// the find
Tencent/WeKnora
Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
WeKnora is a self-hostable RAG + agent platform from Tencent that ingests documents, exposes them via a chat interface, and can auto-generate a structured wiki from your content. It targets teams who want a private knowledge base with LLM Q&A without sending data to a managed SaaS. The scope is enormous — 20+ LLM providers, 9 IM integrations, 7 vector databases, multiple object storage backends.
The CLI is unusually well-built for a project like this — agent-first JSON envelope contract, typed exit codes, dry-run support, and an MCP server surface baked in from day one. The security posture is more serious than typical hobby RAG projects: AES-256-GCM for credentials at rest, SSRF-safe HTTP client explicitly called out, gRPC TLS between app and docreader, and scoped API keys with capability-level grants. The worker-pool architecture (per-stage pools, per-model concurrency governors, DLQ for failed wiki tasks) suggests they've actually run this under real load rather than just demoing it. Docker Compose profiles let you start with just the core and add Neo4j, MinIO, and Langfuse à la carte — that's a reasonable onboarding gradient for a project this large.
The version is 0.7.1 but the feature count already rivals mature enterprise products — that gap between version number and surface area usually means the older subsystems are under-tested when new ones land. It's Go on the backend but the README is heavy with WeChat ecosystem integrations (WeCom, QQBot, Yunzhijia, WeChat Mini Program) that will mean nothing to most Western developers and hints at where the team's attention actually goes. The README changelog is the primary source of architecture knowledge, which means understanding what changed and why requires reading a wall of bullet points across six versions. There's no mention of horizontal scaling — the worker-pool governance is per-instance, so if you outgrow one node you're on your own.