Weknora
Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
Last verified:
What is Weknora?
WeKnora is Tencent WeChat team's open-source large language model (LLM)-based document understanding and semantic retrieval framework, specifically designed for structurally complex and heterogeneous document scenarios. It uses a modular architecture that integrates multimodal preprocessing, semantic vector indexing, intelligent recall, and large model generative reasoning to build an efficient and controllable document Q&A process. The core retrieval is based on the RAG (Retrieval-Augmented Generation) mechanism, combining contextually relevant fragments with language models to achieve higher quality semantic answers.
The framework features a powerful multimodal cognitive engine that accurately parses mixed text content in PDF, Word, and images, extracting semantic information from text, tables, and images. It supports modular RAG assembly line design with free combination of search strategies, large language models, and vector databases. WeKnora provides precision reasoning and trusted decision assurance through private deployment, multi-round deep contextual understanding, and full-link visual assessment.
WeKnora is designed for enterprise knowledge management teams building internal knowledge bases, intelligent customer service developers integrating into customer service systems, legal and financial professionals analyzing complex documents, academic researchers processing papers and research reports, educators and students building intelligent learning systems, and data/intelligence analysts handling multi-source heterogeneous information.
The tool supports multiple deployment options including local deployment, Docker images, and WeChat Dialog Open Platform integration, ensuring data security and privacy. It provides an out-of-the-box interactive experience with one-click startup scripts and an intuitive Web UI interface that lowers the usage threshold for non-technical users.
Weknora pricing
Pricing model: Freemium
WeKnora is completely free and open-source under the MIT license. There are no paid plans or tiers. Users can download from GitHub (https://github.com/Tencent/WeKnora) and deploy locally at no cost. The framework includes all features: document parsing, semantic retrieval, RAG pipeline, knowledge graph visualization, Web UI, and API access. Users only need to provide their own LLM (local models like Qwen/DeepSeek or third-party API keys) and infrastructure (Docker, server hardware).
Weknora pros
- Open-source under MIT license for commercial use
- Supports PDF, Word, TXT, Markdown, and image formats
- Built-in multimodal cognitive engine for text/table/image extraction
- Modular RAG pipeline with flexible component Combination
- Multiple retrieval strategies: BM25, dense vector, GraphRAG
- Supports Qwen, DeepSeek, and custom LLM integration
- Private/local deployment for data security
- Docker-compatible for easy deployment
- Intuitive Web UI interface for non-technical users
- One-click startup scripts included
- Knowledge graph visualization support
- Multi-round contextual understanding capability
- Full-link visual assessment and monitoring
- Compatible with PostgreSQL pgvector and Elasticsearch
- WeChat official account/miniprogram integration support
Weknora cons
- Requires Docker and Docker Compose installation
- Local deployment needs technical setup knowledge
- Requires git for repository cloning
- Multimodal processing may be resource-intensive
- OCR processing adds complexity to pipeline
- Limited to Chinese documentation primarily
- Knowledge graph feature requires additional configuration
- No hosted cloud service available
Frequently asked questions about Weknora
What is WeKnora's retrieval mechanism?
WeKnora has built-in multiple recall strategies including BM25 keyword retrieval, dense vector retrieval (Dense Retrieve), and knowledge graph enhancement (GraphRAG). These strategies can be freely combined to improve accuracy and recall rate. The system uses advanced semantic vector indexing technology to quickly find the most relevant document fragments to the query.
Can I deploy WeKnora locally?
Yes, WeKnora supports local deployment, Docker images, and WeChat Dialog Open Platform integration. This meets private deployment and offline deployment needs, ensuring data control and security. Users can deploy on their own infrastructure without sending data to external servers.
How to optimize retrieval effects?
To optimize retrieval: upload clear, structured documents with reasonable segmentation; choose appropriate recall strategies (BM25/vector/graph enhancement); use embedding models adapted to the scenario (BGE, GTE, etc.); and adjust the number of recalled items and re-ranking strategies.
Which large language models are supported?
WeKnora defaults to supporting Qwen and DeepSeek. Users can also接入 local LLMs or third-party APIs. The inference mode is configurable with thinking/non-thinking options. The system seamlessly integrates with Ollama platforms and allows flexible switching between mainstream models.
What document formats does WeKnora support?
WeKnora supports PDF, Word, TXT, Markdown, and images (including OCR/Caption). The multimodal parsing engine handles mixed text-image layouts and provides structured extraction for complex documents.
How do I install and start WeKnora?
First ensure Docker, Docker Compose, and Git are installed. Then clone the repository with git clone https://github.com/Tencent/WeKnora.git, cd into the directory, copy .env.example to .env and configure environment variables, then run ./scripts/start_all.sh or make start-all to start all services including Ollama and backend containers.
What are the service access addresses after deployment?
After successful startup, access: Web UI at http://localhost, Backend API at http://localhost:8080, and Link tracing (Jaeger) at http://localhost:16686. The Web UI provides document upload, knowledge base construction, and intelligent Q&A functionality.
Does WeKnora support knowledge graph visualization?
Yes, WeKnora supports converting documents into knowledge graphs that show relationships between different document paragraphs. When enabled, it analyzes and builds semantic relationship networks within documents, helping users understand content and providing structured support for indexing and retrieval, improving relevance and breadth of results.
What vector databases are compatible?
WeKnora is compatible with PostgreSQL (pgvector) and Elasticsearch for vector database needs. The modular design allows flexible configuration of the vector database component based on user requirements.
What enterprise scenarios is WeKnora suitable for?
WeKnora is suitable for: enterprise knowledge management (internal document Q&A, policy retrieval), research literature analysis (paper retrieval, academic data organization), technical support (product manual Q&A, troubleshooting), legal compliance (contract clause and regulation retrieval), and medical knowledge assistance (medical literature and treatment guideline Q&A).