WeKnora
An open-source, self-hosted platform from Tencent that turns a pile of company documents into a searchable knowledge base with AI Q&A — built to plug into a very wide range of AI models and vector databases.
🔗 Visit WeKnora
Description
Enterprises accumulate huge document archives that are technically "searchable" but practically useless — nobody can find the right answer buried in thousands of PDFs and wikis. WeKnora addresses that by combining document ingestion, AI-powered question answering, and automatic wiki generation into one self-hosted platform.
WeKnora is an open-source platform built by Tencent that ingests 10+ document formats (PDF, Word, Excel, images, Markdown, EPUB and more), syncs from sources like Feishu, Notion, Yuque and RSS feeds, and answers questions using retrieval-augmented generation (RAG) plus autonomous ReACT-style multi-step reasoning agents. It auto-generates a browsable Wiki with knowledge graphs, offers 4-tier role-based access control and workspace audit logging, and is notably flexible on the backend — it supports 20+ LLM providers (OpenAI, Claude, DeepSeek, Gemini, Ollama and more) and 8+ vector database backends (pgvector, Elasticsearch, Milvus, Qdrant, Weaviate and others). It has over 20,500 GitHub stars.
💬 Our review
The short version: WeKnora is a genuinely capable, feature-dense RAG platform, but its documentation and integrations lean heavily toward the Tencent ecosystem (WeChat, Feishu, WeCom), so its fit depends a lot on whether your organization already lives in that world.
The differentiator versus a typical RAG framework like LlamaIndex or a platform like Dify is breadth of backend support — 20+ LLM providers and 8+ vector databases means you're not locked into a specific AI vendor or database choice, and the ReACT agent layer with 29 official MCP tools goes beyond simple Q&A into genuine multi-step task execution. The auto-generated Wiki with knowledge graphs is also a step up from plain chat-based RAG, giving you a browsable, structured output rather than just a chatbot.
The honest limits: it's operationally heavy — the project itself documents 150+ environment variables and expects Docker/Kubernetes infrastructure, so this is not a five-minute self-host. Its GitHub license field shows no standard SPDX license detected, so teams should check the actual license terms in the repository carefully before adopting it, especially for commercial use. Documentation and integrations also assume familiarity with Feishu/WeChat/WeCom, which is a real friction point outside organizations already using those tools. For a well-resourced team wanting a highly flexible, self-hosted RAG platform with real agentic capabilities, it's worth evaluating; for a smaller team wanting a quick RAG setup, a lighter framework will get you running faster.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Plateforme open-source, auto-hébergée (vérifier les termes de licence exacts sur le dépôt — champ licence GitHub non standard) ; option WeKnora Cloud hébergée mentionnée.
Pros
Supporte 20+ fournisseurs LLM et 8+ bases vectorielles — pas d'enfermement fournisseur
Agents ReACT multi-étapes avec 29 outils MCP officiels, au-delà du simple Q&A
Génération automatique de Wiki avec graphes de connaissances
RBAC à 4 niveaux et journalisation d'audit par espace de travail
Cons
Lourd opérationnellement — 150+ variables d'environnement, infra Docker/Kubernetes requise
Licence GitHub non standard (NOASSERTION) — à vérifier avant usage commercial
Documentation et intégrations orientées écosystème Tencent (WeChat, Feishu, WeCom)