Multi-agent system for data analysis, made by cosmologists, powered by autogen.

Note: This software is under MIT license. We bear no responsibility for any misuse of this software or its outputs.

See our preliminary documentation and examples here to have a preview of our work.

Contributed by:

• Boris Bolliet (Cambridge)
• Andrew Laverick (Independent)
• Inigo Zubeldia (Cambridge)
• Kristen Surrao (Columbia)
• Miles Cranmer (Cambridge)
• Antony Lewis (Sussex)
• Blake Sherwin (Cambridge)
• Julien Lesgourgues (Aachen)

To install cmbagent, follow these steps:

Clone and install our package from the cmbagent repository:

pip install cmbagent

Before pip installing cmbagent, creating a virual environment is envouraged:

python -m venv /path/to/your/envs/cmbagent_env
source /path/to/your/envs/cmbagent_env/bin/activate

You can then pip install cmbagent in this fresh environment.

If you are a cosmologist, you need the RAG data to use cmbagent in your cosmology work.

Do this:

export CMBAGENT_DATA=/where/you/want/the/data

Note that you need to set the CMBAGENT_DATA environment variable accordingly before using cmbagent in any future session. Maybe you want to add this to your .bashrc or .zshrc file, or in your activate script.

RAG agents are defined in a generic way. The core of the code is located in cmbagent.py.

To generate a RAG agent, create a .py and .yaml file and place them in the assistants directory. Additionally, create a directory named after the agent and include associated files in the data directory of cmbagent.

Apart from the RAG agents, we have assistant agents (engineer and planner) and a code agent (executor).

All agents inherit from the BaseAgent class. You can find the definition of BaseAgent in the base_agent.py file.

Before you can use cmbagent, you need to set your OpenAI API key as an environment variable:

For Unix-based systems (Linux, macOS):

export OPENAI_API_KEY="sk-..."

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

For Windows:

setx OPENAI_API_KEY "sk-..."

You can also pass your API key to cmbagent as an argument when you instantiate it:

cmbagent = CMBAgent(llm_api_key="sk-...")

Instantiate the CMBAgent with:

from cmbagent import CMBAgent
cmbagent = CMBAgent(verbose=True)

Define a task as:

task = """
       Get cosmological parameter values from Planck 2018 analysis of TT,TE,EE+lowE+lensing with the Plik likelihood in LCDM. 
       Use Cobaya with Classy_SZ to evaluate the ACT DR6 lensing likelihood for sigma8=0.8 and Omega_m=0.31. Other parameters set to Planck 2018.  
       To set Omega_m, adjust the value of omch2. 
       Give me the value of log-likelihood.
       """

Solve the task with:

cmbagent.solve(task)

If you request any output, it will be saved in the output directory.

Show the plot with:

cmbagent.show_plot("cmb_tt_power_spectrum.png")

Restore session with:

cmbagent.restore()

Push vector stores of RAG agents into the OpenAI platform:

cmbagent = CMBAgent(make_vector_stores=True)

Push selected vector stores of RAG agents into the OpenAI platform:

cmbagent = CMBAgent(make_vector_stores=['act', 'camb'])

Start session with only a subset of RAG agents:

cmbagent = CMBAgent(agent_list=['classy', 'planck'])

Show allowed transitions:

cmbagent.show_allowed_transitions()

cmbagent uses cache to speed up the process and reduce costs when asking the same questions. When developing, it can be useful to clear the cache. Do this with:

cmbagent.clear_cache()