SciAgents

SciAgents is an extensible multi-agent framework designed for scientific research scenarios. It leverages large language models (LLMs) to automate research tasks and can be integrated with robotic systems for advanced scientific workflows.

Features

• Supports various LLM providers (OpenAI, Azure, Gemini, etc.)
• Modular agent and tool system for easy extension
• Flexible configuration for different research needs
• Ready-to-use test scripts and Jupyter notebooks
• Can be combined with robots for automated scientific experiments

Installation

pip install sciagents

Quick Start

Below is a minimal example of using ChatAgent in your project.
Note: Make sure you have a valid config/config.yml with your LLM API keys and model info.

import os
from sciagents.agents.chat_agent import ChatAgent
from sciagents.agents.message import AgentInput, Message, Role
import yaml

# Load config
config_path = os.path.join("config", "config.yml")
with open(config_path, "r", encoding="utf-8") as f:
    config = yaml.safe_load(f)
chat_agent_config = config["agents"]["ChatAgent"]

# Build agent input
messages = [Message(role=Role.USER, content="Introduce yourself, please.")]
agent_input = AgentInput(messages=messages)

# Create ChatAgent instance
agent = ChatAgent(
    name="DemoChatAgent",
    llm_config={
        "model": chat_agent_config["model"],
        "api_key": chat_agent_config["api_key"],
        "api_base": chat_agent_config["url"],
        **chat_agent_config.get("model_config_dict", {})
    },
    stream=True
)

# Get response
output = agent.step(agent_input)
if hasattr(output.content, "__iter__") and not isinstance(output.content, str):
    for chunk in output.content:
        print(chunk, end="", flush=True)
    print()
else:
    print(output.content)

Project Structure

• sciagents/: Core code (agents, tools, LLM interfaces)
• config/: Configuration files
• test/: Test scripts and examples

License

This project is licensed under the MIT License.

Acknowledgement

This repository is inspired by CAMEL: Communicative Agents for "Mind" Exploration of Large Scale Language Model Society. We thank the CAMEL team for their open-source contributions.