Access Your AI PC from Any Laptop with Tailscale

Image: Tailscale

Your most powerful PC doesn't have to be the one you're sitting at. With a free tool called Tailscale and a local model server, you can park a big AI machine in the corner and use it from any laptop in the house, or from a hotel room across the country.

I run an AMD Ryzen AI Halo desktop with 128GB of memory, enough to hold some very large local models. Rather than copy models around or run them on less-powerful hardware, I let my other computers borrow the Halo's brainpower over the network.

Tailscale builds a private, encrypted network between your devices. The AI PC runs a model server, and everything else on that network talks to it like a cloud AI service, except the cloud is sitting on my desk. This walkthrough uses LM Studio on the server and a Mac as the client, but the idea carries over to other servers and other machines. While setting this up, I used Claude to help troubleshoot any issues I ran into, and later used Claude to help format the following step-by-step instructions.

What you'll need

• An AI-capable PC to act as the server. I used an AMD Ryzen AI Halo Developer Platform with 128GB of memory. Any machine with enough memory for your model will do.
• A second PC or Mac to act as the client.
• A free Tailscale account. You can sign up with Google, Microsoft, or GitHub.
• LM Studio on the server. Lemonade Server works along the same lines.
• Optional: an AI agent or chat app on the client. I used Hermes, the open-source agent from Nous Research.

Step 1: Put Tailscale on every machine

Install Tailscale on the AI PC first, then on each device that will connect to it, and sign in with the same account everywhere.

Image: Dan Ackerman

1. Create your free account at tailscale.com.
2. On the AI PC, download the installer from tailscale.com/download, run it, and sign in.
3. Repeat on your other PCs and Macs.
4. Click the Tailscale icon in your menu bar or system tray. Every device you added should appear in the list.

That list is your private network, which Tailscale calls a tailnet. Only devices you have added to it can see each other.

Step 2: Start the model server on your AI PC

Most model servers listen only to the machine they run on by default. You need to tell yours to accept visitors.

1. Open LM Studio and go to the Developer tab.
2. Load the model you want to serve. I used Qwen 3.8 27B.
3. Turn on Serve on Local Network.
4. Set a generous context length when you load the model: 32K tokens or more if your memory allows. Agents send long prompts, and a small default is the most common reason you'll run into trouble.
5. Switch the server to Running. LM Studio's default port is 1234.

Using Lemonade Server instead? The idea is identical, but its default port and API path differ, so the steps are slightly different.

Step 3: Test the connection from another machine

Before you configure any app, confirm the second machine can reach the server. It takes one command.

1. On the client, open Tailscale, find your AI PC in the device list, and copy its address. It starts with 100 and looks like 100.x.x.x. You can also find it at login.tailscale.com/admin/machines.
2. Open Terminal and run the command below, swapping in your address and port:

curl http://100.x.x.x:1234/v1/models

3. You should get back a block of JSON listing the models LM Studio can serve. Copy the exact name of the one you want, including any prefix such as qwen/.

Or use the numeric address instead of the device's name. Friendly names rely on a Tailscale feature called MagicDNS, and the number works either way.

Step 4: Point your apps at the AI PC

Anything that accepts a custom OpenAI-compatible endpoint can now use your server. You need three values:

• Base URL: http://100.x.x.x:1234/v1
• API key: anything. The server doesn't check it, but some apps insist on one, so you might have to type none or just click Return/Enter.
• Model: the exact name from Step 3.

Here's how that looks in Hermes on a Mac (assuming you've already installed Hermes):

1. In Terminal, run "hermes model"
2. From the provider list, choose custom (direct API) at the very bottom. It lets you enter any address, which is what you need here.
3. Paste your base URL. Hermes checks it and reports how many models it can see.
4. Leave the API key blank, or enter none.
5. When asked for an API compatibility mode, choose Chat Completions.
6. Pick your model from the list Hermes pulled from the server.
7. Leave the context length blank for auto-detect, give the connection a display name, and you're done.

Image: Dan Ackerman

Start Hermes and say hello. The first reply can be slow while the model loads into memory on the AI PC.

If something goes wrong

• "Could not resolve host": your client doesn't recognize the device name. Use the numeric 100.x.x.x address instead.
• "Connection refused": the machines can see each other, but nothing is listening on that port. Make sure the server is running and Serve on Local Network is on.
• The command hangs, then times out: a firewall is probably blocking the port. On Windows, add an inbound rule in Windows Defender Firewall with Advanced Security that allows TCP on your server's port, and apply it to all network profiles.
• Replies get cut off, or the agent seems confused: the context length is too small. Reload the model in LM Studio with a larger one.
• The first reply is very slow: that's normal. The model is loading into memory.

Keep it private, keep it simple

Don't forward the server's port on your router. Local model servers usually have no login, so anyone who could reach the port could use your machine. Tailscale avoids the problem: only devices you've added to your network can reach the server, and traffic between them is encrypted.

One limit to keep in mind: your AI PC still has one set of hardware. If several devices send requests at once, they queue up behind each other.

That's the whole process: install, serve, test, connect. Once it's running, every laptop you own can borrow a big machine's brainpower from the couch, the office, or the road.

Want a machine of your own to serve from? Micro Center carries the AMD Ryzen AI Halo Developer Platform, and other Pro AI Solutions like AI-ready desktops, laptops, and parts.