fileless chat), retrieval-only mode, FAQ matching, custom Bots and hybrid BM25-plus-embedding search.
Independent components
PDF parsing, OCR, embedding and rerank services are independent and replaceable, and can be called separately over HTTP.
Advertisement
// run it
Run it yourself
- Clone the repository
Pull the QAnything code with its Docker Compose files.
git clone https://github.com/netease-youdao/QAnything.git cd QAnything - Start with Docker Compose
Launch the stack with the compose file for your platform (Linux shown).
docker compose -f docker-compose-linux.yaml up - Open the web app
Once the log says the backend service is ready, open the front end in your browser.
# Open http://localhost:8777/qanything/ - see README for details
// use cases
What you can build
- Running a private document Q&A service over company manuals and contracts on an internal server.
- Answering questions over a large bilingual (Chinese-English) knowledge base without a GPU.
- Giving support teams a custom Bot grounded in product documentation and FAQs.
- Deploying document Q&A in an air-gapped environment with a local Ollama model.
Advertisement
// faq
Questions, answered
Who maintains QAnything?
QAnything is developed by NetEase Youdao's open-source team and community contributors.
Is it free to use?
The code is AGPL-3.0 licensed; note the README says commercial use should also follow the license of the underlying QwenLM models if you use them.
Does it need a paid API key?
Version 2.0 gets its LLM through an OpenAI-compatible interface: that can be a paid hosted API with your key, or a free local model served by Ollama.
How does QAnything compare with Dify or FastGPT?
Those are broader LLM app platforms; QAnything focuses specifically on local knowledge-base Q&A with strong document parsing and two-stage retrieval, deployable fully offline on CPU.
How do I get started?
Clone the repo, run the Docker Compose file for your OS, open http://localhost:8777/qanything/, set your model endpoint, and upload your first files.
// keep exploring
Similar tools
RAG & Knowledge
wenlan
Agent-fed knowledge base that distills learnings into cited wiki pages.
★ 79 starsApache-2.0
View guide →RAG & Knowledge
KnowNote
A local-first AI knowledge base and NotebookLM replacement built with Electron.
★ 1.2k starsGPL-3.0
View guide →RAG & Knowledge
Langchain-Chatchat
Local knowledge-base RAG and agent app built on LangChain with ChatGLM, Qwen and Llama models
★ 38.7k starsApache-2.0
View guide →RAG & Knowledge
AutoRAG
An agent that finds anything on your computer — federated search over your files, in place.
★ 5.1k starsMIT
View guide →RAG & Knowledge
ecommerce-ai-skills
A trilingual e-commerce knowledge base agents can install.
★ 80 starsCC0-1.0
View guide →// alternatives
Alternatives in RAG & Knowledge
The most-starred other RAG & Knowledge projects in the LibreAI catalogue — the realistic shortlist if QAnything is not the right fit.
- MarkItDown★ 193.5k
MarkItDown — Microsoft tool that converts PDFs, Office files, images and more to clean Markdown for LLMs, MIT
- Dify★ 157.8k
Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or…
- ragflow★ 91.6k
RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context…
- PaddleOCR★ 90.8k
Lightweight OCR toolkit turning PDFs and images into structured, LLM-ready data in 100+ languages
- MinerU★ 82.5k
MinerU — turns PDFs and Office documents into LLM-ready Markdown and JSON for agents and RAG pipelines, Apache-2.0 plus terms
// more in rag
More in RAG & Knowledge
Nearby RAG & Knowledge projects in the catalogue, closest in stars to QAnything — keep exploring the shelf either side of it.
- llmware★ 14.8k
Unified framework for enterprise RAG pipelines with small, specialized models - local, private and secure.
- LEANN★ 13.0k
The smallest vector index in the world: private RAG over everything on your laptop with 97% less storage.
- zvec★ 16.1k
Lightweight, lightning-fast in-process vector database from Alibaba that embeds in your app.
- txtai★ 13.0k
All-in-one AI framework for semantic search, LLM orchestration and language model workflows.
- SurfSense★ 16.3k
Air-gapped, open-source NotebookLM alternative that turns local documents into briefings, decks and podcasts.
- cocoindex★ 11.7k
Incremental engine that keeps RAG indexes and agent context fresh by processing only the delta.
Get QAnything
Free and open — grab the source, star the repo, and start creating.