Agent 001

A simple tool-using AI agent built from first principles in Python.

What it can do

• Accept a user goal
• Decide when to use tools
• Read files from a restricted workspace
• Perform calculations
• Write results to files
• Chain multiple tool calls autonomously
• Recover from tool errors
• Stop after a maximum number of steps

What I learned

This project was built without an agent framework to understand the core architecture:

Goal → Model Decision → Tool Call → Tool Execution → Observation → Next Decision → Completion

It also includes basic safety controls:

• Workspace-restricted file access
• Iteration limits
• Error handling

Setup

Create a .env file containing:

OPENAI_API_KEY=your_key_here

Install dependencies:

pip install openai python-dotenv

Run:

python agent.py