MiroFlow: Towards High-Performance and Robust Open-Source Agent Framework for General Deep Research Tasks

2026-02-26

AI Summary

Current large language models exhibit limited performance on complex tasks requiring interaction with external tools and dynamic environments. Prevailing agent frameworks often suffer from rigid workflows, poor stability, narrow task coverage, and reliance on costly commercial APIs. To address these limitations, this work proposes a high-performance, fully open-source agent framework that introduces a flexibly composable agent graph structure, an optional deep reasoning mode, and a robust workflow execution mechanism. These innovations significantly enhance the system’s autonomy, robustness, and cross-task generalization in complex scenarios. Built upon open-source large language models and integrating diverse tool-calling interfaces without dependence on commercial APIs, the framework achieves state-of-the-art results across multiple authoritative benchmarks—including GAIA, BrowseComp-EN/ZH, HLE, xBench-DeepSearch, and FutureX—establishing a reproducible and easily comparable open-source baseline for agent research.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Multiagent Systems: Agent Communication
• 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
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Despite the remarkable progress of large language models (LLMs), the capabilities of standalone LLMs have begun to plateau when tackling real-world, complex tasks that require interaction with external tools and dynamic environments. Although recent agent frameworks aim to enhance model autonomy through tool integration and external interaction, they still suffer from naive workflows, unstable performance, limited support across diverse benchmarks and tasks, and heavy reliance on costly commercial APIs. In this work, we propose a high-performance and robust open-source agent framework, termed MiroFlow, which incorporates an agent graph for flexible orchestration, an optional deep reasoning mode to enhance performance, and a robust workflow execution to ensure stable and reproducible performance. Extensive experiments demonstrate that MiroFlow consistently achieves state-of-the-art performance across multiple agent benchmarks, including GAIA, BrowseComp-EN/ZH, HLE, xBench-DeepSearch, and notably FutureX. We hope it could serve as an easily accessible, reproducible, and comparable baseline for the deep research community.

Problem

• agent framework
• large language models
• tool integration
• workflow stability
• benchmark diversity

Innovation

• agent framework
• agent graph
• deep reasoning
• robust workflow
• open-source