Skip to content
EN
Add to Chrome

Model apps on your network or at an https address

How-toNew

For anyone whose models run on a bigger machine than the one with the browser: a desktop, a home server, or a box in the office. Riff writes with them the same way it writes with LM Studio or Ollama on this computer Models: pick and manage the AI model.

Add a model app at a network address, allow the one host, then pick its model.

In Riff Home › Models, the Model apps section always has LM Studio and Ollama on this computer. Next to them is Add a model app, for everything else. You can have up to 12 model apps in all.

Add a model app

  1. Click Add on the Add a model app card.
  2. Name: anything that tells it apart, like GB10 (tunnel).
  3. App: LM Studio, Ollama or Other (OpenAI-compatible). Pick Other for vLLM, a llama.cpp server or anything else that speaks the OpenAI chat API. The same menu has agents to chat with Connect a Hermes agent.
  4. Address: where the app answers, like 192.168.1.20:1234 or llm.yourdomain.com (see below).
  5. API key: only if the app asks for one. Leave it empty otherwise.
  6. Click Save and connect. The browser asks once to let Riff reach that one address. Allow it.
  7. The card reads Connected. Pick a model in its menu and click Use this model. Replies, posts and Chat now use it, as with an app on this computer.

To change LM Studio's or Ollama's address, click Edit on its card. Back to this computer puts it back to localhost. An app you added has Remove. If you remove the app Riff writes with, Riff goes back to Auto.

Which addresses work

You typeRiff usesFor
localhost or 127.0.0.1http://, this computerAn app on this computer
192.168.1.20, 10.0.0.5, studio.localhttp://, your networkAnother computer at home or in the office. Tailscale addresses (100.x.x.x) count too
llm.yourdomain.comhttps://Your own server behind a tunnel, like a Cloudflare Tunnel
  • Leave out the port and Riff adds the app's usual one: 1234 for LM Studio, 11434 for Ollama, 8000 for Other.
  • A plain http:// address on the public internet is refused, because what you type would cross the internet unencrypted. Riff says Use an address on your network (192.168.1.20:1234, name.local) or an https tunnel address.
  • Under the card's name, one line says where it is and what it holds, like Your network · 192.168.1.20:1234 · API key saved or Tunnel · llm.yourdomain.com.

Let the app accept other computers

Most apps only answer their own computer until you tell them otherwise. If a new card reads Not found, it says what to change on that computer:

  • LM Studio: start the server and turn on Serve on Local Network (Developer › Settings). Keep Enable CORS on.
  • Ollama: start it with OLLAMA_HOST=0.0.0.0, and still allow Riff with OLLAMA_ORIGINS Models: pick and manage the AI model.
  • Other: start the server listening on host 0.0.0.0.

A tunnel points at the app's port on that computer instead, so the app can keep listening on localhost.

API keys

An app reached from other computers, and anything behind a tunnel, should have a key. Riff sends it with every request, as Authorization: Bearer and the key.

  • In LM Studio, make one in Developer › Server settings › API tokens.
  • If an app wants a key you haven't given, the card reads Needs an API key with a box for it. A wrong one reads Key not accepted A model app or agent needs a key or a setting.
  • Keys stay on this computer, kept apart from your other settings where only Riff's own pages can read them. They are never in a backup Back up and restore.

What leaves this computer

When Riff writes with an app on another computer, the post you reply to, your point and your Voice go to that address, and nowhere else. With a tunnel, they pass through the tunnel's company (Cloudflare, say) on the way to your server. Riff only talks to addresses you add, and only after you click Save and connect or Connect.