skip to content

Hosting Choices

Laptop, rented VPS or a dedicated box such as a Raspberry Pi or Mac mini: where the Gateway lives decides uptime, which channels work and how exposed it is. A common judgment question.

part ofAI coding toolsread it →
on this pageshow

explore

questions

6

Why is a personal laptop a weak always-on host for an OpenClaw Gateway, and what setup do the docs suggest instead?

level: juniorimportance: must knowfreq 18%

answer

  1. the lid closes
  2. one process holds every channel
  3. scheduled work needs a running host
  4. dedicated host, laptop as companion

basics

~20 s

An OpenClaw Gateway works only while its host is awake and online. A laptop sleeps, changes networks and reboots for updates, so chats and scheduled jobs stall; run the Gateway on an always-on host and pair the laptop as a node.

solid answer

~40 s

Everything in OpenClaw runs inside one Gateway process on one host: the channel connections, the agents and the scheduler behind heartbeats and cron. A laptop sleeps when the lid closes, drops connections when it changes networks and reboots for updates, so the bot goes quiet and a missed cron slot runs late once the machine is back, because cron catches up overdue work by default. A laptop is fine for trying it and for visible-browser work. For dependable use the docs suggest a dedicated always-on host (a small VPS, a Raspberry Pi or a Mac mini) with the laptop paired as a **node**, which adds `screen`, `camera` and `system.run` on that device without running a second Gateway.

go deeper

for a junior

Recall that the Gateway is one process on one host and that the assistant is only up while that host is awake and online.

for a middle

Explain what sleep does to channels and to cron, including that missed slots catch up late by default, and how a node differs from a second Gateway.

for a senior

Recommend a host from observed failures: disconnect history, scheduled work that must run on time, browser needs, and the cost of keeping a laptop awake.

for a principal

Frame the move as an availability decision: what the assistant is trusted to do unattended decides whether its host may ever sleep.

## What the Gateway does while you are not looking OpenClaw runs as one long-lived process, the **Gateway**, on one host. It holds the live connections to every chat channel (a WhatsApp linked device, a Telegram bot, a Slack or Discord app), runs the agents that answer them, and owns the scheduler behind heartbeats and cron jobs. Its config is `~/.openclaw/openclaw.json`, and its state and default workspace live under `~/.openclaw` on that same machine. The phone app, the browser dashboard and the CLI are only clients of that process. So the host's uptime **is** the assistant's uptime. While the host is asleep, nothing in OpenClaw runs: no channel is answered, no heartbeat fires, no scheduled report is written. ## Why a laptop is the weak spot A laptop is a fine place to try OpenClaw and a poor place to depend on it. The project's own hosting FAQ lists the costs: - **Sleep.** Closing the lid or idling suspends the Gateway; channels disconnect and replies wait until the machine wakes. - **Network changes.** Moving between home, office and a hotspot drops connections and changes the address remote clients use. - **Updates and reboots.** Each one interrupts the Gateway. A managed service (a LaunchAgent on macOS, a systemd user unit on Linux) brings it back afterwards, but no service can keep a sleeping machine awake. - **Shared use.** It is also your daily computer, so the assistant inherits whatever accounts and files you are signed into. Scheduled work shows the cost most clearly. By default, a recurring cron slot missed while the Gateway was down is caught up once it is running again (`cron.skipMissedJobs` defaults to `false`), and a sleeping Gateway that wakes past a deadline runs the overdue job then. A 7 a.m. briefing therefore arrives whenever you open the lid: late rather than lost, which for a reminder can be just as bad. What a laptop does well is real too: no server bill, direct access to your local files, and a visible browser window for UI automation. The FAQ's default recommendation follows from both lists: run it locally while you actively use the machine; move to a VPS or another always-on host once you have had Gateway disconnects. ## The usual move: an always-on host | Host | Always on | Typical trade-off | | --- | --- | --- | | Laptop | No | Free, local files, visible browser; sleeps and roams | | Small Linux VPS | Yes | Monthly fee, headless, a data-center IP some sites block | | Raspberry Pi | Yes | Cheap one-off hardware; 64-bit OS, cloud models only | | Mac mini | Yes | Higher upfront cost; the one box here that can also host iMessage | The Gateway itself is light, because models run at the providers. For a VPS the FAQ gives an absolute minimum of 1 vCPU, 1 GB RAM and about 500 MB of disk, with 2 GB or more recommended for logs, media and several channels. ## Keep the laptop, as a node The documented best-of-both pattern is to keep the Gateway on the dedicated host and pair the laptop as a **node**: 1. Run the Gateway on the always-on host. 2. Install the macOS app or a node host on the laptop and pair it to that Gateway. 3. Check it with `openclaw nodes status`. A node is a companion device, not a second Gateway. It connects out to the Gateway's WebSocket through device pairing, so the laptop needs no inbound SSH, and it adds device capabilities: `screen`, `camera` and `system.run`, with `system.run` gated by allowlists and approvals on the laptop itself. When the laptop sleeps, those device tools are unavailable for a while; the assistant keeps answering chats from the always-on host. ## Mistakes that show up in practice - Treating the laptop as the server and blaming the model when answers stop overnight. - Running a second Gateway on the laptop to "sync" with the home one. Gateways do not mirror each other's sessions; a companion device joins the always-on Gateway as a node. - Moving to a VPS and publishing the Gateway port to reach it. The documented secure default keeps the Gateway on loopback and reaches the dashboard through an SSH tunnel or Tailscale Serve; the exposure details belong with Gateway hardening. - Buying a large machine for the Gateway. Model calls go to hosted providers, so a modest always-on box is enough unless you plan to run local models.

  • If the laptop has to stay the host for now, what reduces the damage?
    Keep it plugged in with sleep disabled, and install the managed service with `openclaw onboard --install-daemon` so the Gateway returns after reboots; `openclaw gateway status` confirms it is up. Expect overdue cron jobs to run late after downtime, and set `cron.skipMissedJobs: true` if stale recurring reminders at Gateway startup would be worse than skipped ones.
  • What does moving the Gateway from a laptop to a VPS give up?
    The FAQ lists it: a VPS is usually headless, so browser work happens in a headless browser; local files are only reachable remotely; updates need SSH; and some sites block data-center IPs. Pairing the laptop as a node gets device tools and a local browser back while the Gateway stays on the always-on host.

A laptop-hosted Gateway is a shop whose only door is in your house: when you go to bed the shop closes, and the orders that came in overnight are dealt with when you get up.

  • A laptop is an always-on host as long as the Gateway runs as a service.
  • Cron jobs missed while the laptop slept are dropped by default.
  • Pairing the laptop as a node turns it into a backup Gateway.
  • A VPS Gateway must publish its port so the phone app can reach it.
  • The Gateway needs a powerful machine because the model runs on it.
open as a page

When you run OpenClaw from its official container image with Docker Compose, what has to live outside the container, and how do image upgrades work?

level: middleimportance: should knowfreq 7%

basics

~20 s

The state directory mounted at /home/node/.openclaw (config, SQLite auth and session stores, .env) and the workspace must live on the host. Upgrades pull a new ghcr.io/openclaw/openclaw tag; the entrypoint runs openclaw doctor --fix before the Gateway starts.

open as a page

Can an OpenClaw Gateway on a Linux VPS offer iMessage, and what does OpenClaw's iMessage channel need from a Mac?

level: middleimportance: should knowfreq 9%

basics

~20 s

Yes, but a Mac is still required. OpenClaw's iMessage channel drives the imsg tool on a Mac signed into Messages; a Linux Gateway reaches it by pointing channels.imessage.cliPath at an SSH wrapper that runs imsg on that Mac.

open as a page

How do you back up an OpenClaw host with openclaw backup and move it to a new machine, and what does a restore not bring back cleanly?

level: seniorimportance: should knowfreq 6%

basics

~20 s

Stop the Gateway, run openclaw backup create --verify, restore the archive on the new host into an empty staging directory, move it into place offline and run openclaw doctor. Ratcheting channel logins such as WhatsApp, approvals and plugin packages may need attention.

open as a page

For a household's always-on OpenClaw assistant, how would you choose between a rented VPS, a Raspberry Pi and a Mac mini, and what decides it?

level: principalimportance: should knowfreq 12%

basics

~20 s

Start from hard constraints: iMessage needs a Mac signed into Messages, and local models rule out a Pi. Then weigh a VPS's monthly cost and public exposure against home hardware and power, with one Gateway per group that trusts each other.

open as a page

Which Node.js versions does OpenClaw 2026.9.9 support, and why did release 2026.9.3 drop Node 22 and Node 25?

level: seniorimportance: nice to knowfreq 5%

basics

~10 s

OpenClaw 2026.9.9 needs Node 24.16+ or Node 26.1+, with Node 26 recommended. Release 2026.9.3 dropped Node 22 and 25 because their node:sqlite TEXT decoder silently truncates values at embedded NUL characters.

open as a page