MidiCopilot

AI-powered music workspace with visual workflow editor for MIDI generation and manipulation.

https://github.com/user-attachments/assets/5e86de6c-96cb-4d13-a207-acdc044494fd

Architecture

MidiCopilot consists of two main components:

• Frontend: React + TypeScript + Vite with visual workflow editor
• Backend: FastAPI + PyTorch for ARIA model inference

Quick Start

Prerequisites

• Python 3.12+
• Node.js 18+
• UV (Python package manager)
• pnpm (Node package manager)

Installation

Install UV:

curl -LsSf https://astral.sh/uv/install.sh | sh

Install pnpm:

curl -fsSL https://get.pnpm.io/install.sh | sh -

Development Setup

1. Clone the repository

   git clone <your-repo-url>
   cd MidiCopilot
   

2. Install dependencies

   Backend:

   cd backend
   uv sync
   cd ..
   

   Frontend:

   cd frontend
   pnpm install
   cd ..
   

3. Run the development servers

   From the root directory:

   pnpm run dev
   

   This will start:

   • Backend API at http://localhost:8000
   • Frontend UI at http://localhost:5173

Manual Start (Optional)

If you prefer to start servers separately:

Backend:

cd backend
uvicorn src.main:app --reload --host 0.0.0.0 --port 8000

Frontend:

cd frontend
pnpm dev

Project Structure

MidiCopilot/
├── backend/                 # FastAPI backend
│   ├── src/
│   │   ├── dto/            # Pydantic models
│   │   ├── services/       # Business logic (ARIA service)
│   │   └── main.py         # FastAPI app
│   ├── data/
│   │   ├── uploads/        # Uploaded MIDI files
│   │   └── outputs/        # Generated MIDI files
│   └── pyproject.toml      # Python dependencies
│
├── frontend/               # React frontend
│   ├── src/
│   │   ├── components/     # React components
│   │   │   ├── AriaExecutionPanel.tsx
│   │   │   ├── MidiFileUpload.tsx
│   │   │   └── WorkflowEditor.tsx
│   │   ├── pages/          # Page components
│   │   ├── mockData.ts     # Workflow definitions
│   │   ├── types.ts        # TypeScript types
│   │   └── theme.ts        # UI theme
│   └── package.json        # Node dependencies
│
└── README.md

API Documentation

Once the backend is running, you can access:

• Swagger UI: http://localhost:8000/docs