TunnelMind

A distributed platform that enables users to share GPU resources and AI models. "Givers" can share their local Ollama models, and "Takers" can access them through secure tunneling.

Quick Links

• Live App: https://tunnelmind.harshkeshri.com
• NPM Package: https://www.npmjs.com/package/tunnelmind

Installation

TunnelMind ships with a CLI that lets you authenticate, start the local relay, and publish your Ollama models to the cloud.

npm install -g tunnelmind

Prerequisites:

• Node.js 18+
• Ollama installed and running locally (ollama serve)
• At least one model pulled via ollama pull <model>

Usage

Authenticate the CLI:

tunnelmind login

Start sharing your local models:

tunnelmind server

Optional flags:

• --cloud-url to point at a different TunnelMind cloud server
• --ollama-url to target a custom Ollama endpoint

Check the current session:

tunnelmind user

Log out and clear the local session:

tunnelmind logout

Project Structure

• cloud-server/ – Express + MongoDB backend managing givers, takers, chats, and the WebSocket relay.
• local-server/ – CLI client that connects a giver’s Ollama instance to the cloud server.
• frontend/ – Next.js frontend for takers and givers to manage sessions, browse models, and chat.

Development

Install dependencies for all packages:

pnpm install:all

Start individual services:

# Cloud server
eval "cd cloud-server && pnpm dev"

# Local CLI (debug mode)
eval "cd local-server && pnpm dev:cli"

# Frontend
eval "cd frontend && pnpm dev"

Contributing

Pull requests and feature suggestions are welcome. Please open an issue describing your proposal before submitting large changes.

License

MIT © TunnelMind