JoyAgent: Open‑Source Enterprise‑Grade Multi‑Agent Platform from JD

Project Overview

JoyAgent is JD’s self‑developed intelligent‑agent engine platform. The multi‑agent module AutoBots (JDGenie) was open‑sourced in July 2024, and the DataAgent capabilities were added in September. The source code is hosted at https://github.com/jd-opensource/joyagent-jdgenie (over 11 k stars).

Key Characteristics

100% Open‑Source : The entire stack—frontend, backend, framework, engine, and core sub‑agents—is released, allowing rapid deployment of enterprise‑grade multi‑agent solutions.

High Availability : New agents or tools can be mounted directly; built‑in sub‑agents support HTML, PPT and Markdown output formats.

Lightweight : Low coupling with MaaS or cloud platforms enables independent local deployment.

Strong Performance : On the GAIA benchmark (validation set) JoyAgent achieves >75 % accuracy, outperforming OWL and SmolAgent.

Proven Reliability : Tested with >20 000 agent instances in JD’s internal scenarios.

JoyAgent DataAgent

DataAgent addresses gaps in existing data‑related agents by providing end‑to‑end open‑source solutions for data governance (DGP protocol), diagnostic analysis, and work recommendations. It supports adaptive querying across detail and metric tables and visualizes results with charts. In the BirdSQL public leaderboard test set, it ranked 7th out of 84 teams with 75.35 % accuracy.

Core Features

User Guidance – assists users in formulating queries they may not know how to ask.

DGP Protocol – defines principles for table, field, and value design to ensure data quality and consistency.

Data Lineage – parses SQL scripts to build lineage graphs for retrieval‑augmented generation.

Semantic Alignment & Pre‑weaving of Metric Data – unifies dimensions and resolves definition conflicts for precise metric retrieval.

Intelligent Querying – supports both detail and metric tables and returns visual chart outputs.

Diagnostic Analysis – provides attribution methods (trend, cycle, anomaly, correlation, causality) and SOPPlan for custom analysis workflows.

Deployment Experience

After deployment, JoyAgent guides users on possible queries, visualizes answers (e.g., line charts), and generates diagnostic reports in roughly 20 minutes, dramatically reducing manual effort while delivering actionable insights.