Guide · OpenClaw

OpenClaw requirements: RAM, CPU, Node.js and OS.

OpenClaw itself is light: its docs put the absolute minimum at 1 vCPU and 1 GB of RAM. What it does is what needs resources: the tool sandbox, browser automation, media, and above all running models locally. Here's what each one adds, and where to run it.

Short answer. With a hosted model, 2 vCPU and 2–4 GB of RAM is plenty for a personal OpenClaw. Turn on the Docker sandbox or browser tools and plan for 4 vCPU and 8 GB. Local models need a GPU or Apple silicon with lots of memory.

Hardware, by what you run

SetupCPURAMSource
Absolute minimum1 vCPU1 GB, ~500 MB diskOpenClaw docs
Recommended, hosted model1–2 vCPU2 GB+OpenClaw docs
Docker sandbox, browser tools, several busy channels4 vCPU8 GBOur guidance
Several agents, or a shared team claw8 vCPU16 GBOur guidance
Local modelsA GPU, or Apple silicon with plenty of unified memoryDepends on the model

OpenClaw's docs note that Node tools and browser automation "can be resource hungry", and that logs, media and multiple channels are what the extra headroom is for.

Software

  • Node.js: 24.16.0+ on the 24.x line, or 26.1.0+. The docs strongly recommend 24.22.1+ or 26.12.1+ for security. The installer script handles this for you.
  • Operating system: macOS or Linux, and Windows through WSL2. On a server, Ubuntu LTS (or another modern Debian or Ubuntu) is the best-tested path.
  • Docker: only if you turn on the tool sandbox, where it's the default backend. Podman and SSH backends also work.
  • Disk: about 500 MB for OpenClaw itself, plus room for logs, media, session transcripts and sandbox images.

What actually uses the resources

  • The gateway: light. It mostly waits on the network and your model provider.
  • The tool sandbox: each sandboxed session or agent runs in a container, which costs memory. Plan for more RAM if you turn it on, and you should (see step 5 of the safety guide).
  • Browser automation: a headless browser is the hungriest tool OpenClaw runs. If your agent browses a lot, give it 8 GB.
  • Local models: by far the biggest cost. With a hosted model like Claude, GPT or Kimi, the thinking happens on the provider's hardware and you need no GPU at all.

Where to run it

OpenClaw's own advice: run it on a VPS if you want 24/7 reliability, and locally if you want the lowest friction and are fine with it pausing when your machine sleeps. A dedicated host is "not required, but recommended for reliability and isolation".

HostGood forWatch out for
Your laptopTesting, local files, a visible browser windowPauses when it sleeps or updates; reaches all your stuff
A VPSAlways on, stable network, isolated from your filesHeadless browser only; you do the hardening unless the host does
A Mac miniiMessage, Apple apps, local models on Apple silicon$899 up front for the base model; on your home network
A Raspberry PiA cheap, always-on host for hosted modelsToo small for local models or heavy browser use

Need iMessage but want the gateway on a server? OpenClaw's docs describe running the gateway on Linux and reaching a Mac signed into Messages over SSH. You need a Mac for iMessage, not necessarily the machine that runs OpenClaw. More detail: Mac mini or VPS for your AI agent.

OpenClaw on Airbox: which size

SizeSpecGood forPrice
AB‑22 vCPU · 4 GBOne personal OpenClaw, hosted model, a few channels$15/mo
AB‑44 vCPU · 8 GBSandbox on, browser tools, busy channels. Our pick.$29/mo
AB‑88 vCPU · 16 GBA shared claw, or OpenClaw plus Hermes or Claude Code$58/mo

Every Airbox gets dedicated cores, OpenClaw installed on a current Node.js and the safety settings on. Early-access rates. How OpenClaw runs on Airbox.

Questions

How much RAM does OpenClaw need?

OpenClaw's docs give an absolute minimum of 1 GB and recommend 2 GB or more. With the Docker tool sandbox or browser automation turned on, plan for 8 GB.

Can OpenClaw run on a Raspberry Pi?

Yes. OpenClaw's docs list a Raspberry Pi-class box as a valid always-on host and have a Raspberry Pi install guide. It's best with hosted models, since a Pi can't run useful local models.

Does OpenClaw need a GPU?

No, not with hosted models like Claude or GPT, which do the heavy work on the provider's side. You only need a GPU, or Apple silicon with plenty of unified memory, to run models locally.

Which operating system is best for OpenClaw?

On a server, Ubuntu LTS or another modern Debian or Ubuntu release, which OpenClaw calls its best-tested Linux path. It also runs on macOS, and on Windows through WSL2.

Do I need a Mac for OpenClaw's iMessage support?

You need some macOS device signed into Messages, not necessarily a Mac mini, and not necessarily the machine running the gateway. OpenClaw's docs describe running the gateway on Linux and reaching a Mac over SSH for iMessage.

Sources: OpenClaw's providers, hardware and hosting FAQ, its README and its sandboxing docs, checked November 16, 2026. Mac mini price is Apple's US list price for the base M6 model. Next: How to run OpenClaw safely.

The right-sized box for your claw.

Claim an OpenClaw box · from $15/mo