Bog AI Agent

A command-line AI coding agent which can we run via Google's API key. Bog can autonomously inspect, read, edit, and execute code in a project through a tool-calling loop.

Features

• File and directory listing
• File reading and writing
• Python file execution
• Iterative tool-use loop with the model (up to 20 iterations per prompt)
• Verbose mode for token usage and step tracing
• Path sandboxing — all operations are confined to a working directory

Tech Stack

• Python 3
• Google Gemini API (google-genai)
• uv for dependency management

Setup

1. Clone the repository.

2. Create a .env file with your Gemini API key:

   GEMINI_API_KEY=your_key_here
   

3. Install dependencies:

   uv sync
   

4. Change Gemini model from config.py

Usage

uv run main.py "your prompt here"
uv run main.py "your prompt here" --verbose

Example Prompts

• uv run main.py "list files in the calculator directory"
• uv run main.py "read main.py in calculator" --verbose
• uv run main.py "run tests.py"
• uv run main.py "write hello world to notes.txt"

Architecture

• main.py — entry point, agent loop, and message history management
• call_function.py — dispatches model tool calls to the correct function
• functions/ — tool implementations (get_files_info, get_file_content, run_python_file, write_file) and their Gemini function schemas

Try It: End-to-End Agent Workflow

The test_agent/ folder contains a calculator subfolder with a small sample program you can use to see the full agent loop in action.

1. Open one of the files in test_agent/calculator/ and break the code.

2. Run the agent and let it diagnose and fix the issue:

   uv run main.py "fix the calculator for me"
   

3. Use --verbose to watch each tool call and token usage:

   uv run main.py "fix the calculator for me" --verbose
   

The agent will list the directory, read files, run the tests, edit the broken code, and re-run the tests to confirm the fix — all without any further input.