Youdao QAnything
Open-source RAG engine for document Q&A, with private deployment
Overview
QAnything is NetEase Youdao's open-source (AGPL-3.0) local-knowledge-base Q&A system: upload PDFs, Office docs, images or web links and query them via a two-stage retrieval-and-rerank pipeline, self-hosted via Docker or as a hosted SaaS at qanything.ai, with a public API.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 4 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Full RAG pipeline
- Offline/local-LLM support
- OpenAI-compatible API
- Managed hosting
- Capped usage quotas
- Dedicated deployment support
- Custom scale/integration
What you can produce with Youdao QAnything
- Multi-format ingestion (PDF, Word, PPT, Excel, email, images, web links)
- Two-stage retrieval + BCEmbedding reranking (bilingual CN/EN)
- Self-hosted Docker deployment (offline-capable)
- Hosted SaaS at qanything.ai
- OpenAI-compatible REST API
- AGPL-3.0 open-source license
- ~14,000 GitHub stars
QAnything's real selling point is the deployment story: it's genuinely open-source under AGPL-3.0, runs self-hosted via Docker Compose on Linux, Mac or Windows, and can go fully offline with a local LLM like Ollama -- which matters a lot to organizations that can't ship documents to a third-party API. The retrieval layer uses Youdao's own BCEmbedding models with bilingual Chinese-English reranking, and format coverage is wide: PDF, Office files, images, email, web links.
It's a developer tool wearing a SaaS coat rather than the reverse -- non-technical teams will default to the hosted qanything.ai tier, which comes with usage limits, and getting real value out of the self-hosted version means comfort with Docker and, ideally, GPU infrastructure. With roughly 14,000 GitHub stars it's one of the better-known Chinese open-source RAG stacks, but it's a building block, not a finished product.
What people say
QAnything (Question and Answer based on Anything) is NetEase Youdao's open-source retrieval-augmented-generation system, released in January 2024 and still actively maintained under the AGPL-3.0 license -- currently sitting around 14,000 stars on GitHub. The pitch is a local knowledge base that answers questions against a user's own files rather than the open web: upload PDFs, Word, PowerPoint, Excel, Markdown, email (.eml), plain text, images or paste in a web link, and the system indexes and retrieves from all of it.
Under the hood, QAnything runs a two-stage retrieval pipeline built on Youdao's own BCEmbedding and reranking models, which the project specifically markets on bilingual Chinese-English retrieval quality -- a genuine differentiator versus RAG stacks built purely around English-tuned embeddings. The self-hosted version deploys via Docker Compose on Linux, macOS or Windows and can run fully offline paired with a local model such as Ollama, which independent comparisons of open-source RAG projects (alongside RAGFlow, LlamaIndex and others) call out as a meaningful advantage for data-sensitive deployments -- legal, healthcare, government and enterprise users who can't route documents through a hosted API.
Beyond self-hosting, Youdao runs a hosted SaaS version at qanything.ai for users who don't want to manage infrastructure, and a documented REST API (OpenAI-compatible interface) at the self-hosted instance's /qanything/ endpoint for developers building it into other products. There's a listed enterprise contact channel for organizations wanting custom or larger-scale deployment, though -- as with most open-source-plus-enterprise-tier products -- no public enterprise pricing is posted; larger deployments go through a direct sales conversation.
There isn't much in the way of independent consumer reviews, because this isn't really a consumer product -- the audience is developers and IT teams evaluating RAG infrastructure, and the evidence that matters is the GitHub activity, the format/language coverage, and the offline-deployment option, all of which check out. Where it's a genuinely stronger open pick than most is the combination of solid Chinese-English retrieval quality with a real, working self-hosted path rather than an open-source repo that only nominally supports it.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Youdao QAnything's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Youdao QAnything unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Youdao QAnything directly →
Spotted something out of date? Suggest an update →
Alternatives to Youdao QAnything
More in Research & Data
