link-ai

Run an LLM on one machine. Access it from anywhere.

An open-source alternative to LM Studio's LM Link. Run your local LLM on a server and access it from any machine via CLI or web GUI.

What is this?

You have a powerful machine running an LLM (via Ollama, vLLM, llama.cpp, etc.). You want to use it from your laptop, your phone, or another server — without exposing ports to the internet.

link-ai is a lightweight proxy + client that makes this easy.

┌──────────────────────────┐             ┌──────────────────────┐
│ Server (anywhere)        │             │ Your machine         │
│                          │             │                      │
│  Ollama / vLLM / etc    │             │  link-ai chat        │
│  link-ai host :8080      │ ◄─────────► │  link-ai gui         │
│                          │  internet   │  link-ai dashboard   │
└──────────────────────────┘             └──────────────────────┘

Works across countries. Tested from Germany to Poland.

Quick Start

Install

npm install -g @sski/link-ai

On the machine running the LLM

# If using Ollama
link-ai host --upstream http://localhost:11434 --port 8080 --all-interfaces

# If using vLLM
link-ai host --upstream http://localhost:8000 --port 8080 --all-interfaces

# With Tailscale (recommended for remote access)
link-ai host --upstream http://localhost:11434 --port 8080 --all-interfaces --tailscale

On your local machine

# Chat via CLI
link-ai chat 192.168.1.100:8080 -m "hello"
link-ai chat 192.168.1.100:8080 -i  # interactive mode

# Open web GUI
link-ai gui 192.168.1.100:8080

# Open dashboard (manage multiple hosts)
link-ai dashboard

# Test connection
link-ai connect 192.168.1.100:8080

# List models
link-ai models 192.168.1.100:8080

Why link-ai?

You can already access remote LLMs via SSH or RDP. So why another tool?

| | SSH + Ollama | RDP + LM Studio | LM Studio Link | link-ai | |---|:---:|:---:|:---:|:---:| | Setup time | 5 min | 30 min | Waitlist | npm i -g | | Chat UI | Terminal only | Full desktop GUI | GUI | ChatGPT-style web GUI | | Phone/tablet | Painful | Laggy | Yes | Yes, just a browser | | Multiple hosts | Manual | One at a time | Limited | Dashboard with status | | Bandwidth | Low | High (streaming desktop) | Medium | Low (just API calls) | | Headless server | Yes | Needs desktop env | Yes | Yes | | Open source | Yes | N/A | No | Yes | | Auto model detect | No | No | Yes | Yes | | Capability detection | No | No | No | Yes | | Web search | No | No | No | Yes | | File upload | No | No | No | Yes | | Works across countries | Yes (if port open) | Yes (if port open) | Yes (Tailscale) | Yes (Tailscale) | | Security | Key-based | Password | Tailscale | Tailscale + API key |

The bottom line:

• SSH = for sysadmins who live in terminal
• RDP = for when you need the full desktop (overkill for LLM chat)
• LM Studio Link = polished but closed source, waitlist-gated
• link-ai = open source, instant setup, works on any device, purpose-built for LLM access

link-ai isn't trying to replace SSH or RDP. It's the right tool for one specific job — chatting with a remote LLM from any device, with a nice UI, without the overhead.

Features

• OpenAI-compatible proxy — works with Ollama, vLLM, llama.cpp, LocalAI, etc.
• Streaming responses — real-time token-by-token output via SSE
• CLI chat — one-shot messages or interactive conversation with history
• Web GUI — ChatGPT-style browser interface with model selector
• Dashboard — manage multiple hosts, see status, latency, models
• Auto model detection — no need to specify model names
• Optional auth — --auth <key> to protect your host
• Tailscale support — show Tailscale IPs, list peers
• Works across countries — tested from Germany to Poland (105ms latency)

Commands

| Command | Description | |---------|-------------| | link-ai host | Start proxy server for your LLM | | link-ai chat <address> | Chat with a remote LLM (CLI) | | link-ai gui <address> | Open chat GUI in browser | | link-ai dashboard | Host management dashboard | | link-ai connect <address> | Test connection to a host | | link-ai models <address> | List available models | | link-ai tailscale status | Show Tailscale network status | | link-ai tailscale ip | Show this machine's Tailscale IP | | link-ai tailscale peers | List Tailscale peers |

Host Options

link-ai host [options]

  -u, --upstream <url>     Upstream LLM API URL (required)
  -p, --port <number>      Port to listen on (default: 8080)
  -a, --auth <key>         API key for authentication
  -t, --tailscale          Show Tailscale IP
  --all-interfaces         Listen on 0.0.0.0 (default: 127.0.0.1)

Security

• No ports exposed to the internet — use Tailscale or SSH tunnel
• Optional API key auth — --auth <key> on host, same key on client
• CORS enabled — works from browser GUIs
• Your data stays on your machines — link-ai doesn't store or send anything

Using with Tailscale

Tailscale gives your machines stable IPs that work from anywhere, without port forwarding.

# Install Tailscale on both machines
# https://tailscale.com/download

# On both machines
tailscale up

# Get your Tailscale IP
link-ai tailscale ip

# List connected machines
link-ai tailscale peers

# Start host with Tailscale
link-ai host --upstream http://localhost:11434 --all-interfaces --tailscale

# Connect from anywhere
link-ai chat 100.64.0.5:8080 -m "hello from across the world"

Dashboard

The dashboard lets you manage multiple LLM hosts from one place.

link-ai dashboard --port 9090

Then open http://localhost:9090/dashboard.

Features:

• Add/remove hosts with display names
• See online/offline status, latency, available models
• Auto-refresh every 10 seconds
• Click to open chat with any host
• Dark/light theme

Architecture

link-ai/
├── bin/link-ai.js              # CLI entry point
├── src/
│   ├── host/
│   │   ├── server.js           # Proxy server (HTTP + WebSocket)
│   │   └── tailscale.js        # Tailscale CLI integration
│   ├── client/
│   │   ├── api-client.js       # HTTP client + connect/models
│   │   ├── chat-cli.js         # Interactive CLI chat
│   │   └── gui-client.js       # Web GUI server
│   ├── web/
│   │   ├── dashboard-server.js # Dashboard server
│   │   └── public/             # Static frontend files
│   └── shared/
│       ├── config.js           # Config parsing
│       └── utils.js            # Utilities

Supported Backends

Any LLM server with an OpenAI-compatible API:

• Ollama — http://localhost:11434
• vLLM — http://localhost:8000
• llama.cpp server — http://localhost:8080
• LocalAI — http://localhost:8080
• text-generation-webui — http://localhost:5000
• Any OpenAI-compatible API

Development

git clone https://github.com/yourusername/link-ai.git
cd link-ai
npm install
node bin/link-ai.js --help

License

MIT

Install · Commands · Dashboard · Security

Built with Node.js. No frameworks. No build step. Just works.