How to Set Up a Remote Ollama Server + LM Studio Client: The Ultimate Guide

Introduction

Running Large Language Models directly on your primary workstation sounds convenient until your laptop fans start sounding like a jet engine and simple tasks suffer from system lag. Offloading heavy inference to a dedicated remote server (like a home lab server with a high-end GPU, an older gaming rig, or a cloud VPS) frees up your local machine while letting you keep the polished user interface of desktop applications.

Connecting LM Studio on your client laptop to a remote Ollama instance gives you the best of both worlds. Your thin client remains cool, quiet, and responsive, while the heavy lifting happens elsewhere. In this guide, we walk through configuring your server host, securing the network connection, setting up the client, and optimizing performance. Before choosing your stack, you can also review our comparison of vLLM vs llama.cpp vs Ollama benchmarks to understand the performance profiles of each inference engine.

The setup is straightforward. Instead of running the inference engine locally, the client interface communicates with the server via its HTTP API. The remote server runs the background service and holds all the model files. The client workstation sends token payloads and receives the text stream back, using negligible local RAM and CPU.

First, you need Ollama running on your server. If you are setting this up on Linux, run the standard installation command:

curl -fsSL https://ollama.com/install.sh | sh

Setting host environment variables

By default, Ollama binds to localhost (127.0.0.1) on port 11434. To accept connections from other computers, you must configure the service to listen on all interfaces (0.0.0.0).

On Linux (systemd)

Edit the systemd service file:

sudo systemctl edit ollama.service

Add these lines to configure the environment variable:

[Service]
Environment="OLLAMA_HOST=0.0.0.0"

Save the file, reload the systemd manager, and restart the service to apply changes:

sudo systemctl daemon-reload
sudo systemctl restart ollama

On Windows

If your server runs Windows, open the Start menu, search for "Environment Variables", and choose "Edit the system environment variables". Under User variables, click "New", set the variable name to OLLAMA_HOST and the value to 0.0.0.0. Close Ollama from the system tray and restart it.

Verifying the service is accessible

Test the endpoint from another machine on your local network to verify the configuration:

curl http://<server-ip>:11434

If it returns "Ollama is running", the server is listening correctly.

Option A: SSH Tunneling (Recommended for Home Labs)

An SSH tunnel creates an encrypted link between your client machine and the server. This allows you to bind Ollama to localhost on the server, keeping the port completely closed to the outside world.

Run this command on your client laptop to map your local port 11434 to the server:

ssh -L 11434:localhost:11434 user@server-ip

Leave this terminal window open. Your local applications can now talk to the remote server by sending requests directly to localhost.

Option B: Private Overlay Networks (VPN)

If you want a connection that does not require keeping a terminal window open, use a service like Tailscale. Install Tailscale on both the server and your client laptop. Once they are on the same virtual network, you can bind Ollama to your server's Tailscale IP instead of 0.0.0.0.

For example, on Linux, set the variable to your specific Tailscale address:

Environment="OLLAMA_HOST=100.x.y.z"

This restricts connection access only to devices on your private VPN.

With the secure connection established, launch LM Studio on your client laptop.

• Configure the Endpoint:

  Click on the Settings icon in LM Studio. Navigate to the API / Server section. Under custom provider settings, set the target address. If you are using the SSH Tunnel method, set the address to http://localhost:11434. If you are using a VPN like Tailscale, use the server's private network IP (e.g., http://100.x.y.z:11434).

• Verify Model List:

  When you select the model dropdown in LM Studio, it should automatically query the remote server and show the list of models you have downloaded on the host. Select the model you want to run.

• Model Management:

  Remember that the models must be downloaded and stored on the remote server, not your client laptop. To pull a new model, run ollama pull modelname in the server terminal. Once the download completes, the client will immediately recognize it.

Connection refused

If your client cannot reach the server, double-check that you restarted the service after editing environment variables. You should also check your server firewall settings. On Linux, ensure that UFW permits traffic on port 11434 if you are connecting directly over a VPN:

sudo ufw allow 11434/tcp

VRAM allocation and GPU usage

To verify that Ollama is utilizing your remote GPU rather than falling back to the CPU, run the monitoring tool on the server during an active request:

nvidia-smi

You should see the ollama_llama_server process in the GPU process table, along with associated VRAM allocation. If it is missing, verify that your CUDA drivers are correctly installed on the server.

Network bottlenecks

If you are running this over a wireless connection, you might notice input lag. Time-to-first-token is sensitive to network latency. Using a wired ethernet connection for your server or connecting via a low-overhead network topology helps keep latency minimal.

Frequently asked questions