Agentic code editor using Python and Google Gemini — supports function-calling, file editing, and debugging via LLM.

Agentic Code Editor with Google Gemini

A toy code editor agent, inspired by Claude Code and Cursor’s agent mode. This project walks through how to build an LLM-powered code assistant using the free Google Gemini API, Python function calling, and feedback loops.

🔧 The final result is a local tool that can analyze Python files, detect bugs, and modify code — all through intelligent agentic reasoning.

A toy code editor agent, inspired by Claude Code and Cursor’s agent mode. This project demonstrates how to build an LLM-powered code assistant using the free Google Gemini API, Python function calling, and feedback loops.

The agent can:

Analyze code
Detect bugs
Modify files
Run Python code All through natural language prompts and Gemini’s tool-calling capabilities.

🚀 Getting Started

📦 Requirements

Python 3.8+
A free Google AI Studio API key
google-generativeai client

🔧 Installation

git clone https://github.com/yourusername/AI-Agent.git
cd AI-Agent.git
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Set your API key:

export GOOGLE_API_KEY="your-key-here"

▶️ Usage

Ask questions about the codebase like:

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

Expected output:

 - Calling function: get_files_info
 - Calling function: get_file_content
Final response:
<agent explains logic here...>

🧪 Example Project Included

This repo includes an example project in the calculator/ folder that the agent can analyze and debug.

You can try it out immediately with:

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

This will trigger the agent to:

Explore files in the calculator/ project Read and interpret code logic Use tool calls to retrieve and analyze the relevant functions

🧪 Features

✅ Agentic loop with up to 20 iterations

✅ Gemini 1.5 / 2.0 support

✅ Structured function-calling with validation

✅ Real-time feedback from executed code

📂 Project Structure

.
├── main.py                  # Entry point for the agent
├── tools/                   # Functions: read/write files, run code, list dirs
├── agent/                   # Core agent logic and feedback loop
├── config.py                # System prompts and setup
├── calculator/              # 🔹 Example Python project for testing the agent
│   ├── main.py
│   ├── pkg/
│   │   ├── calculator.py
│   │   ├── render.py
│   │   └── morelorem.txt
│   ├── tests.py
│   └── README.md
└── README.md

🛠️ Built With

Google Gemini API
Python 3
Function calling (types.FunctionDeclaration, Part.function_call)
Feedback loop design pattern

AI-Agent