Best VPS for OpenClaw in 2026

The short answer

Buy a 2 vCPU / 4 GB RAM x86 Linux VPS with 40 GB or more of SSD, billed month to month, and run the official pre-built Docker image. You do not need a GPU, because OpenClaw sends model calls to a remote provider. If you only want a chat bot on Telegram or Discord, a 2 GB plan is enough. If you are on a budget and can commit for a year, OVHcloud VPS-1 gives you 4 GB for $4.54 a month. If you want to experiment for free, Oracle's Always Free Arm tier works with caveats covered below.

This page is about raw servers. For ranked managed platforms see best OpenClaw hosting, and for the install steps themselves see OpenClaw on a VPS.

What OpenClaw needs from a server

The OpenClaw gateway is a Node.js process that owns your state and workspace, keeps your channel connections (Telegram, Discord, WhatsApp and others) open, and serves a control UI. The install documentation asks for Node 24.16+ or 26.1+ (Node 26 is recommended) on macOS, Linux or Windows. Neither the install page nor the Linux server page publishes a minimum RAM figure, so the numbers below come from the places that do give one, plus our own fleet.

Our own breakdown is on the system requirements page. The takeaway is the shape: a typical agent is small, but the tail is more than three times the median, so size for the tail.

Four things that change the sizing for OpenClaw

Generic “how much RAM for a bot” advice misses what is specific to this project.

• Do not build from source on a small box. Several hosting guides say you need 6 GB for OpenClaw. That figure is for a local source image build, and a build that runs out of memory is killed with exit code 137. The official images on ghcr.io/openclaw/openclaw and Docker Hub avoid the build entirely, which is why a 2 GB or 4 GB VPS is enough for running it.
• The browser image is a bigger download. The -browser variants (for example latest-browser) bake in Chromium for the Gateway-controlled browser. On Docker Hub the default 2026.9.8 image is about 1.2 GB compressed and the browser variant about 1.7 GB, so budget disk for a few image versions as you update. Slim variants (slim) are also published.
• Plugins install from npm on your server. Channels and skills that are not bundled are fetched at install time, which is a short spike of CPU, RAM and disk. A box with headroom absorbs it; a 1 GB box with swap makes it slow.
• Arm64 works. The official images are published for both amd64 and arm64, including the browser variant, and OpenClaw's Oracle guide notes that Node.js, Telegram and WhatsApp are pure JavaScript. The exception is helper binaries that a skill downloads, so check for a linux-arm64 release before you rely on one.

Disk is rarely the constraint. Chat history, workspace files and image versions grow slowly, and every plan in the table below has at least 40 GB. Prefer SSD or NVMe: the OpenClaw Linux server guide recommends SSD-backed disks for state and cache paths to avoid slow cold starts.

VPS plans compared (October 2026)

Prices checked October 2026 on each vendor's own pricing page or public pricing feed, in the currency the vendor shows. Hetzner's feed lists prices before VAT and Contabo says its price includes applicable taxes; for the others, check tax at checkout. Headline prices often depend on a long term, so read the last column.

Hetzner's cheaper CX and Arm CAX plans showed as unavailable when we checked, so we left them out rather than quote an old number. Note that Vultr's 2 vCPU / 4 GB plan at $20 is slightly cheaper month to month than Hetzner's CPX22, so Hetzner is not the price leader here; it is a good choice for EU data centres. Provider walkthroughs: Hetzner, DigitalOcean, Hostinger, Contabo and OVHcloud.

Our picks

• Best overall: a month-to-month 2 vCPU / 4 GB plan, which means Vultr ($20), Hetzner CPX22 (€19.49 before VAT), DigitalOcean ($24) or Akamai ($24). You pay more per month than the long-term deals but can resize or delete the server without losing a prepaid term, which matters while you learn how much your agent uses. This is our judgement on price and flexibility, not a benchmark of CPU or network quality.
• Cheapest comfortable: OVHcloud VPS-1. Four gigabytes and two vCores for $4.54 a month is the lowest price we found for a size that runs the browser image without strain. It is billed annually, so it is a commitment rather than a trial. Contabo Cloud VPS 4 gives 8 GB for an introductory €4.40 and is the better buy if you will also run other services, but that price is for the first 24 months only.
• Cheapest month to month: Vultr's 1 vCPU / 2 GB plan at $10 is the lowest 2 GB plan we read. It is enough for a chat-only bot without the browser image.
• Best free option: Oracle Cloud Always Free. Oracle's Always Free allowance is 1,500 OCPU hours and 9,000 GB hours of Ampere A1 a month, equal to 2 OCPUs and 12 GB running continuously, and OpenClaw has its own Oracle guide plus arm64 images. The same Oracle page says idle instances can be reclaimed after 7 days of under 20% CPU, network and memory use, and “out of host capacity” errors on creation are common. A quiet agent looks exactly like an idle instance. Fine for trying OpenClaw, risky for a bot you rely on. See free OpenClaw hosting options and is OpenClaw free.
• EU versus US: choose the region closest to you. Latency to your model provider dominates how fast the agent feels, so region mostly affects the control UI and webhooks. Hetzner's quoted plans are EU locations; Vultr, DigitalOcean and Akamai have many regions on both sides of the Atlantic.

Running OpenClaw next to n8n

A frequent setup is OpenClaw for conversation and n8n for scheduled workflows on the same server. The only sizing implication is additive: n8n is a separate service with its own memory use, so start from the 4 GB we recommend for OpenClaw with the browser image and add n8n's needs on top. In the table, the 8 GB plans (Contabo Cloud VPS 4 and Hostinger KVM 2) are the cheapest ways to get that room. Run each in its own container, give n8n its own volume, and watch docker stats for a week before deciding whether to resize. Do not publish either UI directly; put both behind the same private network.

24/7 setup checklist

Buying the server is the easy half. The step-by-step install is in OpenClaw on a VPS and OpenClaw with Docker; this is the short list that keeps it alive and private.

• Restart automatically. With Docker, set a restart policy so the gateway returns after a reboot. The image has a built-in health check that pings /healthz, and the Docker guide documents /readyz for channel-aware readiness. Without Docker, openclaw onboard --install-daemon installs a systemd user service, and the Linux server guide suggests TimeoutStartSec=90 on slow hosts.
• Firewall everything except SSH. Chat channels connect outward, so a bot needs no inbound ports. If you use Docker, read OpenClaw's security hardening note on the DOCKER-USER chain, because published container ports can bypass a simple firewall.
• Keep port 18789 private. The gateway defaults to loopback. Reach the control UI with ssh -N -L 18789:127.0.0.1:18789 user@host or with Tailscale. Auth is required by default, and a tunnel does not bypass the token.
• Treat the token and pairing as secrets. The Docker setup writes a gateway token to .env. A new browser must be approved with openclaw devices approve, and unknown Telegram senders get a pairing code rather than a reply when DM pairing is required.
• Update on your schedule. Pin a version tag rather than latest if you want predictable upgrades, enable unattended OS security updates, and keep the mounted state volume intact when you recreate the container.
• Back up the state. OpenClaw's guide says to treat the VPS as the source of truth and back up state and workspace regularly. A provider snapshot is not a substitute for a copy you can restore elsewhere.

VPS versus managed: the total cost

A VPS gives you root, any region and the lowest long-term monthly price. It also hands you every job around the agent. The table counts where the hours go, not just the invoice.

OpenClaw Launch hosts OpenClaw and Hermes Agent. Lite has 1 vCPU, 2 GB RAM, 10 GB storage, 1 instance and $1 a month of AI credits. Pro has 2 vCPU, 4 GB RAM, 40 GB storage, up to 3 instances and $10 a month of AI credits. Lite matches the 2 GB size this guide calls workable for a chat-only bot, and Pro matches the 4 GB we recommend with the browser image. Choose a VPS if you want root access, a specific region or other services on the same machine. Choose managed if you would rather spend the afternoon using the agent. More in managed vs self-hosted, cheapest OpenClaw hosting and the hosting page.

Frequently Asked Questions

Related Guides

• OpenClaw on a VPS — the install walkthrough
• OpenClaw System Requirements — measured RAM, CPU and Node versions
• OpenClaw with Docker and Docker Compose
• OpenClaw on Hetzner, DigitalOcean, Hostinger, Contabo, OVHcloud
• OpenClaw with Tailscale — keep the gateway private
• Best OpenClaw Hosting — managed platforms ranked
• Cheapest OpenClaw Hosting
• Best VPS for Hermes Agent