OpenClaw System Requirements
OpenClaw needs roughly 2 CPU cores, 2 GB RAM, and 2 GB storage, and runs on any OS, with Ubuntu 24.04 recommended. It's lightweight because the heavy lifting happens in the LLM API, not on your server.
OpenClaw is deliberately lightweight. The minimum specs are roughly 2 CPU cores, 2 GB RAM, and 2 GB of storage, and it runs on any operating system. Ubuntu 24.04 is the recommended OS, mostly because it is well-supported and what most guides assume.
It can be this modest because the expensive computation, the model inference, happens in the LLM provider’s cloud over an API. Your server orchestrates the agent loop and holds state; it does not run a model. That is why a small, cheap box is enough.
How it works
OpenClaw runs as a process (via Docker or Node) that manages the agent loop: receiving messages, calling the LLM, executing skills, and persisting memory. None of that is CPU- or RAM-hungry, so the bottleneck is almost never your hardware. It is your LLM API quota and your network connection.
A practical baseline:
- CPU: ~2 cores.
- RAM: ~2 GB (more headroom helps if you run many skills).
- Storage: ~2 GB (memory files and logs grow slowly).
- OS: any; Ubuntu 24.04 recommended.
- Network: stable outbound HTTPS to your LLM and channel APIs.
Almost any modern VPS clears this bar. For picking one, see best VPS for an AI agent.
Worked example
| Resource | Minimum | Comfortable |
|---|---|---|
| CPU | 2 cores | 2 cores |
| RAM | 2 GB | 4 GB |
| Storage | 2 GB | 10 GB+ |
| OS | Any | Ubuntu 24.04 |
Because requirements are low, the cost driver is not the server. Infra plus LLM API typically stays under $100/month, with the LLM usage doing most of the spending. See how much OpenClaw costs to run.
Try this in Liv
If you would rather not size, provision, or maintain a server at all, Liv runs the OpenClaw agent loop on managed infrastructure.
- Start a 14-day free trial at app.liv4all.com, no credit card.
- Message Liv on Telegram, the default channel.
- Connect Gmail and Calendar via Google OAuth.
- Optionally link WhatsApp (invite-only, needs a dedicated eSIM).
Onboarding is early access and batched.
Common questions
Can I run OpenClaw on a Raspberry Pi?
If it meets ~2 cores and 2 GB RAM, yes. The requirements are low enough for small single-board computers.
Does it need a GPU?
No. Inference happens via the LLM API, so no local GPU is required.
What OS should I use?
Any works; Ubuntu 24.04 is recommended for support and documentation alignment.
How much storage will it use over time?
Little. Memory files and logs grow slowly; 2 GB is a fine starting point.
What VPS is best?
Any that meets the minimum specs. Best VPS for an AI agent compares options.
Do I need a server at all?
Not if you use a managed service. Liv is an independent managed service built on OpenClaw, not affiliated with the project, and runs the infrastructure for you.