AI Agent: CLI Code Editor

A minimal, terminal-based AI coding agent powered by Google's Gemini API — inspired by Claude Code and Cursor's Agent Mode. I built this project to explore agentic workflows, function calling, and LLM-powered code editing directly from the command line.

Overview

This project is a CLI terminal tool that uses function calling and feedback loops to autonomously find and fix bugs in a codebase. It interacts with the gemini-2.5-flash model to analyze code, execute custom functions, and iteratively repair issues in target projects (such as the included calculator package).

Core Components

The architecture of the agent is built on four main capabilities:

1. LLM Integration

The agent connects to the Gemini API using the Python client library. It handles prompts and maintains conversational context directly within the CLI environment.

2. File and Execution Functions

The agent is equipped with custom tools to interact with the file system and run Python code. For example, functions/get_files_info.py provides the agent with secure directory traversal and file listing capabilities.

3. Function Calling

Through its system prompt and context window, the AI is granted the ability to dynamically select and execute the appropriate tools based on the user's terminal commands.

4. Agentic Feedback Loop

A proper feedback loop makes the tool truly "agentic." It can analyze the results of its own function calls, re-evaluate the codebase's state, and iteratively work towards resolving bugs without manual intervention.

Project Structure

• pyproject.toml / uv.lock: Project dependencies (managed via uv), requiring Python 3.12+ and packages like google-genai and python-dotenv.
• main.py: The CLI entry point for the agent. Use it to send prompts and initiate debugging sessions.
• functions/: Contains the tools exposed to the agent.
  • get_files_info.py: Allows the agent to list directory contents safely within a restricted working directory.
• calculator/: A sample target codebase used for testing the agent's ability to debug and repair broken logic.
• .env: Environment variables (add your GEMINI_API_KEY here).

Setup and Usage

1. Install dependencies: Ensure you have Python 3.12 or above. Install the requirements from pyproject.toml:

   uv sync
   # or pip install -e .

2. Environment Variables: Create a .env file in the root directory and add your API key:

   GEMINI_API_KEY=your_gemini_api_key_here

3. Run the CLI:

   python main.py "Find and fix the bug in the calculator package"

   For verbose output with token usage:

   python main.py "Hello, Gemini!" --verbose