AI Agent System

An extensible agent system that combines the power of language models with shell command execution and file manipulation capabilities.

Overview

This project implements a modular AI agent system that can:

• Process natural language commands
• Execute shell commands
• Answer questions using language models
• Manipulate files

The system supports multiple modes of operation:

1. Direct Mode: Loads model weights from disk
2. API Mode: Connects to a custom API server
3. Ollama Mode: Uses a local Ollama server with any available model

Project Structure

WorkSpace/Agent/                     # Root directory for the project
├── deploy_api_server_scripts/       # Directory for scripts to launch the API server
│   └── deploy_api_server_qwen25_72b.py  # Script to start the API server
├── local_model_weights/             # Directory for model weights (for local deployment)
├── log/                             # Directory for logs
├── requirements.txt                 # Python dependencies
├── src/                             # Source code
│   ├── agent.py                     # Agent class for task handling
│   ├── llm.py                       # LLM integration
│   └── logger.py                    # Logging setup
├── start.py                         # Main entry-point script
└── static/                          # Static assets (if applicable)

Installation

1. Clone the repository:

git clone https://github.com/romgenie/ai-agent-system.git
cd ai-agent-system

2. Install dependencies:

pip install -r requirements.txt

Usage

Starting the Agent

Direct Mode (with local model weights)

python start.py --model-path ./local_model_weights

API Mode (connecting to a running API server)

python start.py --api-url http://localhost:8760

Ollama Mode (using models from Ollama)

python start.py --ollama --model-name llama3

Command Execution

You can run the agent in two modes:

Interactive Mode

python start.py --ollama --model-name deepseek-coder --interactive

Single Command Mode

python start.py --ollama --model-name deepseek-coder --command "What is the current date?"

Starting the API Server

If you want to run your own API server:

python deploy_api_server_scripts/deploy_api_server_qwen25_72b.py

Extending the System

The modular architecture makes it easy to extend the system with new capabilities:

1. Add new LLM providers in llm.py
2. Implement additional command types in agent.py
3. Create custom endpoints in the API server

Research Foundation

This implementation is based on research from the paper:

Frontier AI systems have surpassed the self-replicating red line

• Authors: Junxiao Song, Xuming Hu, Wenbo Guo, Zheng Li, Fan Yang, Dongkuan Xu, Yongfeng Zhang, Heng Ji, Jiliang Tang and Xia Hu
• arXiv:2412.12140v1

The paper introduces frameworks for developing and benchmarking AI agent systems that can collaborate and perform complex tasks through natural language interaction.

License

MIT

Credits

This project uses various open-source components and LLM providers:

• Ollama for local model inference
• FastAPI for the API server
• Various LLM models like deepseek-coder, Qwen, etc.