How to run OpenClaw without a Mac mini: cloud and VPS options

#Does OpenClaw actually need a Mac mini?

No. OpenClaw's install page lists macOS, Linux and Windows as supported operating systems, and the same one-line installer covers macOS, Linux and WSL2. The Mac mini is popular for always-on agents because it is quiet, cheap to idle and sitting on a shelf you own, not because OpenClaw needs Apple silicon.

If you want the Mac mini case made properly, we have written it up separately in Running OpenClaw on a Mac mini. This post is the other half of that decision: what to do when you do not want a box in your house at all.

OpenClaw publishes a provider picker for exactly this. Its VPS page covers Azure, Daytona, DigitalOcean, exe.dev, Fly.io, GCP, Hetzner, Hostinger, Northflank, Oracle Cloud, Railway, Render, Raspberry Pi and Upstash Box, and the install page frames the same choice as "Deploy OpenClaw on a cloud server or VPS."

#What does OpenClaw need to run?

Four things, and none of them are exotic.

A supported OS. macOS, Linux or Windows. On Linux, note one packaging detail before you pick a distro: OpenClaw's Linux page says published AMD64 AppImages "are built on Ubuntu 22.04 and require glibc 2.35 or newer plus a libstdc++ that provides GLIBCXX_3.4.30." Ubuntu 22.04 and Debian 12 clear that bar; RHEL 9 and Rocky Linux 9 ship glibc 2.34 and cannot run the published AppImages. That constraint is about the desktop app rather than the CLI and gateway you will run on a server, but it is the kind of thing worth knowing before you standardise on a distro.

A Node runtime. OpenClaw requires Node.js 24.16+ or 26.1+. The docs say "Node 26 is recommended; the installer provisions Node 26 on macOS and Node 24 LTS on Linux when Node is missing." pnpm is only needed if you build from source.

The gateway process. This is the thing that has to stay up. The docs describe it as "One always-on process for routing, control plane, and channel connections" that hosts multiple agents through a single multiplexed port. It is a long-lived daemon, which is the entire reason people reach for dedicated hardware or a VPS instead of their laptop.

A port, bound carefully. The gateway's default port is 18789. Its default bind mode is loopback, and the docs add that "Inside a detected container environment the effective default is auto." Loopback by default is the right instinct: the gateway is not something you expose to the internet.

#How much RAM does an OpenClaw server need?

The lowest numbers OpenClaw publishes anywhere are on its Raspberry Pi page: "Minimum: 1 GB RAM, 1 core, 500 MB free disk, 64-bit OS", with "Recommended: 2 GB+ RAM, 16 GB+ SD card (or USB SSD), Ethernet." The per-provider pages agree with that shape. DigitalOcean's page recommends "Basic, Regular, 1 vCPU / 1 GB RAM / 25 GB SSD" plus a 2 GB swapfile; the Fly.io page is firmer, saying "512MB is too small; 2GB recommended."

So: 1 GB works, 2 GB is the comfortable floor. One figure that is easy to misread, the Docker page's "At least 6 GB RAM for a local source image build", is a build requirement rather than a runtime one. Pull the published image and it does not apply.

All of this assumes the agent calls a hosted model. Running models locally is a different sizing problem entirely, and a 1 GB VPS is not where you solve it.

#The documented Linux install

The shell installer is the same line you would run on a Mac:

On a server you usually want to skip the interactive wizard on the first pass and install the daemon explicitly. That is the flow OpenClaw documents on both its Raspberry Pi and DigitalOcean pages:

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --no-onboard
openclaw onboard --install-daemon

The Linux page documents an npm route as well, for when you want to control the Node version yourself:

npm i -g openclaw@latest --allow-scripts=openclaw
openclaw onboard --install-daemon

Then confirm the install with the three commands the docs name:

openclaw --version
openclaw doctor
openclaw gateway status

The full walkthrough, including systemd and firewall configuration, lives in OpenClaw on Linux (Ubuntu/Debian). For the provider-agnostic version, see OpenClaw on a VPS.

#The documented Docker install

If you would rather ship a container than manage a Node install, OpenClaw publishes images on two registries: ghcr.io/openclaw/openclaw:latest on the GitHub Container Registry and openclaw/openclaw:latest as a Docker Hub mirror. There is a ghcr.io/openclaw/openclaw:latest-browser variant that bundles Chromium.

The documented path is a script rather than a compose incantation, and it runs from the repo root, so start with a checkout:

git clone https://github.com/openclaw/openclaw.git
cd openclaw
./scripts/docker/setup.sh

The docs describe what that script does: it builds the gateway image locally as openclaw:local, or uses a pre-built image when OPENCLAW_IMAGE is exported; runs onboarding and prompts for provider API keys; generates a gateway token and writes it to .env; creates the legacy auth-profile secret key directory; and starts the gateway via Docker Compose.

Once it is up, open http://127.0.0.1:18789/ and paste the token from .env into the Control UI settings. If you lose the dashboard URL, the docs give you a one-liner under "Need the URL again?":

docker compose run --rm openclaw-cli dashboard --no-open

There is a manual flow too, for when you want to see each stage. It is four commands, in this order: build the image, run onboarding, write the gateway config, then bring it up.

BUILD_GIT_COMMIT="$(git rev-parse HEAD)"
BUILD_TIMESTAMP="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
docker build \
  --build-arg "GIT_COMMIT=${BUILD_GIT_COMMIT}" \
  --build-arg "OPENCLAW_BUILD_TIMESTAMP=${BUILD_TIMESTAMP}" \
  -t openclaw:local -f Dockerfile .
docker compose run --rm --no-deps --entrypoint node openclaw-gateway \
  dist/index.js onboard --mode local --no-install-daemon
docker compose run --rm --no-deps --entrypoint node openclaw-gateway \
  dist/index.js config set --batch-json '[{"path":"gateway.mode","value":"local"},{"path":"gateway.bind","value":"lan"},{"path":"gateway.controlUi.allowedOrigins","value":["http://localhost:18789","http://127.0.0.1:18789"]}]'
docker compose up -d openclaw-gateway

Gateway state is mounted at /home/node/.openclaw, and the browser variant adds a Playwright cache at /home/node/.cache/ms-playwright. Configuration comes in through environment variables including OPENCLAW_IMAGE, OPENCLAW_GATEWAY_TOKEN, OPENCLAW_EXTENSIONS and OPENCLAW_INSTALL_BROWSER. Two health endpoints, which the docs note need no authentication, tell you whether it came up:

curl -fsS http://127.0.0.1:18789/healthz
curl -fsS http://127.0.0.1:18789/readyz

Our own container walkthrough is OpenClaw + Docker.

#Which hosting route should you pick?

Four routes, priced from each vendor's own page as of September 19, 2026. Every one of them runs the gateway on Linux, so the install above is the same in all four.

#Oracle Cloud Always Free

This is the cheapest route that exists, because it costs nothing. Oracle's own Always Free documentation states that "All tenancies get the first 1,500 OCPU hours and 9,000 GB hours per month for free for VM instances using the VM.Standard.A1.Flex shape, which has an Arm processor. For Always Free tenancies, this is equivalent to 2 OCPUs and 12 GB of memory." You also get "a total of 200 GB of Block Volume storage, and five volume backups", plus "10 TB per month of outbound data."

Twelve gigabytes of RAM for nothing is six times OpenClaw's comfortable floor.

OpenClaw's own Oracle page matches those numbers: shape VM.Standard.A1.Flex, "2 (or up to 4)" OCPUs, "12 GB (or up to 24 GB)" of memory, a "50 GB (up to 200 GB free)" boot volume, and Ubuntu 24.04 (aarch64) as the image.

Because the shape is Arm, check your architecture before installing CLI helpers. The docs tell you to run uname -m, which "should print aarch64", and to check for an "aarch64 / linux-arm64 release before installing" anything without pre-built Arm support. What the Oracle page does say about OpenClaw itself is reassuring: "Most OpenClaw features work fine" on Arm, with some optional native binaries needing to be sourced separately. The documented image for the instance is Ubuntu 24.04 (aarch64), and the documented install is the ordinary Linux path.

Our step-by-step version is OpenClaw on Oracle Cloud Free Tier.

#DigitalOcean

The most boring option, which is a compliment for something that has to stay up. DigitalOcean's pricing page lists a basic droplet with "1 GiB" memory, "1 vCPU", "25 GiB" SSD and "1,000 GiB" transfer at $6.00/month. That is exactly the size OpenClaw's DigitalOcean page recommends, and the docs put the cost at "~$6/month for the 1 GB Basic plan."

If you want more headroom, the same page lists "2 GiB" / "1 vCPU" / "50 GiB" / "2,000 GiB" at $12.00/month. There is a "512 MiB" / "1 vCPU" / "10 GiB" tier at $4.00/month, but 512 MB is below what any OpenClaw page recommends, so the $6 tier is the real entry point.

The documented setup is six steps, in this order: create a droplet on Ubuntu 24.04 LTS with SSH key auth; connect and install, which covers Node 24 LTS, OpenClaw and a non-root openclaw user; run onboarding; add swap, recommended for 1 GB droplets; verify the gateway; and access the Control UI over an SSH tunnel or Tailscale Serve. The docs are specific about avoiding third-party marketplace images.

#Fly.io

Fly is the route to pick if you would rather define the deployment in a file than administer a server. The core of the documented flow starts from a checkout, and fly deploy comes after you have edited fly.toml and set your secrets:

git clone https://github.com/openclaw/openclaw.git
cd openclaw
fly apps create my-openclaw
fly volumes create openclaw_data --size 1 --region iad
# edit fly.toml to match your app name and requirements
fly secrets set OPENCLAW_GATEWAY_TOKEN=$(openssl rand -hex 32)
fly secrets set ANTHROPIC_API_KEY=...
fly deploy

After the deploy, the docs have you fly ssh console in, create /data/openclaw.json, and restart the machine before the gateway is actually configured.

The docs specify a shared-cpu-2x Machine with 2048mb of memory, and say a 1 GB volume is "usually enough". Fly's pricing page puts shared-cpu-2x with 2 GB at $11.83 per month and 4 GB at $22.22 in the Amsterdam region, with volumes at "$0.15/GB per month of provisioned capacity" and North American and European egress at "$0.02 per GB". Prices vary by region, and Fly notes that a Machine costs "the price of a named CPU/RAM preset, plus about $5 per 30 days per GB of additional RAM." OpenClaw's own estimate for the recommended configuration is "roughly $10-15/month", which lines up.

Four fly.toml settings carry the deployment. OPENCLAW_STATE_DIR = /data puts state on the volume; --bind lan lets the Fly proxy reach the gateway; --allow-unconfigured lets it start before a config file exists; and auto_stop_machines = false keeps it awake, which matters a great deal for a process whose whole job is being always-on. The docs also set internal_port = 3000 and note it must match the gateway port for health checks, so keep those two in step.

#A Linux box you already own

The honest fourth option: the machine in the cupboard. An old laptop, a NUC, a Raspberry Pi. OpenClaw's Raspberry Pi page grades the hardware for you. A Pi 5 with 4 or 8 GB is "Best", a Pi 4 with 4 GB is the "sweet spot", a Pi 4 with 2 GB is "OK" with swap, and the Pi Zero 2 W is "Not recommended."

Two rules from that page apply to any small Linux box. Use a 64-bit OS, because the docs say plainly "do not use 32-bit". And put the state on real storage: the docs recommend "Use a USB SSD" to avoid SD card slowness and wear. Add swap on anything with 2 GB or less.

Our Pi-specific guide is OpenClaw on a Raspberry Pi.

#How do you reach the gateway once it is remote?

The same way on all four routes, because the gateway defaults to loopback everywhere except inside a container.

The documented pattern on OpenClaw's DigitalOcean page is an SSH tunnel as the simplest method, or Tailscale Serve when you want several devices reaching it over a trusted network. Neither one requires opening 18789 in a cloud firewall, which is the whole appeal.

Once you are in, the operator commands are the same set you would use on a Mac mini:

openclaw gateway status
openclaw gateway status --deep
openclaw gateway restart
openclaw logs --follow

The latest release listed in the OpenClaw docs as of September 19, 2026 is v2026.9.5.

#What to read next

• OpenClaw on Linux (Ubuntu/Debian) is the full server build, with systemd and firewall configuration.
• OpenClaw on Oracle Cloud Free Tier walks the free Arm instance end to end, including the capacity dance.
• OpenClaw on a VPS is the provider-agnostic version of the DigitalOcean section above.
• OpenClaw + Docker covers the container path with volumes and health checks.
• Running OpenClaw on a Mac mini is the other side of this decision, if a box at home still appeals.