OpenClaw Hosting on Your Own VPS
A private AI agent instance with its own sandbox and persistent memory, on hardware you control.
Size a server for OpenClaw
Every stop is a real configuration you can deploy. Priced per 30 days, billed by the hour, exclusive of 18% GST.
What OpenClaw is
OpenClaw is a self-hosted AI agent. It runs autonomously against tasks you give it, keeps memory between sessions, and uses a browser sandbox to reach the web — which makes it materially different from a chat interface, and materially different to host.
The defining characteristic for hosting purposes is that agent workloads execute code and make outbound network requests that you did not write in advance. That is the whole point of an agent, and it is also why a dedicated instance matters: you want that activity in an environment where it cannot reach anything else you care about.
Model access is separate. OpenClaw calls model APIs with your own keys, so inference is billed by whichever provider you use, not by the server. The VPS runs the agent, its memory store and its sandbox — not the model.
Why run OpenClaw on a VPS
A dedicated instance, so agent activity is isolated from anything else you run.
Your model keys stay in your own environment.
Persistent storage for agent memory across restarts.
Root access and firewall control, which matter more than usual for a workload that makes its own outbound requests.
Snapshots before version changes, so an upgrade is reversible.
OpenClaw features
Autonomous task execution with persistent memory
Browser sandbox for web access
Bring your own model provider keys
Isolated environment with no shared tenancy
Snapshot and restore around version changes
Custom domain with TLS
Recommended server
A starting point for OpenClaw, not a hard floor — size it on the sliders.
| Memory | 8 GB — the browser sandbox is the memory-hungry part |
|---|---|
| CPU | 4 vCPU for responsive sandbox sessions |
| Storage | 80 GB for memory, artefacts and snapshots |
| Operating system | Ubuntu 22.04 or 24.04 LTS |
How deployment works
OpenClaw is installed on the instance with its browser sandbox and a persistent store for agent memory, behind TLS on your own domain. You supply model provider keys after handover. Because agent workloads run arbitrary code and reach the open internet, our acceptable use policy applies from the moment the instance starts.
What people run it for
Research and monitoring
Long-running tasks that watch sources and report, without a person driving each step.
Private automation
Agent workflows over data you would not send to a hosted agent service.
Development and evaluation
A disposable instance for testing agent behaviour, snapshot and roll back.
Team tooling
An internal agent behind your own domain and access controls.
OpenClaw hosting questions
Related applications
n8n Hosting Workflow automation with unlimited executions and your credentials on your own server. 
Windmill Hosting Scripts become APIs, workflows and internal UIs — with executions bounded by your server, not a plan. 
LM Studio Hosting Serve open-weight models from your own machine, with an OpenAI-compatible API. 
Anaconda Hosting Conda environments and Jupyter on a server, for work that outlasts a laptop session. 
MCP Server Hosting Expose your own tools and data to AI assistants, from a server you control. Deploy OpenClaw on your own server
Root access, your choice of size, and no container between you and the application.