Skip to content
Aventail™ investigation into UK government AI dependence featured in Computer Weekly.Read the story
Aventail™
All guides

Peer-to-peer AI routing: use models on another computer

4 min read

Use a model on another computer running Aventail Link. Add a same-account, same-network device as a backup and verify where requests run.

Aventail™ peer-to-peer AI routing guide, showing a pink request path from a laptop to a separate workstation above silver chainmail.

Peer-to-peer AI routing sends a model request to another computer running Aventail™ Link. Your application talks to Link’s local endpoint on your own computer; Link can forward the request to the other computer’s model and computing power.

In the current simple Control workflow, you add the other computer as a backup destination. Link can try it when the destination before it cannot be reached. This guide follows that setup, which is available to all accounts.

Use your own software or a compatible OpenAI-compatible chat-completions client, following the local API walkthrough. For supported apps, Aventail™ Connect configures their connection to Link automatically. API clients and Connect apps use the routing setup on the originating computer.

What you need

  • Two computers on the same network. They must be able to reach one another. A guest Wi-Fi network or device-isolation setting can prevent that.

  • The same Aventail™ account on both. Install Link, sign in and register both computers to that account.

  • A suitable model on the other computer. Prepare a compatible model there and keep Link running. Select a model that supports what your app needs.

Cross-network and cross-account routing are not part of the current setup. Use the computers already registered to your own account; the feature is not a public marketplace for other people’s machines.

For example, a laptop can use a model installed on a desktop on the same office network. The desktop executes the request. The computers do not pool their memory or split one request across their GPUs.

1. Prepare the other computer

Open Aventail™ Link on the computer that will provide the model. Sign in to the same account as your working computer and finish registering it. Download and prepare a model that can run on its hardware.

Check that the model is available on that device in Control. Keep the computer awake and reachable while you want to use it. An account entry by itself does not show that a machine can accept requests now.

2. Add it to your working computer’s route

  1. Open Aventail™ Control and select the computer where you use your apps under Devices.

  2. Open Routing, then Set up routing or Edit routing. Keep or choose the first model for that computer.

  3. At Add a backup?, select Add a backup and Another device.

  4. Choose the other computer under Device, then choose one of the models available on it.

  5. Select Add backup, review the destination order, then Apply to [device].

If Control shows No other devices yet, register the second computer with Link using the same account. If a model is missing, check its availability on the second computer.

You can also keep a downloaded model on your working computer as the final backup. That is a separate destination from the model on the other machine.

3. Wait for the settings to apply

Wait for Applied on device before testing. If an update is waiting or has failed, the previously applied choices remain active.

Link makes routing decisions on the computer where the request originates. Control manages the settings outside the model request path.

4. Verify a request reaches the peer

Start with a harmless prompt from your client through the local Link endpoint. Check the normal primary route first.

Then test a situation where the destination before the peer is unreachable, while both computers stay connected to each other. If your primary destination is a cloud provider, use a controlled network test that removes internet access but preserves the local network. Turning off Wi-Fi on the laptop would also remove its path to the peer.

Check the request records reported in Control’s Requests view alongside Link’s routing status. Look for the peer’s model and reported execution destination. If the available record does not establish where it ran, check diagnostic information or contact support.

Restore the original connection after testing. Each new request starts at the top of the applied order; a peer backup is used when an earlier destination cannot be reached.

If the connection does not work

Check both computers are on the same network and account, Link is running, the peer is awake, and the selected model is available. Check local network permissions and firewall rules with whoever manages the network. Do not expose the model endpoint to the public internet to work around a local connectivity problem.

Any permitted destination that receives a request processes the data sent to it. Choose a peer you are authorised to use for the work, and check the connected app’s storage and other network activity separately.

New to the terminology? Read What is routing? For a cloud-first setup that can finish on your own computer, see cloud routing with a local backup.

We use optional cookies and tracking to identify visiting companies, load our contact form, measure Google Ads download clicks and understand page use with PostHog on our homepage, Connect and How it works pages, Ollama and LM Studio comparison guides and campaign pages. Cookie policy