cmbagent

Autonomous Research System, Powered by AG2.

🎉 News: Cmbagent won a first place award at the NeurIPS 2025 Fair Universe Competition.

Cmbagent is the autonomous research engine for Denario, our end-to-end research system.

We are currently deploying cmbagent on the cloud, it will be in production soon!

Check our demo videos on YouTube!

Join our Discord Server to ask all your questions!

This is open-source research-ready software.

• Check the demo notebooks.

• Best performances are obtained with top-scoring models.

We emphasize that cmbagent is under active development and apologize for any bugs.

The backbone of cmbagent is AG2. Please star the AG2 repo ⭐ and cite Wu et al (2023)!

Strategy

Cmbagent acts according to a Planning and Control strategy with no human-in-the-loop.

You give a task to solve, then:

Planning

• A plan is designed from a conversation between a planner and a plan reviewer.
• Once the number of feedbacks (reviews) is exhausted the plan is recorded in context and cmbagent switches to control.

Control

• The plan is executed step-by-step.
• Sub-tasks are handed over to a single agent in each step.

Installation

With Python 3.12 or above:

python3 -m venv cmbagent_env
source cmbagent_env/bin/activate
pip install cmbagent

Optional Dependencies

# Development tools (pytest, ipython, jupyter)
pip install cmbagent[dev]

# Local execution (if not using the web UI)
pip install cmbagent[local]

# Domain-specific packages (for specialized agents or local execution)
pip install cmbagent[materials]  # pymatgen, ASE, phonopy, matminer
pip install cmbagent[biochem]    # BioPython, RDKit, MDAnalysis, ProDy
pip install cmbagent[astro]      # CAMB, AstroPy, HEALPix, Cobaya
pip install cmbagent[data]       # scipy, scikit-learn, seaborn, plotly

# Everything
pip install cmbagent[all]

# Or combine multiple domains
pip install cmbagent[materials,biochem,data]

Installation for developers

git clone https://github.com/CMBAgents/cmbagent.git
cd cmbagent
python3 -m venv cmbagent_env
source cmbagent_env/bin/activate
pip install -e .

You can then open the folder in your VSCode/Cursor/Emacs/... and work on the source code.

To install optional domain-specific packages (examples provided, add your own in pyproject.toml):

# Single domain
pip install -e .[materials]
pip install -e .[biochem]
pip install -e .[astro]
pip install -e .[data]

# Or combine multiple domains
pip install -e .[materials,biochem,data]

Run

We assume you are in the virtual environment where you installed cmbagent.

Here is a one-liner you can run in terminal:

python -c "import cmbagent; task='''Draw two random numbers and give me their sum'''; results=cmbagent.one_shot(task, agent='engineer', engineer_model='gpt-4o-mini');"

If you want to run the notebooks, first create the ipykernel (assuming your virtual environment is called cmbagent_env):

pip install ipykernel jupyterlab
python -m ipykernel install --user --name cmbagent_env --display-name "Python (cmbagent_env)"

Then launch jupyterlab:

jupyter-lab

Select the cmbagent kernel, and run the notebook.

API Keys

Before you can use cmbagent, you need to set your OpenAI API key as an environment variable. Do this in a terminal, before launching Jupyter-lab.

For Unix-based systems (Linux, macOS), do:

export OPENAI_API_KEY="sk-..."  ## mandatory for the RAG agents
export ANTHROPIC_API_KEY="sk-..." ## optional
export GOOGLE_APPLICATION_CREDENTIALS="/path/to/service-account-key.json" ## optional for Vertex AI

(paste in your bashrc or zshrc file, if possible.)

For Windows, use WSL and the same command.

By default, cmbagent uses models from oai/anthropic/google. If you want to pick different LLMs, just adapt agent_llm_configs as above, or the default_agent_llm_configs in utils.py.

References

@software{CMBAGENT_2025,
            author = {Boris Bolliet},
            title = {CMBAGENT: Open-Source Multi-Agent System for Science},
            year = {2025},
            url = {https://github.com/CMBAgents/cmbagent},
            note = {Available at https://github.com/CMBAgents/cmbagent},
            version = {latest}
            }

@misc{xu2025opensourceplanning,
    title={Open Source Planning & Control System with Language Agents for Autonomous Scientific Discovery},
    author={Licong Xu and Milind Sarkar and Anto I. Lonappan and Íñigo Zubeldia and Pablo Villanueva-Domingo and Santiago Casas and Christian Fidler and Chetana Amancharla and Ujjwal Tiwari and Adrian Bayer and Chadi Ait Ekiou and Miles Cranmer and Adrian Dimitrov and James Fergusson and Kahaan Gandhi and Sven Krippendorf and Andrew Laverick and Julien Lesgourgues and Antony Lewis and Thomas Meier and Blake Sherwin and Kristen Surrao and Francisco Villaescusa-Navarro and Chi Wang and Xueqing Xu and Boris Bolliet},
    year={2025},
    eprint={2507.07257},
    archivePrefix={arXiv},
    primaryClass={cs.AI},
    url={https://arxiv.org/abs/2507.07257},
}


@misc{Laverick:2024fyh,
  author = "Laverick, Andrew and Surrao, Kristen and Zubeldia, Inigo and Bolliet, Boris and Cranmer, Miles and Lewis, Antony and Sherwin, Blake and Lesgourgues, Julien",
  title = "{Multi-Agent System for Cosmological Parameter Analysis}",
  eprint = "2412.00431",
  archivePrefix = "arXiv",
  primaryClass = "astro-ph.IM",
  month = "11",
  year = "2024"
}

Acknowledgments

Our project is funded by the Cambridge Centre for Data-Driven Discovery Accelerate Programme. We are grateful to Mark Sze for help with AG2.