Tencent WeKnora open-sources enterprise LLM knowledge platform
For users needing to manage and query internal documents, WeKnora provides a robust, open-source alternative to proprietary solutions, with features like secure multi-tenant access and built-in evaluation tools.
Tencent's WeKnora, an enterprise-level LLM knowledge platform, has been open-sourced.
It transforms scattered documents into three assets: a RAG knowledge base for question-answering, a ReAct Agent for autonomous reasoning, and an automatically evolving Wiki.
The platform is built on the WeChat Open Platform's core technology framework and offers features like private deployment, data autonomy, and multi-tenant support with RBAC.
腾讯微信团队开源了「企业级 LLM 知识平台」：WeKnora WeKnora 把散落的原始文档转化为三种知识库资产：可问答的 RAG 知识库、可自主推理的 ReAct Agent、可持续演进的自动 Wiki。并且它是微信对话开放平台的核心技术框架，有真实生产背书。 项目地址： github.com/Tencent/WeKnora WeKnora 解决什么问题？ 企业知识天然是碎片化的：文档散落在飞书、GitLab、Notion、语雀、钉钉、企微、RSS 等各处。WeKnora 的答案是把这些内容“沉淀为可查询、可推理、可持续演进的专属知识资产”，同时强调私有化部署、数据完全自主可控。 # 三个核心能力 1. RAG 快速问答：日常查询的主路径。检索上支持 BM25 稀疏检索、稠密向量、GraphRAG、父子分块等策略组合；向量层默认用 pgvector（HNSW 索引，1024 维），可切换 8 种向量库并做多库扇出检索。问答带引用溯源、引用气泡、建议问题，还内置了用召回命中率 + BLEU/ROUGE 做 E2E 效果评测的工具——这在同类项目中少见，说明团队认真对待“检索质量可度量”这件事。 2. ReAct Agent 智能推理：面向复杂多步任务。Agent 可自主编排知识检索、MCP 工具、网络搜索（11 家搜索源）、租户技能目录，以及 v0.8 新增的会话级持久沙箱（Docker / E2B / Cube 三种后端，v0.8 起移除了本地宿主进程后端，安全上更收紧）。配套跨会话长期记忆（画像/偏好/事实/任务/兴趣五类，自动抽取 + 用户确认）。 3. Wiki 模式：WeKnora 最差异化的能力。Agent 自动把文档蒸馏成相互链接的 Markdown 知识库，配可视化知识图谱、手动编辑、版本历史与一键回滚，官方称可支撑 4 万文档级的知识库。大多数 RAG 项目止步于“问答”，WeKnora 额外给出了“知识再生产”的输出形态；知识库不只是被查，而是被持续整理成结构化资产。 企业级能力 · 多租户 + 四级 RBAC（Owner / Admin / Contributor / Viewer），按知识库的 ownership 隔离，租户级审计日志； · API Key 作为一等主体：可限定作用域的密钥 + principal 模型，适合 headless/CI 集成； · 安全：凭据 AES-256-GCM 加密、gRPC/Redis TLS、SSRF 加固的 HTTP 客户端、OIDC JWKS 验证、沙箱隔离； · 知识库工程化：分块可编辑、带逐版本 diff 与回滚、自动重索引，自动打标、批量重解析——把知识库当“代码仓库”来管理，这个思路成熟。 WeChat @Weixin_WeChat Put your team’s knowledge to work. Developed by Tencent’s Weixin team, WeKnora is an open-source, LLM-powered knowledge framework built for enterprise-grade document understanding, semantic retrieval, and autonomous reasoning. Build assistants that find answers in your docs, show their sources and use skills to create files you can preview and download. Keep useful context across chats with opt-in long-term memory. You approve suggested memories before they’re used. Self-host it and run skills in a sandbox with network limits you set. Explore WeKnora: github.com/Tencent/WeKnora Have a project in mind? Join our community to share ideas and build with us: github.com/Tencent/WeKnor…
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include WeKnora, Tencent, RAG, ReAct Agent.
Source attribution: curated from shao__meng (@shao__meng). Cross-source corroboration: 3 independent sources have covered this development (including pandaily, Yangyi), confirming industry-wide signal strength.
