Cognitive Kernel-Pro: A Framework for Deep Research Agents and Agent Foundation Models Training

2025-08-01

AI Summary

Existing AI agent systems predominantly rely on proprietary models or paid APIs, severely limiting accessibility and reproducibility. Method: This paper introduces the first fully open-source, zero-cost multi-module agent framework supporting complex reasoning, web interaction, code execution, and autonomous research. It innovatively constructs a high-quality agent training dataset spanning queries, interaction trajectories, and verifiable answers. For the first time, it jointly integrates test-time reflection and multi-path voting within an open-source agent architecture. The framework leverages open-weight LLMs and employs four complementary data construction techniques—web, file, code, and general reasoning—to enhance capability and robustness. Contribution/Results: On the GAIA benchmark, our 8B-parameter model outperforms WebDancer and WebSailor, achieving state-of-the-art performance among open-source, free agents. This work establishes a new paradigm for open, robust, and reproducible AI agent research.

Technology Category

• Multiagent Systems: Adversarial Agents
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

General AI Agents are increasingly recognized as foundational frameworks for the next generation of artificial intelligence, enabling complex reasoning, web interaction, coding, and autonomous research capabilities. However, current agent systems are either closed-source or heavily reliant on a variety of paid APIs and proprietary tools, limiting accessibility and reproducibility for the research community. In this work, we present extbf{Cognitive Kernel-Pro}, a fully open-source and (to the maximum extent) free multi-module agent framework designed to democratize the development and evaluation of advanced AI agents. Within Cognitive Kernel-Pro, we systematically investigate the curation of high-quality training data for Agent Foundation Models, focusing on the construction of queries, trajectories, and verifiable answers across four key domains: web, file, code, and general reasoning. Furthermore, we explore novel strategies for agent test-time reflection and voting to enhance agent robustness and performance. We evaluate Cognitive Kernel-Pro on GAIA, achieving state-of-the-art results among open-source and free agents. Notably, our 8B-parameter open-source model surpasses previous leading systems such as WebDancer and WebSailor, establishing a new performance standard for accessible, high-capability AI agents. Code is available at https://github.com/Tencent/CognitiveKernel-Pro

Problem

• Develop open-source framework for advanced AI agent training
• Improve accessibility and reproducibility in AI agent research
• Enhance agent robustness via reflection and voting strategies

Innovation

• Open-source multi-module agent framework
• High-quality training data curation
• Novel reflection and voting strategies