Best hosting for OpenClaw and Hermes Agent
When to run OpenClaw or Hermes on an Orgo template, and when a VPS is enough.
TL;DR
A good home for OpenClaw or Hermes Agent stays on, keeps the messaging gateway running, and gives the agent a desktop it can use. Orgo's curated templates set that up in one launch: system/openclaw@1.1.8 for OpenClaw and system/hermes-agent@1.1.13 for Hermes, each on 2 vCPU and 8 GB RAM (OpenClaw guide, Hermes guide). Any paid plan can launch them, and a free account can run one of them by the hour on its starting credit.
Pick a template over a general VPS when you want the desktop and the agent without maintaining the OS, VNC, browser and gateway restarts yourself. Keep a VPS when the job is a headless backend with no desktop or messaging gateway (Migrate to Orgo).
What does "hosting" OpenClaw or Hermes mean?
OpenClaw is an open-source AI assistant that controls applications, automates tasks, and talks to you through a messaging gateway. Hermes Agent is Nous Research's self-improving agent: it builds skills from experience and keeps a persistent model of the user. Neither is made by Orgo.
Orgo provides the computer those agents run on: a full desktop with a browser and the standard Linux userland, controlled over HTTP and viewable over VNC. A computer stays on until you stop it. An hourly computer also stops when its credit runs out.
With a curated template, you launch an Orgo computer where the agent is already installed and its gateway is supervised: it restarts on a crash and on every boot. You give it a model and connect your messaging apps. You do not write service files, set up VNC, or rebuild the desktop on each machine.
Why pick a curated Orgo template over a general VPS?
A general VPS is a blank Linux server. You own the OS, firewall, desktop, VNC, browser, fonts, TLS, monitoring and backups, and turning it into a desktop an agent can operate is your job. An Orgo template arrives with the desktop, the agent and the control API (click, type, screenshot, shell) already in place.
| General VPS | Orgo OpenClaw or Hermes template | |
|---|---|---|
| What you set up | OS, firewall, desktop, VNC, browser, fonts, TLS, monitoring, backups | A model and your messaging apps |
| Getting started | Install the agent and a desktop stack yourself | One create from the dashboard or the API |
| Click, type, screenshot | Build it yourself | Part of the Orgo API |
| Agent and gateway | You install and supervise them | Preinstalled; the gateway restarts on a crash and on every boot |
| systemd | Usually available | Not available; the template supervises the gateway instead |
| Inbound webhooks | You can expose a public hostname | No public inbound hostname; run gateways in outbound mode |
| Best for | Headless backends you fully own | An always-on agent with a desktop and messaging |
OpenClaw vs Hermes on Orgo
Both templates are supervised, run on 2 vCPU and 8 GB RAM, and launch the same way. The difference is the agent.
| OpenClaw template | Hermes Agent template | |
|---|---|---|
| Template ref | system/openclaw@1.1.8 | system/hermes-agent@1.1.13 |
| What ships | OpenClaw and the Claude Code CLI, gateway supervised | Hermes Agent, gateway supervised |
| Model | Dashboard chat uses ANTHROPIC_API_KEY from your workspace's Secrets, or run openclaw onboard in the terminal | Dashboard chat uses ANTHROPIC_API_KEY from Secrets, or Orgo's models billed to your AI credits when there is no key; in the terminal, hermes setup (hermes setup --portal signs in with Nous Portal) |
| Messaging | Discord, Matrix, Microsoft Teams, Signal, Slack, Telegram, WhatsApp and more | Telegram, Discord, Slack, WhatsApp, Signal, email and more, through hermes gateway setup |
| State | ~/.openclaw/, kept across stops and restarts | ~/.hermes/, kept across stops and restarts |
| Extras | Skills from ClawHub | hermes claw migrate imports an existing OpenClaw setup |
Those refs are the current versions as of October 2026. List curated templates returns the published curated templates and their refs.
How do I launch OpenClaw on Orgo?
-
Use any paid plan, or a free account by the hour on its starting credit (Pricing).
-
Create the computer. In the dashboard, create a computer and choose OpenClaw. From the API, pass the template ref:
curl -X POST https://www.orgo.ai/api/computers \ -H "Authorization: Bearer $ORGO_API_KEY" \ -H "Content-Type: application/json" \ -d '{"workspace_id": "'"$WORKSPACE_ID"'", "name": "openclaw", "template_ref": "system/openclaw@1.1.8"}'On a free account, add
"billing_mode": "hourly"to the body to run it on your credit, and send anIdempotency-Keyheader with a UUID so a retry launches it once (Create computer). -
Give it a model. Set
ANTHROPIC_API_KEYin Settings → Secrets and chat with OpenClaw from the dashboard, or open the computer's terminal, where the template startsopenclaw onboard. The gateway starts once OpenClaw has a config. -
Connect messaging in outbound mode, such as Slack Socket Mode or Telegram long polling. Webhooks cannot reach the gateway because the computer has no public inbound hostname.
-
Prefer the template to a manual install. A manual install needs Node 24.16+ or 26.1+, and without systemd its gateway runs in a detached
tmuxsession that does not survive a restart.
How do I launch Hermes Agent on Orgo?
- Use any paid plan, or a free account by the hour on its starting credit (Pricing).
- Create the computer. In the dashboard, choose Hermes Agent. From the API, send the same request as above with
"template_ref": "system/hermes-agent@1.1.13". - Give it a model. With
ANTHROPIC_API_KEYin Settings → Secrets, dashboard chat uses your key. Without one, it uses Orgo's models billed to your AI credits. In the terminal, runhermes setup, orhermes setup --portalto sign in with a Nous Portal account. - Connect messaging with
hermes gateway setup. The template's gateway reads its settings when it starts, so restart it afterwards withsupervisorctl restart hermes-gateway. From then on, the template keeps it running. - When you serve many people, give each user their own Hermes computer, as the docs recommend. Identity, memory and skills live in
~/.hermes/and survive stops.
How do I migrate an existing agent from a VPS?
Migrate to Orgo has the full commands for DigitalOcean, Fly.io and Hetzner. For a worked example from one provider, see moving from Hetzner to Orgo. The shape is the same for both agents:
- Stop the gateway on the VPS and archive the agent's state directory:
~/.openclaw/for OpenClaw,~/.hermes/for Hermes. That one directory holds identity, memory, skills and API keys. - Create an Orgo computer and copy the archive onto it. The file upload takes files up to 10 MB, and the two-step large upload takes files up to 2 GB (Upload file). You can also pull the archive from inside the computer, for example from a private repository or a temporary URL.
- Install the agent with its upstream installer, then restore the state over the fresh install. For Hermes, do not run
hermes setupagain afterwards, because it rewrites.envand can overwrite the keys you brought over. - Start the gateway in a detached
tmuxsession (Hermes needshermes gateway run --no-supervise). Atmuxsession does not come back after a restart; the curated templates are the way to get a supervised gateway. - Re-pair device-bound channels. WhatsApp and Signal pairings do not carry across machines, so scan the QR code again. OpenClaw's iMessage channel needs a Mac signed in to Messages, so it does not move onto a Linux computer on its own.
- Switch messaging to outbound mode. There is no public per-computer hostname for inbound webhooks.
- Keep the source gateway stopped while you verify. Two gateways on the same bot token race for incoming messages. Delete the source once the agent behaves on Orgo.
Moving from OpenClaw to Hermes at the same time? hermes claw migrate imports an OpenClaw setup into Hermes.
Stay on the VPS when the workload is a headless backend script with no desktop or gateway. The Orgo docs say a VPS serves that case equally well.
What are the hard limits?
- No systemd. Supervised gateways come from the curated templates. Manual installs run the gateway in
tmux, which does not survive a restart. - Outbound gateways only. There is no public inbound hostname for webhooks.
- Device-bound messaging. WhatsApp and Signal need a fresh QR scan after a move.
- Isolation. The agent reaches only what you put on the computer. Your laptop's files, browser sessions and keys stay out of reach.
- The computer, not the agent. Orgo provides the computer. You bring OpenClaw, Hermes, or your own agent loop.
- More than a browser. If you only need a headless browser session, that is a different kind of product. See cloud desktops vs. headless browsers.
FAQ
Do I need systemd for the gateway on Orgo?
No. Orgo computers do not run systemd. Launch the OpenClaw or Hermes template so the gateway is supervised, or run it in a detached tmux session on a manual install, knowing it will not survive a restart.
Can Slack or Telegram webhooks reach my gateway? No. The computer has no public inbound hostname. Use outbound modes such as Slack Socket Mode or Telegram long polling.
Will WhatsApp or Signal keep working after I migrate? Not until you pair again. These pairings are tied to the device, so scan the QR code on the new computer.
Which plans can launch these templates? Any paid plan can launch both. A free account can run one of them by the hour on its starting credit. When the credit runs out, the computer stops, and it is deleted 7 days later unless you add credit and start it again or choose a plan. See Pricing for current plans.
Can I run OpenClaw and Hermes on the same computer?
Each template ships one agent. You could install the other by hand, but the documented manual install runs its gateway in tmux, without supervision. Separate computers keep both gateways supervised. Give each agent its own bot token, because two gateways on one token race for messages.
Does agent state survive stop and start?
Yes. ~/.openclaw/ and ~/.hermes/ are on the computer's disk and survive stops and restarts. Deleting the computer deletes them.
What about Nous Research's Hermes Cloud? As of October 2026, Nous Research also sells Hermes Cloud, a hosted Hermes Agent with its own container per agent and a web dashboard. Its page lists Telegram, Discord, Slack, email and CLI channels. It suits you if you want Nous to run the agent and those channels cover your needs. An Orgo computer suits an agent that needs a full desktop and browser to operate, or a channel that page does not list, such as WhatsApp or Signal.
When should I keep a VPS instead? When the job is a headless backend with no need for a desktop or a messaging gateway.