yuezhiai/jonex

What it solves

Jonex is an enterprise-grade AI knowledge platform designed to turn raw, multimodal content into structured, reusable knowledge services. It addresses the challenge of transforming unstructured data (documents, videos, audio) into a format that AI agents can use for reasoning and source-grounded retrieval, ensuring that enterprise knowledge is traceable and governed.

How it works

The platform uses a pipeline that combines multimodal parsing and knowledge compilation. It integrates with tools like MinerU and RAG-Anything for content processing, and employs a "dual engine" approach using a Wiki and Graph Ontology to compile domain reasoning into the knowledge layer before retrieval. For storage and retrieval, it utilizes Milvus for vector indexing and Neo4j for graph persistence, supporting hybrid search modes that provide answers with reasoning traces and structured source references.

Who it’s for

It is built for organizations and developers creating enterprise AI agents that require high-precision, domain-specific knowledge bases grounded in multimodal data.

Highlights

• Multimodal Parsing: Supports processing of documents, audio, and video (via ASR and vision-language models).
• Ontology-First Retrieval: Compiles domain reasoning into a knowledge layer to improve retrieval accuracy.
• Enterprise Ready: Features multi-tenant support, a unified API gateway, and independent scaling of parsing workers.
• Extensible Integration: Connects with LightRAG, Neo4j, Milvus, and any OpenAI-compatible LLM/embedding provider.