MiroThinker is an open-source deep research agent framework with models supporting a 256K context window and up to 300 to 600 tool interactions per task. It targets researchers and developers working on complex deep research and prediction tasks requiring long-horizon reasoning.
Project overview
The framework applies interactive scaling alongside long-horizon reasoning to address complex research tasks, supported by tools for web search, content extraction, and code execution.
Project type
AI Agent · AI Search · Infrastructure
Use cases
Search & Research
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Researchers, developers, and AI engineers who need an open-source framework for long-horizon reasoning tasks with extensive tool interactions and have coding knowledge and paid API access.
Key capabilities
Supports a 256K context window for deep multi-step analysis.
Handles up to 300 to 600 tool interactions per task.
Fully open-source framework providing tools for web search, content extraction, and code execution.
Comprehensive logging and analysis of agent interactions for SFT and DPO.
Supports uploading and processing file formats such as .pdf, .doc, .ppt, .xls, and .jpg in the online version.
Generates, previews, and shares deep research reports.
Limitations and risks
The agent's Chinese language capability is limited because the training data mainly contains English.
The framework depends on paid external API services (Serper, E2B, Jina) for core functionality.
Getting started
Clone the repository, install dependencies using the uv package manager, configure the .env file with necessary API keys (SERPER, JINA, E2B, etc.), and run the evaluation script. Python is required.
Evidence and sources
GitHub project description: MiroThinker is a deep research agent optimized for complex research and prediction tasks. Our latest models, MiroThinker-1.7, achieves 74.0 and 75.3 on the BrowseComp and BrowseCo…
README: MiroThinker will continue to be maintained and iteratively upgraded, with the goal of becoming the best Research Agent you'll ever use!
README: Unlike previous agents that scale only model size or context length, MiroThinker v1.0 introduces **interactive scaling** at the agent level, systematically training the agent to h…
README: **MiroThinker**: A deep research agent optimized for research and prediction. It achieves a 88.2 on the challenging BrowseComp benchmark. See [Quick Start](#-quick-start).
README: 🚀 MiroThinker-1.7 supports a 256K context window, long-horizon reasoning, and deep multi-step analysis.