WeKnora is an open-source LLM-powered knowledge framework developed by Tencent that transforms raw documents into queryable knowledge assets through three core capabilities: RAG-based question answering, autonomous ReAct agents, and an auto-generating Wiki mode. Built primarily in Go, the platform is designed for enterprise-grade document understanding and semantic retrieval with a focus on self-hostability and data sovereignty.
The framework's architecture is fully modular, allowing users to swap LLMs, vector databases, and storage backends according to their needs. It supports ingestion from multiple sources including Feishu, Notion, Yuque, and RSS feeds, with compatibility for over 10 document formats including PDFs, Word documents, images, and Excel files. The platform integrates with 20+ LLM providers spanning OpenAI, DeepSeek, Qwen, Zhipu, Hunyuan, Gemini, MiniMax, NVIDIA, and Ollama, ensuring flexibility in model selection.
WeKnora's Wiki Mode represents a distinctive feature where agents autonomously distill raw documents into self-maintaining, interlinked markdown knowledge bases with interactive knowledge graphs. The ReAct Agent orchestrates multi-step reasoning by combining retrieval, MCP tools, and web search capabilities. For immediate needs, the RAG-based quick Q&A mode handles everyday lookups with semantic search and reranking. The platform includes website embed widgets for publishing agents to external sites and supports serving Q&A through multiple IM channels including WeCom, Feishu, Slack, and Telegram.
Enterprise readiness is addressed through multi-tenant RBAC with a four-tier role matrix (Owner, Admin, Contributor, Viewer), per-resource ownership, and per-tenant audit logging. The system features AES-256-GCM credential encryption and supports both local and private cloud deployment. Langfuse integration provides comprehensive observability into agent reasoning, token usage, and pipeline tracing.
According to GitGenius activity tracking, the repository shows strong engagement with a median issue and PR response latency of 0.0 hours and a mean of 11.5 hours across 696 tracked items. The most active issue categories are bugs (237), questions (146), and enhancements (136). Primary contributors include lyingbug with 633 tracked events, begoniezhao with 260 events, and voidkey with 27 events. The repository shares overlapping contributors with langgenius/dify, xerrors/yuxi, and hiyouga/llamafactory, indicating cross-pollination within the LLM and knowledge management ecosystem.
Recent versions have introduced significant capabilities including website embed widgets with secure token exchange, document multi-tagging and batch reparsing, Wiki folder hierarchies, RSS data sources, MCP OAuth2 support, EPUB and MHTML parsing, and per-upload process configuration. The platform classifies as a distributed system with message queue, asynchronous messaging, and high-performance characteristics, supporting data streaming and publish-subscribe patterns through its modular middleware architecture. Version 0.6.3 represents the current stable release with ongoing development focused on expanding data source connectors and enhancing the agent reasoning pipeline.
