🤖 AI CLI Agent: Autonomous Agentic Code Editor

Description

A lightweight, terminal-native AI agent engineered to interface seamlessly with Google's Gemini models. Built in Python, this tool functions as a toy agentic code editor capable of reading, updating, and executing code autonomously. By utilizing a modular function-calling architecture and a dynamic ReAct (Reasoning and Acting) feedback loop, the agent can navigate local file systems, evaluate its own execution outputs (stdout/stderr), and iterate until a task is successfully resolved.

Motivation

Standard chat interfaces often lack the context and permissions necessary for real software development tasks. The motivation behind this project is to build an LLM-powered command-line program from scratch that bypasses these limitations. By providing the AI with direct filesystem access, dynamic script execution, and a self-correcting feedback loop, this agent serves as a transparent, high-speed alternative for local codebase manipulation and bug fixing. It prioritizes speed, deterministic dependency management, and secure state handling.

Quick Start

Prerequisites

• Python 3.14+
• uv package manager (pip install uv)
• Google Gemini API Key

Installation

1. Clone the repository:

git clone https://github.com/Shreyansh15624/ai-bot
cd ai-bot

2. Initialize the environment: This project utilizes uv (pyproject.toml / uv.lock) for lightning-fast, fully deterministic dependency resolution.

uv sync

3. Configure Environment Variables: Add your API key to your environment or a .env file:

export GOOGLE_GENAI_API_KEY="your_api_key_here"

Usage

Execute natural language commands directly from your terminal. The agent will autonomously decide which tools from the functions/ directory to use to accomplish your prompt.

Target specific workspaces by pointing the agent to the projects/ directory, which safely isolates target data from the core orchestrator.

Analyze an Isolated Project:

uv run main.py "Scan the projects/ directory and find the appropriate calculator project directory, and read the python files, then summarize how the compound interest is calculated."

Autonomous Bug Fixing & Execution Loop:

uv run main.py "Review projects/calculator/main.py, find any logical errors, fix them, and run the script to verify the output is correct."

Note: The agent will capture the terminal output. If an error occurs during execution, it will read the traceback, self-correct the code, and retry until successful.

Contributing

The system cleanly separates the core orchestrator (main.py) from the modular toolsets (functions/) and test environments (projects/).

• To add a new tool (e.g., web scraping, Git commands): Define a new Python script in the functions/ directory. The orchestrator will automatically parse the function signature and expose it to the LLM's context window.
• To add a new test workspace: Create a new folder inside the projects/ directory to act as an isolated sandbox for the agent to manipulate.