AI agents to help working with CAMELS (Cosmology and Astrophysics with MachinE Learning Simulations) data. CAMELS Agent can help with:

• Writing your CAMELS section if you are using CAMELS data for a paper
• Ask questions about the CAMELS documentation
• Search for CAMELS papers. E.g. What CAMELS papers have studied the impact of AGN feedback on the Lya forest
• Search general papers on the arXiv.
• Coding. E.g. write code to analyze any aspect of CAMELS snapshots, galaxy catalogs...etc.
• A standard LLM for outside CAMELS questions.

The easiest way to use CAMELS Agents is via the associated website. You will only need the LLM API keys to use the agents.

Installation

Requirements:

• python 3.10 or above. Needed to properly use the UI.
• An API for the LLM model. CAMELS Agents supports:
  • ChatGPT-4o
  • ChatGPT-3o-mini
  • Gemini-2-flash
  • Gemma2-9b
  • Sonnet-3.7
  • Llama3-70b
  • DeepSeek-R1-distill-llama-70b
  • DeepSeek-R1-distill-qwen-32b
• A Vertex AI setup file for the embeddings.
• A langchain API key (optional).
• A semantic scholar API key (optional).

Instructions:

• git clone https://github.com/franciscovillaescusa/CAMELS_Agents.git
• cd CAMELS_Agents
• python -m venv CA_env
• source CA_env/bin/activate
• pip install .

At this point, the library should be installed together with all its dependencies. Next we need to create a .env with this content:

# Vertex AI (used for embeddings)
GOOGLE_APPLICATION_CREDENTIALS=gemini.json

# Gemini parameters (needed if using Gemini)
GOOGLE_API_KEY=your_google_api_key

# OpenAI parameters (needed if using ChatGPT)
#OPENAI_API_KEY=your_openai_api_key

# Anthropic parameters (needed if using Sonnet-3.7)
#ANTHROPIC_API_KEY=your_anthropic_api_key

# Groq parameters (needed if using Llama3, Gemma2, or DeepSeek-R1)
#GROQ_API_KEY=your_groq_api_key

# LangChain parameters (optional)
#LANGCHAIN_TRACING_V2=true
#LANGCHAIN_API_KEY=your_langchain_api_key
#LANGCHAIN_ENDPOINT=https://api.smith.langchain.com
#LANGCHAIN_PROJECT=Tools

# Semantic scholar key (optional)
#SEMANTIC_SCHOLAR_KEY=your_semantic_schoolar_api_key

If, for example, you are going to use ChatGPT, comment the gemini parameters, comment out the OPENAI_API_KEY and put your own key.

How to get the keys

LLMs:

• Gemini. Get the API key from the Gemini API key.
• OpenAI. Get the API key from the openai website.
• GROQ. Get the API key from the Groq website.
• Anthropic. Get the API key from Anthropic website

Vertex AI. CAMELS agents use an embedding model (text-embedding-005) for RAG tasks. To enable the google embeddings, follow these instructions:

• login into google cloud: https://console.cloud.google.com/
• create a project
• go to IAM & Admin
• go to Service Accounts
• create Service Account
• choose a name
• Create and Continue
• Grant the "Vertex AI User" role
• click Continue and then Done
• click in the three dots and Manage keys
• Add key --> Create new key --> JSON --> Create
• Download the JSON file and place it inside the CAMELS_Agent subfolder
• change the name of that file to gemini.json

Langchain (optional). Go to the langsmith website and login. Click on settings (bottom left) and create an API key.

Semantic scholar (optional). Go to the semantic schoolar website and sign in/create an account. At the bottom of the page, go to API overview. At the end of the page there is a form to request an API key. Note that unless you made use of semantic schoolar heavily, you dont need an API key.

Run

Once the code is installed and all the API keys in place type:

streamlit run app.py

Tuning prompts.

CAMELS Agents use several text files to context and/or guide the LLM behaviour. You may want to tune these for your particular needs.

CAMELS section

The original text describing CAMELS is located in Input_Text/CAMELS_section.txt. Modify that section if you want a different initial description.

CAMELS papers

The agents are able to identify CAMELS papers for a given query; for instance to see if an idea has been already carried out. The list of CAMELS papers is located in Input_Text/CAMELS_papers_links.py. Add any paper you want to the list for the agents to work with them. One the list is updated, go to the src folder and run python update_db_CAMELS_papers.py to add those papers to the Chroma database.

CAMELS coding

The guidelines to use CAMELS data are in Input_Text/CAMELS_coding.txt. Add or modify that file for your needs. Take into account that the longer the file, the larger the context and the more expensive the call to the LLM will be.

Team

• Francisco Villaescusa-Navarro (Flatiron)
• Boris Bolliet (Cambridge)
• Pablo Villanueva-Domingo (Barcelona)
• ChangHoon Hahn (Arizona)