Python AI Agent

This is a command-line AI agent powered by Google Gemini that analyzes user requests, calls local tools (functions), and iteratively works toward a final response.

Features

• Tool Calling: The AI can choose and execute local Python functions
• Iterative Agent Loop: The model reasons step-by-step until the task is complete
• Conversation Memory: Each iteration remembers prior messages and tool results
• Verbose Mode: Optional debugging output showing token usage and tool execution
• Safe Execution: Iteration limits prevent infinite loops and excessive token usage

Requirements

• Python 3.x
• Google Gemini API key
• Python packages:
  • google-genai
  • python-dotenv

Installation

No installation required! Just clone or download this repository:

git clone <your-repo-url>
cd python-ai-agent

Install dependencies:

pip install google-genai python-dotenv

Create a .env file:

GEMINI_API_KEY=your_api_key_here

Usage

Run the AI assistant by providing a prompt (uv):

uv run main.py "<your prompt>"

Enable verbose debugging output:

uv run main.py "<your prompt>" --verbose

Example

uv run main.py "how does the calculator render results to the console?"

Output

The AI Code Assistant will:

• Send your prompt to Gemini
• Decide whether tools are needed
• Execute local functions when requested
• Feed results back to the model
• Produce a final response

Example verbose output:

User prompt: what files are in the root?

Calling function: get_files_info({})
-> {'files': ['main.py', 'functions', 'README.md']}

Response:
The project contains main.py, a functions directory, and README.md.

If the agent exceeds its iteration limit:

Error: reached maximum iterations (20) without producing a final response.

Available Tools

• get_files_info — Lists files in a directory
• get_file_content — Reads file contents
• run_python_file — Executes a Python script
• write_file — Writes content to a file

Each tool includes:

• A Gemini function schema (for the model)
• A Python implementation (for execution)

Project Structure

main.py              — Entry point and agent loop
call_function.py     — Tool dispatcher and execution logic
prompts.py           — System prompt instructions
functions/
├── get_files_info.py
├── get_file_content.py
├── run_python_file.py
└── write_file.py

How It Works

1. User prompt is sent to Gemini.
2. The model may request a function call.
3. The tool is executed locally.
4. Tool results are added back into the conversation.
5. The model continues reasoning until a final answer is produced.
6. The agent loop runs for a maximum of 20 iterations to ensure safe execution.

License

Open source — feel free to use and modify as you like.