oneShotCodeGen - AI Retool/ERP App Generation

A modular agent-based system for generating and editing app configurations through AI-driven steps organized in flows.

Overview

This project provides an advanced AI-powered platform for automatically generating and editing enterprise application configurations through a series of intelligent agent flows. Built with a modular architecture, it leverages OpenAI's APIs and structured generation to produce consistent, high-quality app specifications.

Repository: https://github.com/vivek100/oneShotCodeGen

Project Structure

• backend/: Modular FastAPI backend with agent flow system
• frontend/: React-based frontend interface with real-time updates

Quick Start

Backend Setup

1. Clone the repository:

   git clone https://github.com/vivek100/oneShotCodeGen.git
   cd oneShotCodeGen
   

2. Set up Python environment:

   cd backend
   python -m venv venv
   # On Windows
   venv\Scripts\activate
   # On macOS/Linux
   source venv/bin/activate
   

3. Install dependencies:

   pip install -r requirements.txt
   

4. Create environment file:

   cp sample.env .env
   

5. Configure your environment variables in .env:

   # API Configuration
   HOST=0.0.0.0
   PORT=8000
   
   # Database Configuration
   DB_URL=sqlite:///./agent.db
   
   # OpenAI API Configuration
   OPENAI_API_KEY=your_openai_api_key_here
   OPENAI_MODEL_NAME=gpt-4.1-nano
   
   # Frontend Configuration (for CORS)
   FRONTEND_ORIGIN=http://localhost:3000
   

6. Start the backend server:

   # Development mode with auto-reload
   python main.py
   # Or using uvicorn directly
   uvicorn main:app --host 0.0.0.0 --port 8000 --reload
   

The backend API will be available at http://localhost:8000.

Frontend Setup

1. Navigate to the frontend directory:

   cd frontend
   

2. Install Node.js dependencies:

   npm install
   

3. Create environment file:

   cp .env.example .env.local
   

4. Configure your environment variables in .env.local:

   NEXT_PUBLIC_BACKEND_URL=http://localhost:8000
   

5. Start the development server:

   npm run dev
   

The frontend will be available at http://localhost:3000.

Backend Architecture

The backend is built with a modular architecture focused on managing agent flows and step execution:

Core Modules

1. Agent Router: Routes user messages to appropriate flows based on the main agent decision.
2. Flow Registry: Manages flow definitions and their steps.
3. Flow Runner: Orchestrates flow execution by running steps in sequence.
4. Step Executor: Executes individual steps (AI or tool calls).
5. Tool Call Module: Executes non-AI operations like entity reorganization, merging, etc.
6. Prompt & Schema Store: Manages prompt templates, schemas, and examples.
7. Message Dispatcher: Handles system messages during flow execution.
8. Replay Engine: Re-runs steps with original or modified inputs for testing and debugging.
9. WebSocket Manager: Provides real-time communication for chat messages and flow status updates.

Flow Types

• Main Agent Flow: Decides what to do with user input
• Create App Flow: Generates a complete app configuration
• Edit Decision Flow: Decides how to handle edit requests
• Partial Edit Flow: Updates specific parts of an app configuration

Database Structure

The system uses SQLite with the following key tables:

• Projects: Stores project metadata
• Messages: Stores chat messages for each project
• AppVersions: Stores app configuration versions
• Flows: Stores flow definitions
• Steps: Stores step definitions within flows
• FlowRuns: Tracks executions of flows
• StepRuns: Tracks executions of individual steps
• Prompts: Stores prompt templates
• Schemas: Stores JSON schemas for validation
• OneShots: Stores example inputs/outputs for steps
• PydanticSchemas: Stores Pydantic model references

Frontend Architecture

The frontend is built with React and Next.js, providing a modern and responsive user interface:

Key Features

1. Project Management: Create, view, and manage application configuration projects
2. Real-time Chat: Interact with the AI agent through a WebSocket-powered chat interface
3. Configuration Preview: View and explore generated app configurations
4. Flow Visualization: Explore the agent flows and steps
5. Step Replay: Debug and experiment with individual AI steps
6. Application Preview: Preview the generated application (work in progress)

Technology Stack

• React & Next.js: Core frontend framework
• Tailwind CSS: Utility-first CSS framework
• ShadcnUI: Component library based on Radix UI
• React Query: Data fetching and caching
• Zustand: State management
• WebSockets: Real-time communication

How It Works

The AI ERP App Configuration Generator works through a series of interconnected agent flows:

1. User Interaction: Users interact with the system through a chat interface, describing the application they want to build or the changes they want to make.

2. Main Agent Decision: The main agent analyzes the user's message and decides which specialized flow to execute (create new app, edit existing app, etc.).

3. Flow Execution: The selected flow runs through a series of steps, combining AI reasoning with specialized tool calls.

4. AI Steps: These steps use large language models (via OpenAI's API) to:

   • Generate use cases from requirements
   • Design entity models and relationships
   • Create interface specifications
   • Plan data model implementations

5. Tool Call Steps: Non-AI steps perform operations like:

   • Reorganizing entities to handle foreign key relationships
   • Merging changes into existing configurations
   • Validating configuration integrity
   • Generating final output structures

6. Structured Generation: The system uses both JSON schema validation and Pydantic models to ensure consistent, well-structured outputs from the AI.

7. Real-time Updates: WebSockets provide real-time chat messages and status updates to the frontend.

Pydantic Structured Generation

The system leverages Pydantic models for structured LLM generation, providing robust type validation:

from typing import List, Optional, Literal
from pydantic import BaseModel, ConfigDict, Field

class Entity(BaseModel):
    model_config = ConfigDict(extra="forbid", exclude_none=True)
    
    name: str = Field(..., description="Entity name")
    description: str = Field(..., description="Entity description")
    fields: List["Field"] = Field(..., description="Entity fields")

class Field(BaseModel):
    model_config = ConfigDict(extra="forbid", exclude_none=True)
    
    name: str = Field(..., description="Field name")
    type: str = Field(..., description="Field type")
    description: str = Field(..., description="Field description")
    required: bool = Field(..., description="Whether field is required")

Template Rendering with Jinja2

The system uses Jinja2 for flexible prompt templates:

You are designing an entity model for a business application.

# Requirements
{{ requirements | json }}

# Guidelines
- Create entities based on the business domain
- Consider relationships between entities
- Follow naming conventions

# Expected Output
{{ schema | json }}

Deployment Options

Local Development

Run both the frontend and backend in development mode as described in the Quick Start section.

Production Deployment

1. Backend:

   • Build a Docker container for the FastAPI backend
   • Deploy to a cloud service like AWS, GCP, or Azure
   • Set up appropriate environment variables

2. Frontend:

   • Build the Next.js application: npm run build
   • Deploy to Vercel, Netlify, or another hosting service
   • Configure environment variables to point to your deployed backend

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add some amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• This project uses OpenAI's API for large language model capabilities
• Inspired by the potential of AI-assisted software engineering
• Built with a focus on modular, maintainable architecture