OpenClaw on SwissGPU

OpenClaw is a self-hosted AI agent gateway with a browser dashboard. You can run it on a SwissGPU Linux server and connect it to a local model runtime such as Ollama or vLLM.

Install and Configure

Connect using SSH, then run the official installer:

curl -fsSL https://openclaw.ai/install.sh | bash

Follow the onboarding wizard and choose a local gateway. If you need to return to configuration, run openclaw configure. Follow the official getting-started guide for the current prerequisites and service setup.

Connect a Model

Set up Ollama or vLLM on the same server, then follow OpenClaw's Ollama provider guide or vLLM provider guide to configure it. Choose a model that supports tool calling and fits your GPU memory. Small chat models are not necessarily suitable for agent tasks.

For Python environments on Ubuntu, install python3-venv rather than a package tied to a different Python version. Keep the model endpoint and gateway bound to loopback for this setup.

Open the Dashboard

Run this on your local computer, replacing the SSH destination with the one from your dashboard:

ssh -N -L 18789:127.0.0.1:18789 ubuntu@<server-host>

Keep the tunnel open and visit http://127.0.0.1:18789/. Use the gateway authentication configured during onboarding. The SSH connection needs IPv6 by default; see Network Specs for IPv4 options.

Troubleshooting

openclaw gateway status\nopenclaw health\nnvidia-smi

Check the provider configuration and available GPU memory if the model cannot load. Back up your workspace and configuration before stopping or reinstalling the bare metal server.