AI Agent

A simple AI agent to explore how this stuff works.

Uses Google's Gemini API through the google-genai package`.

This is a guided project part of Boot.dev's curriculum.

Code committed after this tag is my own personal additions beyond the scope of Boot.dev's curriculum.

Setup

1. Create a virtual environment and install dependencies

uv venv
source .venv/bin/activate
uv sync

2. Create a .env file in the main directory (alongside .gitignore) and insert your Gemini API key like this:

GEMINI_API_KEY=my_api_key_123456789

Other setup option are available in config.py.

Usage

uv run main.py <prompt> [--verbose]

Try with these prompts:

• "show me what's in the root directory"
• "how does the calculator render results to the console?"
• "please fix the bug in the calculator" (after creating a simple bug like changing operator precedence)

Stats

Print API usage stats. Configured by default for Gemini API free plan for the 2.0 Flash model.

uv run main.py stats

Sample output

Usage stats:
Tokens 24h:       5.2%    10345 / 200000
Requests 24h:     5.5%    11 / 200
Requests 60s:     0.0%    0 / 15

Run tests

pytest

Sample project: Calculator

The agent needs some project to work on, so Boot.dev provides a "calculator" package.

uv run calculator/tests.py
# Ran 9 tests in 0.001s

$ uv run calculator/main.py "3 + 5 * 2"
# 3 + 5 * 2 = 13

Extra credits

Some ideas for some features outside the scope of the guided project that I'd like to add:

• Use logging instead of printing
• Use pytest, plain assertions are more readable than unittest methods
• API usage DB so you can easily see know many token/request you used (and can still use within free plan limits)
  • stats subcommand
  • Test
  • Conversation ID
  • Save request prompt/result
• argparse for argument parsing
• Simplify calling main (uv run main "prompt"is too many words, maybe with pyproject.toml there's a way to install a shell script)
• Move code inside an src directory
• Fix or remove all TODOs
• Ability to install this library in any Python project I make so I can use the agent on any codebase