A Go knowledge platform that auto-writes its own wiki

What it does

WeKnora is a Go-based knowledge framework that ingests documents from Feishu, Notion, Yuque, and local files, then exposes them through three modes: a standard RAG chat, a ReAct agent that can call MCP tools and search the web, and a Wiki mode where agents automatically generate interlinked Markdown pages and a knowledge graph. It wraps the whole thing in enterprise multi-tenancy with a four-tier RBAC matrix, audit logs, and AES-256-GCM credential encryption.

The interesting bit

The Wiki mode is the unusual part: instead of just retrieving chunks, the system distills raw documents into structured, interlinked markdown that it manages itself, complete with a visual graph. That, plus the fact that every component—LLM, vector DB, object storage—is swappable, makes it feel like a knowledge operating system rather than a chat-with-PDF wrapper.

Key highlights

• Supports 20+ LLM providers including OpenAI, DeepSeek, Qwen, Gemini, and Ollama, plus seven vector databases from pgvector to Milvus and Tencent VectorDB.
• Enterprise RBAC with Owner/Admin/Contributor/Viewer roles, per-knowledge-base ownership, and per-tenant audit logs.
• Native IM integrations: WeCom, Feishu, Slack, Telegram, DingTalk, Mattermost, and WeChat.
• Full Langfuse observability for tracing ReAct loops, token usage, and tool calls.
• Runs fully offline via local or private-cloud deployment with Helm charts and full data sovereignty.

Verdict

Worth evaluating if you need an on-premise, multi-tenant knowledge platform with agentic reasoning and auto-generated documentation. Skip it if you just want a quick Python script to chat with one PDF.