Coder Agent

An AI-powered autonomous code generation system that takes natural language prompts and generates complete, structured projects with multiple files and proper dependencies.

Overview

Coder Agent uses a multi-agent architecture powered by LangGraph and Google's Gemini 2.5 Flash to:

1. Plan - Convert user requirements into a structured project plan
2. Architect - Break down the plan into sequential implementation tasks
3. Code - Execute each task using a ReAct agent with file system tools

Features

• 🤖 Autonomous Code Generation - Describe what you want, get a complete project
• 📁 Project Management - Organized project structure in generated_projects/
• 🔄 Multi-Agent Pipeline - Planner → Architect → Coder workflow
• 🛠️ Tool-Using Agent - Coder agent uses tools to read, write, and navigate files
• 🔗 Dependency Resolution - Automatically orders tasks to handle dependencies
• 📊 Structured Output - Uses Pydantic models for type-safe agent outputs

Architecture

User Prompt → Planner Agent → Architect Agent → Coder Agent (loops) → Complete Project
                    ↓               ↓                  ↓
                  Plan         Task Plan        File Implementation

Agents

1. Planner Agent (planner_agent)

   • Generates project name
   • Creates project plan with tech stack, features, and file structure
   • Initializes project directory

2. Architect Agent (architect_agent)

   • Converts plan into sequential implementation tasks
   • Ensures proper task ordering for dependency resolution
   • Provides detailed task descriptions with context

3. Coder Agent (coder_agent)

   • ReAct agent with file system tools
   • Implements each task sequentially
   • Maintains consistency across files
   • Loops until all tasks are completed

Installation

This project uses uv for dependency management.

Prerequisites

• Python 3.13+
• uv package manager
• Google API Key (for Gemini)

Setup

1. Clone the repository

   git clone <repository-url>
   cd coder

2. Install dependencies with uv

   uv sync

3. Set up environment variables

   cp .env.example .env

   Edit .env and add your Google API key:

   GOOGLE_API_KEY=your_api_key_here
   

Usage

Interactive Mode

Run the agent interactively:

python main.py

You'll be prompted to enter your project description:

Enter your project prompt: Build a colorful modern todo app in HTML, CSS and JS

Example Prompts

• "Build a colorful modern todo app in HTML, CSS and JS"
• "Create a Python Flask REST API with SQLite database for a blog"
• "Make a React dashboard with charts for sales data"
• "Build a CLI tool in Python for file encryption"

Project Structure

coder/
├── agent/
│   ├── graph.py          # LangGraph workflow definition
│   ├── prompts.py        # Agent system prompts
│   ├── states.py         # Pydantic state models
│   └── tools.py          # File system tools and project management
├── generated_projects/   # Output directory for generated projects
├── main.py              # CLI entry point
├── pyproject.toml       # uv project configuration
├── .env.example         # Environment variables template
└── README.md           # This file

Generated Projects

All generated projects are saved in the generated_projects/ directory with sanitized names based on your prompt and a unique hash:

generated_projects/
├── build_a_colorful_abc123/
├── create_a_python_def456/
└── make_a_react_789012/

Development

Running with Debug Output

The agent includes verbose logging by default. You'll see:

• 📁 Project initialization messages
• 📋 Task plan summaries
• 🔨 Current task being worked on
• ✅ Completion status

Modifying Agents

Each agent can be customized by editing:

• Prompts: agent/prompts.py
• State Models: agent/states.py
• Tools: agent/tools.py
• Graph Flow: agent/graph.py

Adding New Tools

Add tools to the coder agent in agent/tools.py using the @tool decorator:

@tool
def your_custom_tool(param: str) -> str:
    """Tool description for the LLM."""
    # Implementation
    return result

Configuration

LLM Model

Change the model in agent/graph.py:

llm = ChatGoogleGenerativeAI(model="gemini-2.5-flash")

Supported models:

• gemini-2.5-flash (default, fast)
• gemini-2.5-pro (more capable)

Troubleshooting

"No project initialized" Error

Make sure the planner agent runs first in the workflow.

File Write Errors

The agent can only write within the project root for security. All paths are validated.

API Rate Limits

If you hit rate limits, consider:

• Using a different model
• Adding delays between requests
• Increasing recursion limit to allow more retries

Dependencies

Key dependencies (managed by uv):

• langgraph - Agent workflow orchestration
• langchain-google-genai - Google Gemini integration
• pydantic - Data validation and settings
• python-dotenv - Environment variable management

See pyproject.toml for the complete list.

Acknowledgments

Built with:

• LangGraph by LangChain
• Google Gemini for LLM capabilities
• uv for Python package management