WeKnora — Open Source LLM Knowledge Platform by Tencent

Introduction

WeKnora is an open-source knowledge platform developed by Tencent that turns unstructured documents into intelligent, queryable knowledge bases. It combines RAG retrieval, autonomous reasoning agents, and auto-generated wiki pages to make organizational knowledge accessible through natural language, going beyond simple document chat to provide structured, maintained knowledge systems.

What WeKnora Does

• Ingests documents in PDF, Markdown, HTML, and other formats into a structured knowledge base
• Provides RAG-powered Q&A with source citations and confidence scoring
• Generates and maintains wiki pages automatically from ingested documents
• Supports autonomous reasoning agents that chain multiple knowledge lookups
• Offers multi-tenant access control for team and enterprise deployments

Architecture Overview

WeKnora is built with a Go backend serving a React frontend. Documents are processed through a pipeline that extracts text, chunks it intelligently, generates embeddings, and stores them in a vector database alongside metadata. The query engine uses a hybrid retrieval strategy combining dense vector search with sparse keyword matching, followed by a reranking step. An agent layer orchestrates multi-hop reasoning by breaking complex questions into sub-queries and synthesizing answers from multiple retrieved passages.

Self-Hosting & Configuration

• Deploy with Docker Compose for quick setup with all dependencies included
• Requires PostgreSQL for metadata and Milvus or Weaviate for vector storage
• Supports OpenAI, Anthropic, and Ollama as LLM backends
• Configurable chunking strategies including semantic, fixed-size, and recursive splitting
• Multi-tenant mode with organization-level isolation and RBAC

Key Features

• Hybrid RAG retrieval combining vector similarity and keyword search with reranking
• Auto-generated wiki that stays synchronized with source document updates
• Multi-hop reasoning agent for complex questions spanning multiple documents
• Support for 20+ document formats with intelligent structure preservation
• Built-in evaluation tools for measuring retrieval quality and answer accuracy

Comparison with Similar Tools

• Dify — LLMOps platform with RAG; WeKnora focuses specifically on knowledge management with auto-wiki generation
• AnythingLLM — All-in-one knowledge base; WeKnora offers more sophisticated multi-hop reasoning and enterprise multi-tenancy
• Quivr — RAG framework; WeKnora adds autonomous agents and self-maintaining wiki pages beyond simple retrieval
• Onyx — Broad AI chat with connectors; WeKnora specializes in deep knowledge structuring and reasoning

FAQ

Q: How many documents can WeKnora handle? A: The architecture scales horizontally. Production deployments handle millions of document chunks across distributed vector stores.

Q: Does the auto-wiki feature require manual curation? A: Wiki pages are generated automatically and updated when source documents change. Manual editing is supported for refinements.

Q: Can I use local LLMs instead of cloud APIs? A: Yes, WeKnora supports Ollama and any OpenAI-compatible endpoint for fully private deployments.

Q: Is there an API for programmatic access? A: Yes, all features are accessible via a REST API with OpenAPI documentation.

Sources

• https://github.com/Tencent/WeKnora
• https://weknora.tencent.com