joyagent-jdgenie
开源的端到端产品级通用智能体
A complete, runnable AI agent system from JD.com that takes a plain-language request and delivers a finished report or slide deck by coordinating specialized sub-agents, with no AI plumbing code required from you.
JoyAgent-JDGenie is an open-source AI agent system from JD.com, a major Chinese technology and e-commerce company, that takes a plain-language request and delivers a finished result without requiring you to write any AI glue code yourself. Where most AI agent projects release a toolkit that still demands significant programming before it does anything useful, this project ships a complete working product you can run from the first day.
The clearest way to grasp what it does: if you type something like "give me a trend analysis of the US dollar and gold over the past month," the system goes to work and hands back a finished report, either as a web page or a PowerPoint presentation. It handles the information gathering, reasoning, and formatting without you needing to supervise each step.
Under the hood, a central orchestrator coordinates several specialized sub-agents: one that writes and runs code, one that produces written reports, one that builds slide decks, one that manages files, and others. If your use case needs something the default setup does not cover, you can attach your own sub-agents or tools to the framework without rewriting the core. JD's team tested it on the GAIA benchmark, a standard measure of general AI agent capability, and it scored 75.15% on the validation set and 65.12% on the test set, ahead of several well-known systems from Huawei, Hugging Face, and Hong Kong University.
The full project is open-sourced, including the frontend interface, the backend server, the scheduling engine, and all built-in sub-agents. A recent addition is a multimodal knowledge management module that handles documents, images, and other mixed-format content for question-answering and content generation.
The system is written in Java and does not depend on any proprietary cloud vendor's AI platform, making it more self-contained than some alternatives that require specific infrastructure to function.
Where it fits
- Ask for a trend analysis report on any topic and receive a finished web page or PowerPoint by typing one sentence
- Attach a custom sub-agent or tool to the JDGenie orchestrator without rewriting the core system
- Run a self-contained AI agent stack that does not depend on any specific cloud vendor's AI platform
- Ask questions over a mix of documents and images using the built-in multimodal knowledge module