WeKnora

Tencent's open-source RAG framework that turns raw documents into a queryable knowledge base, ReAct agent, and self-maintaining wiki.

Free· Free, open-source (self-hosted)RAGMulti-model

Visit website →

Best for

Pick WeKnora if you want a Tencent-backed, fully modular open-source RAG stack you can deploy on your own infrastructure with strict data-sovereignty.

Skip if

Skip it if you want a hosted, click-to-deploy RAG SaaS or a small team without the appetite to run a vector DB and LLM endpoints yourself.

WeKnora is an enterprise-grade open-source LLM knowledge platform from Tencent's WeChat team, built around three modes: classic RAG Q&A over a document corpus, a ReAct agent that orchestrates retrieval plus MCP tools and web search for multi-step questions, and a Wiki mode that distills uploaded files into an interlinked, self-maintaining markdown knowledge base with an interactive knowledge graph. It ingests 10+ formats (PDF, Word, Excel, images, etc.) and can auto-sync from Feishu, Notion, and Yuque.

The project is aimed at teams that want a production RAG stack they fully control rather than a hosted SaaS. It is fully modular - you can swap the LLM (OpenAI, DeepSeek, Qwen, Zhipu, Hunyuan, Gemini, MiniMax, NVIDIA, Ollama), the vector database, and the storage backend - and is designed for local or private-cloud deployment so data never leaves your network. Answers can be served back through WeCom, Feishu, Slack, and Telegram, which makes it a natural fit for internal knowledge-bot use cases.

Being a Tencent OSS project, documentation and primary marketing skew Chinese-first, and operating it at scale requires real infra work (vector DB, embeddings, LLM endpoints, observability). It is best thought of as a framework you deploy, not a turnkey product.

Editor's take

WeKnora is one of the more ambitious open RAG frameworks of 2026 - the Wiki and ReAct modes go beyond the usual chat-over-PDF template. It is squarely for engineering teams, not end users, and the China-first docs are real friction, but if you need an auditable, on-prem alternative to hosted RAG platforms it is a serious option.

— The AI Tool Bible editorial team

Pros

✅ Three modes in one stack: RAG Q&A, ReAct agent, and self-maintaining wiki with knowledge graph
✅ Backed by Tencent and actively maintained on GitHub
✅ Pluggable LLMs, vector DBs, and storage; runs fully on-prem
✅ Native connectors for Feishu, Notion, Yuque, plus IM delivery via WeCom/Slack/Telegram
✅ Handles 10+ document formats including PDFs, Office docs, and images