NeuroForge is a sovereign AI fine-tuning studio. It provides a sleek, ChatGPT-style interface (hosted on Vercel or locally) that connects to a local Python Orchestrator running on your GPU machine.
- Frontend: Next.js 14+ (App Router, Tailwind CSS, Shadcn/UI).
- Backend: Python FastAPI (Runs locally on your training machine).
- Orchestrator: The backend acts as an agent, using a local LLM to interpret your commands ("Train Llama 3 on my dataset") and execute training scripts.
- Node.js 18+
- Python 3.10+
- NVIDIA GPU (Recommended for training) or Apple Silicon (Mac M1/M2/M3).
The backend handles the "brains" and the heavy lifting.
cd backend
pip install -r requirements.txtConfiguration:
Open backend/config.py and set the path to your local orchestrator model:
# backend/config.py
ORCHESTRATOR_MODEL_PATH = "/path/to/your/local/model.gguf"
# Or set to "MOCK_MODE" to test the UI without a GPU.Run Server:
python main.py
# Server will start at http://0.0.0.0:8000You can run the frontend locally or deploy to Vercel.
Local Run:
# In the root directory (where package.json is)
npm install
npm run dev
# Open http://localhost:3000Vercel Deployment:
- Push this repo to GitHub.
- Import into Vercel.
- Important: Ensure the project Root Directory is set correctly (it should be the repository root).
- Connecting to Local Backend:
- Since Vercel is on the cloud and your backend is local, you need a Tunnel.
- Use ngrok:
ngrok http 8000 - Update the frontend's API URL in
lib/api.ts(currently hardcoded tohttp://localhost:8000).
- Chat Interface: Interact with your infrastructure using natural language.
- Local Control: Your data and models never leave your machine.
- Adaptive UI: Works on Desktop and Mobile.
- Extensible: Add new training scripts in
backend/and register them in the Orchestrator.
Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.