WeKnora is an LLM-powered framework designed for deep document understanding, semantic retrieval, and context-aware question answering, leveraging the RAG paradigm. It targets enterprise knowledge management, research, technical support, legal, and medical domains, offering efficient and controllable document Q&A pipelines.

How It Works

WeKnora employs a modular architecture for a complete document understanding and retrieval pipeline. It integrates multi-modal preprocessing, semantic vector indexing, intelligent retrieval, and LLM inference. The core retrieval process uses RAG, combining contextually relevant document snippets with LLMs for high-quality semantic answers. This approach allows for flexible configuration and extension of each component.

Quick Start & Requirements

• Installation: Clone the repository, copy .env.example to .env and configure it, then run ./scripts/start_all.sh or make start-all.
• Prerequisites: Docker, Docker Compose, Git.
• Access: Web UI at http://localhost, API at http://localhost:8080.
• Documentation: API 文档

Highlighted Details

• Supports PDF, Word, Txt, Markdown, and images (with OCR/Caption) for document parsing.
• Integrates with various embedding models (local, BGE/GTE API) and vector databases (PostgreSQL/pgvector, Elasticsearch).
• Offers hybrid retrieval strategies including BM25, Dense Retrieve, and GraphRAG.
• Allows integration with LLMs like Qwen and DeepSeek, supporting local deployments (e.g., via Ollama).
• Provides end-to-end testing support with visualization and metrics (Recall, BLEU/ROUGE).
• Features a Web UI and RESTful API for ease of use.

Maintenance & Community

The project welcomes community contributions via Issues and Pull Requests, with guidelines for bug fixes, new features, documentation improvements, and UI/UX optimizations. Code contributions should follow Go Code Review Comments and use gofmt. Commit messages should adhere to Conventional Commits.

Licensing & Compatibility

The project is released under the MIT license, allowing free use, modification, and distribution with attribution.

Limitations & Caveats

The README does not explicitly detail limitations or known issues. The project is developed by Tencent, suggesting potential enterprise-grade backing but also a possible focus on internal use cases.