Tencent Open-Sources WeKnora Autonomous Knowledge Base Framework with Model Context Protocol Support

Self-Maintaining Knowledge Graphs and Wiki Mode

Tencent has open-sourced WeKnora, an autonomous enterprise knowledge engine designed for document understanding, retrieval-augmented generation, and recursive agent reasoning. Rather than treating document repositories as passive vector embeddings, WeKnora introduces a Wiki Mode where autonomous agents synthesize incoming raw files into an interlinked markdown knowledge base complete with snapshot diffs, rollback support, and an interactive graph view.

Sandboxed Runtimes and Model Context Protocol 2.x

The framework is engineered around flexible enterprise integration points and strict sandboxing:

• Sandbox Runtimes: Version 0.8.0 isolates execution inside session-persistent Docker, E2B, or Cube containers with granular per-tenant network policies, removing unisolated host-process execution.
• Model Context Protocol Tools: Through the tencent-weknora-mcp package, WeKnora provides 29 tools conforming to the MCP 2.x high-level API, including create_knowledge_from_text and list_shared_knowledge_bases.
• Observability and Ingestion: Native Langfuse OTel tracing tracks pipeline latency and token usage, while connectors support GitLab, Notion, Feishu, and standard Office formats parsed in-process.
• Provider Agnostic: Supports over 20 large language model providers, including LiteLLM, Ollama, DeepSeek, OpenAI, and Gemini, alongside pgvector embeddings.