Released · 2026-10-19 · OpenClaw on a cloud server

OpenClaw on a Cloud Server: Run the Gateway

As of 2026-10-19, yes: the official docs say the Gateway process can run on a server; put it on a QCode cloud server, set it up as a systemd service and enable lingering, and it keeps running after you log out.

Updated 2026-10-19

Point this tool at the QCode endpoint — one key for Claude, GPT and Chinese models.

Billed per token — live rates on /models
Pay by card (Visa / Mastercard / AMEX), Apple Pay, Google Pay or crypto
Self-service: top up and activate instantly

Create account View pricing

#Dedicated public IPv4#Gateway as a background service#1-day trial#1, 3, 6 or 12 months

Four points: address, access, trial, term

A fixed public address

One per server, assigned at setup

Includes 1 dedicated public IPv4 and KVM. It is assigned when the server is ready and it cannot be changed.

You run the system

Full root access

Log in with the root password you set when ordering; you administer the system. Once the Gateway is set up as a systemd user unit and lingering is enabled (sudo loginctl enable-linger $(whoami)), it keeps running after you log out. In the dashboard you can check status, start, stop, reboot and renew, and the server image does not include a hosting panel.

1-day trial

Try it, then keep it (price by size)

A trial lasts 1 day (24 hours), and the price depends on the size. When it ends, the server is deleted and the data is not kept. During the trial you can keep it from the panel by picking 1, 3, 6 or 12 months, and each account can have only one trial server at a time.

1, 3, 6 or 12 months

Term and automatic renewal

You can renew 1, 3, 6 or 12 months by hand in the panel at any time. With automatic renewal on, one month is renewed from your balance 3 days before expiry; we remind you early if the balance is short, and if it cannot cover the renewal the server is not renewed.

Can OpenClaw run on a cloud server

As of 2026-10-19, yes. On a server the Gateway owns the state and the workspace on that machine, and you connect to it from your laptop or phone through the control UI or Tailscale/SSH, while the chat apps you already use are the way in. Model calls go through QCode's endpoint, and cloud servers and AI services are sold separately, so you do not have to buy an AI plan first.

The official docs and the tool, checked on 2026-10-19

As of 2026-10-19, the OpenClaw docs say it is an open-source AI assistant that runs on your own hardware and meets you in every chat app you already use. The install page gives the command curl -fsSL https://openclaw.ai/install.sh | bash; the installer detects your OS, installs Node if needed, installs OpenClaw, and then starts onboarding. QCode's OpenClaw guide says a self-hosted Gateway process connects Discord, Telegram, Slack, iMessage and other chat apps, the configuration lives in ~/.openclaw/openclaw.json (JSON5, hot-reloaded by the gateway) or is set through the openclaw onboard wizard, and a QCode key starts with cr_, so no separate account with a model vendor is needed.

Three dates

2026-09-11

OpenClaw 2026.9.4 is released; this is the version QCode's OpenClaw guide is based on.

2026-09-22

Anthropic released the new flagship claude-opus-5-5 on 2026-09-22, and the QCode model page lists it as available.

2026-09-29

gpt-6.1-sol is released, the GPT-6 Sol update from OpenAI dated 2026-09-29; the QCode models page lists it as newly available.

What is confirmed and what is still open

Confirmed points

First, provisioning and trial: the server is usually ready within a few minutes; the trial lasts 1 day (24 hours), the price depends on the size, the server is deleted when the trial ends and the data is not kept. Second, renewal: with automatic renewal on, one month is renewed from your balance 3 days before expiry, we remind you early if the balance is short, and if it cannot cover the renewal the server is not renewed. Third, the runtime: the official docs ask for Node.js 24.16+ or 26.1+ (Node 26 is recommended), and they also say that running the Gateway and its agent commands as root is unsafe and unsupported, so you create a regular user. Fourth, the full refund and forbidden-use terms are on qcode.cc/en/cloud.

What the official pages do not state

Neither the cloud server page nor the OpenClaw docs state latency in milliseconds or bandwidth numbers, and none of them says how long the machine can keep running; the QCode cloud server page says nothing about snapshots or backups, and the OpenClaw docs do not set out how much CPU and memory the server needs, so the configuration is picked from the line comparison.

On your own machine or on a cloud server

OpenClaw on your own machine

The configuration travels with you: ~/.openclaw/openclaw.json and the key sit on your own computer. It suits an occasional try-out; switch the machine off and the Gateway stops.

The Gateway on a cloud server

The machine is yours, and the Gateway process, set up as a background service, runs on that machine, which suits long jobs and background work; in the dashboard you can check status, start, stop, reboot and renew.

Getting OpenClaw running on the server

1) On qcode.cc/en/cloud pick a region, a line, a term and a size; starting prices are in the line comparison. 2) Order it, then log in with the connection details from the dashboard: ssh root@ followed by your server IP, with the root password you set when ordering. 3) The official docs say running the Gateway and its agent commands as root is unsafe and unsupported, so create a regular user while you are logged in as root: on Debian/Ubuntu run adduser openclaw and then add that user to the sudo group (usermod -aG sudo openclaw; usermod's -a/--append only works together with -G/--groups). Run the steps below as that user. 4) As that user run the official install script curl -fsSL https://openclaw.ai/install.sh | bash; it detects your OS, installs Node if needed, installs OpenClaw and starts onboarding. 5) Write the QCode provider into ~/.openclaw/openclaw.json: set mode "merge" in the models section, which keeps the built-in providers and appends QCode, and put the provider into models.providers next to it; every provider needs a models array with the model ids and names; Claude models use baseUrl https://api.qcode.cc/api with api set to anthropic-messages, and GPT and the models in the Chinese catalog use baseUrl https://api.qcode.cc/openai/v1 with api set to openai-completions, while apiKey references ${QCODE_API_KEY}. 6) Or run openclaw onboard --install-daemon, which also sets up the background service, pick Custom Provider in the wizard and fill in the base URL, API key, compatibility and the model ID. 7) Whether you take step 5 or step 6, the Gateway is set up as a systemd user unit; if you wrote the configuration by hand, run openclaw gateway install as well. To keep it running after you log out you have to enable lingering in either case (sudo loginctl enable-linger $(whoami)), and on a server without a desktop session set XDG_RUNTIME_DIR first (export XDG_RUNTIME_DIR=/run/user/$(id -u)). 8) Check it with openclaw gateway status. The Gateway listens on the loopback address by default, the docs suggest Tailscale/VPN, and the fallback is an SSH tunnel (for example ssh -N -L 18789:127.0.0.1:18789 user@gateway-host): map the port to your own computer and open the control UI in the browser there, a bind beyond loopback needs auth, and you connect one chat platform first if you want to message the Gateway from a chat app.

How to connect it on QCode

On the server, OpenClaw talks to QCode's endpoint: the provider goes into the models.providers section of ~/.openclaw/openclaw.json, and step 5 above shows how to write it (set mode "merge" in the models section, which keeps the built-in providers and appends QCode). You can also run openclaw onboard --install-daemon and use the Custom Provider wizard; there the compatibility field takes anthropic for Claude models, while the api field in the configuration file takes anthropic-messages, and the two spellings differ. The QCode docs note that the https://api.qcode.cc/api form is untested; if a request returns 404, change baseUrl to the full path https://api.qcode.cc/api/v1/messages and try again. apiKey references ${QCODE_API_KEY}, and the official docs recommend a reference instead of a plain key. The models available through QCode are claude-sonnet-5, claude-opus-5-5 (the Claude models, on the endpoint with the Anthropic protocol) and gpt-6.1-sol, glm-5.3 (on the endpoint with the OpenAI protocol). Calls are billed per token and each model's price is listed on /models. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 to 90 days (top-ups start at $1).

Frequently asked questions

Can OpenClaw run on a cloud server?

Yes, as long as the Gateway is set up as a background service (see the next question). On a server the Gateway owns the state and the workspace on that machine, and you connect to it from your laptop or phone through the control UI or Tailscale/SSH; model calls go through QCode's endpoint, and AI service and cloud servers share one account balance.

How do I pay and how do I order?

You can pay with Alipay, WeChat, bank cards and cryptocurrency; cloud servers and AI services share one account balance, and you do not have to buy an AI plan first, because they are sold separately and share only the account and the balance. The cloud server console is available in Chinese and English. After you pick a region, a line, a term and a size, the server is usually ready within a few minutes; if provisioning fails or the server is not delivered within 2 hours of the order, the full amount goes back to your balance.

How do I configure the models for OpenClaw on the server?

See steps 5 and 6 above: the point is that every provider needs a models array with the model ids and names. The file is JSON5, and the gateway reloads it automatically after you save it, so no restart is needed.

How is the trial counted, and can I get a refund after the server is provisioned?

A trial lasts 1 day (24 hours), and the price depends on the size. When it ends, the server is deleted and the data is not kept; during the trial you can keep it from the panel by picking 1, 3, 6 or 12 months, and each account can have only one trial server at a time. No refund is possible after the server is provisioned; if provisioning fails or the server is not delivered within 2 hours, the full amount goes back to your balance, and if it is shut down for a rule violation the paid amount is not refunded.

How much configuration does OpenClaw need on a cloud server?

The official docs require Node.js 24.16+ or 26.1+ (Node 26 is the recommended runtime), and the installer puts Node in place if it is missing. The docs do not state how much CPU and memory the server needs, and the QCode cloud server page lists no configuration numbers either, so the configuration is picked from the line comparison. You cannot change the line or upgrade the size online at the moment, so order another server if you need a different one.

How does the Gateway keep running after the SSH session ends?

See step 7 above: the Gateway is set up as a systemd user unit with lingering enabled, so it keeps running after you log out, and on a server without a desktop session you set XDG_RUNTIME_DIR first. Check the state with openclaw gateway status.

Sources

The QCode cloud server page (qcode.cc/cloud, fetched and checked on 2026-10-19): regions and lines, IPv4 and root access, trial and terms, the refund rule, automatic renewal, dashboard features, forbidden uses, payment methods and how to connect. The OpenClaw guide in the QCode docs (fetched and checked on 2026-10-19): where the configuration lives, the two endpoint forms, the adapter values, the key reference and the onboard wizard. The OpenClaw official docs: home page, install page, getting-started page and the Gateway runbook (fetched and checked on 2026-10-19): the positioning sentence, the install command, the background service, the lingering step, the default loopback bind and the remote access options. The QCode packs page and the models page (fetched and checked on 2026-10-19): how long a pack is valid, the remaining quota is forfeited, the refund rule after the purchase and the model release dates.

Get OpenClaw running on a cloud server

A machine you administer, with a 1-day trial; OpenClaw calls claude-sonnet-5, claude-opus-5-5, gpt-6.1-sol and other models through QCode's endpoint, billed per token, and each model's price is listed on /models.

Create an account

Models and live rates View plans