JoyAgent-JDGenie: Technical Report on the GAIA

TL;DR

JoyAgent-JDGenie presents a unified generalist agent architecture that fuses a heterogeneous Plan-Execute/ReAct ensemble, a hierarchical memory stack, and a schema-driven tool suite to achieve robust performance across diverse tasks. On the GAIA benchmark, it attains competitive scores (e.g., Pass@1 of 75.2 and Pass@3 of 82.4 on validation) and narrows the gap to proprietary systems, illustrating the value of system-level integration. The work demonstrates substantial gains from ensemble coordination, memory-aided long-horizon reasoning, and a disciplined tool ecosystem, while also examining exploratory patterns and level-based task biases. It points to promising future directions including dynamic self-improvement via reinforcement learning, autonomous tool evolution, and cross-domain modular transfer to push generalist agents toward real-world, scalable applicability.

Abstract

Large Language Models are increasingly deployed as autonomous agents for complex real-world tasks, yet existing systems often focus on isolated improvements without a unifying design for robustness and adaptability. We propose a generalist agent architecture that integrates three core components: a collective multi-agent framework combining planning and execution agents with critic model voting, a hierarchical memory system spanning working, semantic, and procedural layers, and a refined tool suite for search, code execution, and multimodal parsing. Evaluated on a comprehensive benchmark, our framework consistently outperforms open-source baselines and approaches the performance of proprietary systems. These results demonstrate the importance of system-level integration and highlight a path toward scalable, resilient, and adaptive AI assistants capable of operating across diverse domains and tasks.

Figures (2)

• Figure 1: The overview of the fusion agent architecture.
• Figure 2: The distribution between original level and reassigned level.