OpenClaw runs on a small server when it uses an online AI model, and memory, not processor power, is what runs out first. OpenClaw's own DigitalOcean guide starts with 1 GB of memory plus swap, and suggests 2 GB if that runs out. Two open reports in the project's issue tracker describe the Gateway's memory growing over days.
This guide uses OpenClaw's documentation, its issue tracker and each host's own OpenClaw page, read on 27 October 2026. The table further down compares the routes with this week's prices.
Who this is for
- you want OpenClaw running all the time, reachable from Telegram, WhatsApp or another chat app
- you are choosing between a managed OpenClaw plan, a one-click server and installing it yourself
- you want to know how much memory to rent, from OpenClaw's own guide and the hosts' advice
Not for you if you have not picked an agent app yet: see where to host an AI agent.
Key numbers
- 1 GB of memory, plus swap, in OpenClaw's own DigitalOcean guide; 2 GB if that runs out Official page,
- 2 GB the memory Kamatera's OpenClaw page suggests for a personal assistant Official page,
- 2026.10.8 version the newest OpenClaw release on 27 October 2026; DigitalOcean's 1-Click app had 2026.10.7 Official page,
- 18789 port the Gateway's port, which OpenClaw keeps on the server itself by default Official page,
Which route fits you
Our guide to where to host an AI agent explains the four routes in general. For OpenClaw the choice comes down to two questions:
Where should OpenClaw run?
Question 1 Should someone else install updates and keep backups?
- Yes: a managed OpenClaw plan
- No: go to question 2
Question 2 Are you comfortable with a terminal and Docker?
- Your chat app (Telegram, WhatsApp, Slack): connects to OpenClaw Gateway on your server (messages).
- OpenClaw Gateway on your server (state, skills, workspace): connects to AI model (prompts over HTTPS).
- AI model (Claude, GPT or another, by API)
Managed OpenClaw plans: what the host does for you
Three hosts we cover run OpenClaw for you. You add your own model key and connect your chat apps; the host installs, updates and backs up the server.
- Cloudways runs OpenClaw for you, with managed SSL, automated daily backups and updates it tests before release. Its public pages name no regions, so its prices are not in our records yet
- Hosting.com sells a fully managed server that starts with OpenClaw installed, and patches, monitors and backs it up
- Hostinger's AI automation apps plan covers OpenClaw or one of its other apps, and you can switch between them
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| Hosting.com Managed Server 4GB (OpenClaw pre-installed) 2 vCPU, 4 GB | $36/month 1-year term, paid up front; introductory first-year price, then $40 checked 25 Oct 2026 | No offer | Visit |
Model usage is paid separately to the model provider.
One-click OpenClaw servers: what each host's page says
Hostinger, DigitalOcean, Kamatera and Contabo can put OpenClaw on a new server when you order it. Their own pages describe the set-up in different detail:
None of the four pages says how the Gateway is reached, so check its port and sign-in before you connect accounts.
| Host | What its page says it sets up | Memory it suggests | HTTPS |
|---|---|---|---|
| Hostinger | a one-click app in its Docker manager, with a built-in firewall | not stated | not stated |
| DigitalOcean | a 1-Click app: OpenClaw 2026.10.7 on Ubuntu 24.04 | not stated on the listing | not stated |
| Kamatera | a marketplace image, openclaw_latest on Ubuntu Server 24.04 | 2 GB for a personal assistant, 4 GB for a developer set-up | not stated |
| Contabo | OpenClaw from the order page, at no extra cost | its OpenClaw plans start at 8 GB | you add a domain and a Let's Encrypt certificate, its FAQ says |
OpenClaw's own DigitalOcean guide suggests starting from a clean Ubuntu 24.04 image rather than a third-party marketplace image you have not reviewed. If you use an image, check its OpenClaw version and its open ports before you connect your accounts.
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| Hostinger KVM 1 1 vCPU, 4 GB | $6.49/month 24-month term, paid up front; Black Friday first-term price, then $11.99 checked 25 Oct 2026 | Black Friday prices on VPS plans and AI automation apps, for the first subscription term only checked 30 Oct 2026 | Visit |
| Hostinger KVM 2 2 vCPU, 8 GB | $8.79/month 24-month term, paid up front; Black Friday first-term price, then $14.99 checked 25 Oct 2026 | Black Friday prices on VPS plans and AI automation apps, for the first subscription term only checked 30 Oct 2026 | Visit |
| DigitalOcean Basic Droplet, 2 vCPUs, 4 GiB 2 vCPU, 4 GB | $24/month billed per second (60-second minimum), capped at this monthly price · VAT not stated checked 25 Oct 2026 | New accounts get a $5 credit that lasts 90 days and covers every product except SaaS add-ons checked 29 Oct 2026 | Visit |
| Kamatera Cloud server Type B (General Purpose), 2 cores, 4 GB 2 vCPU, 4 GB | $39/month monthly, no long-term contract · VAT not stated checked 25 Oct 2026 | A 30-day trial of one cloud server in a configuration worth up to $100 checked 29 Oct 2026 | Visit |
| Contabo Cloud VPS 4 4 vCPU, 8 GB | €4.40/month 24-month term, paid up front; the price applies to the first 24 months and to the EU region in Germany; other regions cost more each month checked 25 Oct 2026 | No offer | Visit |
Some of the lowest prices need a long term paid in advance; the term is under each price.
OpenClaw on a VPS you set up yourself
On a plain server you install OpenClaw with its own installer, or in Docker as its Hetzner guide does. Its install page asks for Node 24.22.1 or later, or 26.12.1 or later, and its installer adds Node if it is missing.
Its server guide keeps the Gateway on the server itself and reaches it through an SSH tunnel or Tailscale. If you let the Gateway listen on your local network or tailnet instead, it needs a shared secret. That is a token or a password, unless a trusted proxy handles sign-in.
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| Hosting.com Self Managed Linux 4GB 2 vCPU, 4 GB | $6.49/month 3-year term, paid up front; introductory price for the first three years, then $19 checked 25 Oct 2026 | No offer | Visit |
| Liquid Web Cloud VPS (General), 2 vCPUs, 4 GB 2 vCPU, 4 GB | $17/month monthly · VAT not stated checked 25 Oct 2026 | No offer | Visit |
| UpCloud Starter, 2 CPU cores, 4 GB 2 vCPU, 4 GB | €12/month billed by the starting hour, at most 28 days a month; no minimum term checked 25 Oct 2026 | No offer | Visit |
| Cherry Servers Cloud VPS 4 (Gen 2) 4 vCPU, 4 GB | €9/month monthly, no long-term contract checked 25 Oct 2026 | No offer | Visit |
Prices in euros and US dollars as each provider shows them.
How much memory OpenClaw needs
OpenClaw's install page lists no memory figure. Its DigitalOcean guide and the hosts' own pages give these starting points:
2 GB is the floor most of these sources suggest; a local model needs far more.
| Source | What it suggests |
|---|---|
| OpenClaw's DigitalOcean guide | 1 GB with swap; a 2 GB server, or API models instead of local ones, if memory runs out |
| Kamatera's OpenClaw page | 2 GB for a personal assistant; 4 GB for a developer set-up; 8 GB or more for local models |
| Contabo's OpenClaw page | plans from 8 GB, named for personal and power-user tasks |
Why memory can grow
Two open reports in OpenClaw's issue tracker describe the Gateway's memory growing until the system stops it. In the first, opened on 9 June 2026, a user saw it grow from about 350 MB to 15.5 GB over two to three days. That user ran it on a Mac with 8 GB of memory and custom plugins. The project marks it as its highest priority.
The second, opened on 29 September 2026, reports growth on version 2026.9.5 until the system killed the process at about 3 GB. Both were still open on 27 October 2026, when the newest release was 2026.10.8. Both came from Macs with plugins, so a plain server may behave differently.
Three habits limit the damage if it happens to you. Rent memory headroom above the guide's minimum, run OpenClaw as a service that restarts it when it stops, and stay on the newest release.
Does OpenClaw need a GPU?
No, if it uses a hosted model such as Claude or GPT, which its home page names. OpenClaw can also use local models; then the machine needs the memory, and often the GPU, that the model needs. Our guide to where to run an open-source LLM covers that.
Security comes first
OpenClaw can act on your e-mail, calendar and accounts, so treat its server as you would your own laptop. Keep the Gateway off the open internet, give the agent its own accounts and keys, and set a spending limit on the model API. Our guide to securing OpenClaw on a server goes step by step.
What it costs each month
The server is a fixed monthly price; the model is paid per token and grows with use. Contabo's own OpenClaw page warns that token costs can add up fast with an autonomous agent, so set a spending limit with your model provider. Our guide to what OpenClaw or Hermes Agent costs each month adds the server and the model together with a calculator.
The routes side by side
Only the managed plan takes updates off your hands; the rest leave the server to you.
| Route and plan | Who sets OpenClaw up | Who installs updates | Monthly price |
|---|---|---|---|
| Managed: Hosting.com Managed Server 4GB | Hosting.com | Hosting.com | $36 a month (checked 25 October 2026) |
| One-click: Hostinger KVM 2 | Hostinger's app | you | $8.79 a month (checked 25 October 2026) |
| One-click: DigitalOcean Basic, 2 vCPUs, 4 GiB | DigitalOcean's 1-Click app | you | $24 a month (checked 25 October 2026) |
| Own install: UpCloud Starter, 2 cores, 4 GB | you | you | €12 a month (checked 25 October 2026) |
How we checked
We read OpenClaw's documentation and each host's OpenClaw page on 27 October 2026. We checked the two memory reports and the newest release through GitHub's own interface the same day.
Our own measurements will be added here, with their dates, when our test bench runs them.
Mistakes to avoid
Renting the smallest plan with no headroom
Instead: Rent at least 2 GB, the floor most sources here suggest, plus a margin.
Exposing the Gateway to reach it from anywhere
Instead: Use an SSH tunnel or Tailscale, as OpenClaw's guide does.
Running an image you have not checked
Instead: Check its OpenClaw version and open ports first.
Running OpenClaw without a restart policy
Instead: Run it as a service that restarts it when it stops.
Skipping updates
Instead: Stay on the newest release; memory fixes arrive in updates.
Other questions
How much RAM does OpenClaw need?
OpenClaw's DigitalOcean guide starts with 1 GB plus swap and suggests 2 GB if that runs out. Kamatera's page suggests 2 GB for a personal assistant.
Can I run OpenClaw in Docker?
Yes. OpenClaw's documentation lists a Docker set-up among its server guides. Keep the Gateway reachable only through an SSH tunnel or Tailscale, as its server guide advises.
Which port does OpenClaw use?
The Gateway listens on port 18789, on the server itself by default, so it is not reachable from the internet unless you change that.
Sources and test notes
- OpenClaw documentation: install, read
- OpenClaw documentation: Linux server and VPS, read
- OpenClaw documentation: DigitalOcean, read
- OpenClaw home page, read
- OpenClaw issue #91588: Gateway memory leak, read
- OpenClaw issue #161108: Gateway memory growth on 2026.9.5, read
- OpenClaw releases, read
- Cloudways managed OpenClaw, read
- Hosting.com OpenClaw hosting, read
- Hostinger: OpenClaw application, read
- Hostinger AI automation apps, read
- DigitalOcean Marketplace: OpenClaw, read
- Kamatera: OpenClaw, read
- Contabo: OpenClaw hosting, read
Every page read on 27 October 2026; the issues and releases through GitHub's API.
Guides about OpenClaw hosting
-
Securing OpenClawA plain checklist that matches each OpenClaw risk to a setting, from OpenClaw's own security, update and backup guides, with the hosts' one-click set-ups.
-
Mac mini or VPSMac mini or rented server for OpenClaw: the Mac-only features, the server-plus-Mac set-up for iMessage, electricity from Apple's figures and VPS prices.