Script Planner Assistant▸ 
AI Apps 
AI News 
MiroFlow: A High-Performance Multi-Agent Framework by MiroMind
MiroFlow: A High-Performance Multi-Agent Framework by MiroMindIntroduction
MiroFlow is a high-performance multi-Agent framework publicly released by the MiroMind team. It turns various large models into AI assistants capable of deep research, similar to Openai Deep Research. Simply put, it enables models to arrange their own steps, use tools, check results, and complete multi-step complex tasks—just like a person doing research.
MiroFlow Core Capabilities
Multi-tool collaboration: It can combine web search, Python execution, file reading, code invocation, charting, and other tools, putting the entire workflow to use.
Main-sub Agent division of labor: The main Agent breaks down tasks and assigns them to sub-Agents that specialize in browsing, writing code, or doing calculations, then collects the results. This allows different things to be done in parallel and more professionally.
High concurrency and fault tolerance: By using asynchronous queues along with retries, timeouts, and rollback, even if an interface gets stuck or the network is unstable, tasks can still be completed smoothly.
Transparent and explainable: There is a web panel where you can watch the model's reasoning process and the tools it uses in real time on the left, and view the final answer on the right, making it easy to debug issues and build trust.
Streaming output: The model's thinking, tool progress, and intermediate results can be seen immediately, making the experience feel more like ChatGPT's advanced mode.
Key Metrics
In the GAIA deep research benchmark, the pass rate is 72.2%–82.4%, which is the best reproducible result among currently available open-source Agent frameworks.
Low deployment cost: With the open-source MiroThinker model, a single RTX 4090 can run the entire service stack—no need to purchase closed-source paid models.
MiroFlow Use Cases
Academic research: Automatically find literature, extract key points, write surveys, and assist with experiments.
Business analysis: Combine financial reports, news, and social data to produce investment reports or competitive analyses.
Education or programming: Help with homework, write and debug code, and create data visualizations.
MiroFlow Ecosystem
MiroFlow is part of the MiroMind AI ecosystem, which also includes:
MiroThinker: An open-source base model natively capable of reasoning with tools.
MiroVerse: Contains 147,000 high-quality open-source training data entries for research on agent training.
MiroTrain / MiroRL: Tools that help with agent model training and stability.
MiroFlow Quick Start
1. Clone the repository and set up the environment:
git clone https://github.com/MiroMindAI/MiroFlow && cd MiroFlow uv sync
2. Configure the API key:
cp .env.template .env # Edit .env and add OPENROUTER_API_KEY
3. Run your first agent:
uv run main.py trace --config_file_name=agent_quickstart_reading --task="What is the first country listed in the XLSX file that have names starting with Co?" --task_file_name="data/FSI-2023-DOWNLOAD.xlsx"
GitHub repository: https://github.com/MiroMindAI/Miroflow
Demo: https://miromindai.github.io/MiroFlow/
