Short answer. Plan on at least 2 GB of RAM for OpenClaw 2026.11.1. OpenClaw's FAQ gives 1 GB as the absolute minimum and 2 GB or more as the recommendation, but in our tests 1 GB without swap was not enough: the Gateway from the npm package took 74 seconds to answer and lost two of its processes to the out-of-memory killer, and the official Docker image was killed outright. At 2 GB both ran, using about 1.1 to 1.5 GB at rest before any conversation, which leaves little room. Busy agents, several chat apps, browser automation or local models need more.
Checked on 6 November 2026, accurate for OpenClaw 2026.11.1.
These numbers come from our own tests on a Linux server, each in a throwaway container limited to 1 CPU and with no swap, and with no AI model answering: the state an agent is in before its first message. Figures in MiB and GiB are as Linux reports them (1 GiB is about 1.07 GB). We say where a number comes from OpenClaw's docs instead.
What OpenClaw's docs say
VPS: absolute minimum 1 vCPU, 1 GB RAM and about 500 MB of disk; recommended 1 to 2 vCPU and 2 GB+ RAM "for headroom (logs, media, multiple channels)". Node tools and browser automation "can be resource hungry".
VM: the same 1 GB minimum, 2 GB+ for multiple channels, browser automation or media tools.
1 GB servers: OpenClaw's DigitalOcean guide adds a 2 GB swap file and suggests a smaller model if you hit out-of-memory errors on large prompts.
Building the Docker image from source needs at least 6 GB of RAM. The pre-built image avoids that.
What we measured
Setup
Memory cap
Result
npm package, Node.js 26 container, no chat app
2 GB
Web listener up after 9.9 seconds, ready after 18.6 seconds. The OpenClaw process used 1,062 to 1,183 MiB at rest, and the container touched 2,033 of its 2,048 MiB once, 30 seconds after start
npm package, Node.js 26 container, no chat app
1 GB
Web listener up only after 73.5 seconds. OpenClaw printed its own memory-pressure warning at 850 MiB and two of its child processes were killed for lack of memory
Official image ghcr.io/openclaw/openclaw:2026.11.1, no chat app
2 GB
1,349 MiB 30 seconds after it answered its health check, 1,106 MiB two minutes after, highest 1,439 MiB. No out-of-memory events
Official image, no chat app
1 GB
Reached "ready", then the whole container was killed for lack of memory about 54 seconds after start (exit code 137)
Our panel's start script in a test container (npm package, Node.js 26), a chat app set up with a test token
2 GB
1.24 to 1.36 GiB at rest, whole container
Installing with npm
2 GB
Highest use during npm install: 1,559 MiB. The install itself took 648 MB of disk, plus 328 MB of npm cache
Why so much before you have said a word: at start, OpenClaw 2026.11.1 loads 14 plugins (anthropic, browser, canvas, cua-computer, device-pair, file-transfer, geolocation, github, linux-node, memory-core, ollama, openai, talk-voice and xai, in the official image's log), and the Gateway process alone holds about 1 GiB.
What makes it use more
We have not yet measured an agent in daily use, so this list is what OpenClaw's docs name, not numbers of ours:
Several chat apps and media: the docs' reason for 2 GB+ "headroom".
Browser automation: it needs Chromium (the -browser Docker images include it), which the docs call resource hungry.
Local models: OpenClaw's DigitalOcean guide says local LLM inference does not fit in 1 GB and recommends API-based models instead.
The install and updates: npm install peaked at 1,559 MiB in our test, more than the Gateway at rest.
How to tell you are short of memory
OpenClaw logs a memory-pressure warning.
A process or container ends with exit code 137 (killed by the kernel's out-of-memory killer). OpenClaw's Linux docs explain that its systemd service keeps the Gateway alive when only a child process is killed, and that repeated kills mean you should raise the memory limit or reduce concurrency.
The Gateway is very slow to come up: its web listener answered after 74 seconds at 1 GB, against 10 seconds at 2 GB in our test.
Memory on PlayNHost
OpenClaw plans have 2 GB, 4 GB or 8 GB. There is no 512 MB or 1 GB OpenClaw plan, because of the results above, and the start refuses to run OpenClaw on a server with less than 1,900 MB, with a sentence that says so instead of an agent that keeps getting killed.
2 GB is the minimum. It runs OpenClaw at rest: in our tests the whole server used 1.24 to 1.36 GiB of its 2 GiB before any conversation, and touched almost all of it once while starting.
4 GB is what our panel recommends, because 2 GB leaves little room for what the docs list above. That is our judgment from the at-rest numbers: we have no measurements of busy agents yet. If your agent is killed for memory on any plan, the next size up is the fix.
Every start installs or updates OpenClaw only when the version changed, so the 1,559 MiB install peak happens on the first start and after a new release, not every time. See how to host OpenClaw 24/7 for the full setup, or create an OpenClaw server.
PlayNHost is not affiliated with the OpenClaw Foundation. OpenClaw is open-source software under the MIT licence.
We use cookies to keep the site working, to understand how it is used, and to measure our marketing. You can accept all, accept essential cookies only, or choose per category.We use cookies to keep the site working, to understand how it is used, and to measure our marketing. They are on by default here. You can switch off anything you do not want.Read our privacy policy
