What it does

WeKnora is a Go-based knowledge platform that ingests documents from Feishu, Notion, Yuque, or direct upload, then exposes them through three modes: plain RAG Q&A, a ReAct agent that orchestrates retrieval with web search and MCP tools, and a “Wiki Mode” where agents auto-generate interlinked Markdown pages with a visual knowledge graph. It ships with a Web UI, REST API, CLI (weknora), Chrome extension, WeChat mini-program, and plugs into IM channels including Slack, Telegram, and WeCom.

The interesting bit

The Wiki Mode is the unusual angle: instead of just retrieving chunks, agents distill raw documents into structured, interlinked pages that maintain themselves. Combined with v0.6.0’s four-tier tenant RBAC (Owner/Admin/Contributor/Viewer), per-KB ownership, and audit logs, it’s clearly built for teams that actually need access control—not just hobbyist RAG demos.

Key highlights

• Modular to a fault: swap LLMs (20+ providers including Ollama), vector DBs (pgvector, Milvus, Weaviate, Qdrant, etc.), and object storage backends
• Enterprise guardrails: AES-256-GCM credential encryption, gRPC TLS, SSRF-safe HTTP client, sandboxed agent skills, and Langfuse observability for tracing ReAct loops
• Document breadth: PDF, Word, images, Excel, PPT, plus audio ASR; adaptive 3-tier chunking with live preview
• Deployment flexibility: local, Docker, Kubernetes via Helm, or WeKnora Cloud for hosted LLM/parsing
• CLI v0.4 GA: includes mcp serve for running MCP servers directly

Caveats

• Some connectors are marked “more data sources coming soon”—the roadmap is visible but not fully delivered
• Wiki ingest at 40k-document scale only arrived in v0.5.2; earlier versions presumably struggled with large KBs
• The README is thorough but reads like feature inventory; actual performance benchmarks (latency, throughput) aren’t provided

Verdict

Worth evaluating if you’re building team-facing knowledge tools and need multi-tenancy plus IM integration out of the box. Skip it if you want a lightweight personal RAG—this is closer to a self-hosted Glean competitor with Go ops ergonomics.