Agent S 🐝 👁️

An experimental AI agent leveraging state-of-the-art LLM models with a focus on simplicity and Unix-based tool interactions.

Overview

Agent S demonstrates that modern LLM models can achieve sophisticated agentic capabilities through:

• Carefully crafted prompting
• Direct Unix tool integration
• Console-based I/O instead of native function calling
• Continuous learning loops

Philosophy

The project is built on the hypothesis that reasoning-focused LLM models, when properly tooled, can achieve high-level agentic performance. This approach may provide insights into paths toward AGI as these models continue to evolve.

Installation

This project uses uv as its package manager.

uv run main.py

Environment Setup

Create a .env file in the root directory with the following API keys:

OPENAI_API_KEY=your_openai_key
DEEPSEEK_API_KEY=your_deepseek_key
SERPER_API_KEY=your_serper_key

License

MIT