Your own AI assistant, always on.

It is the same self-managed KVM line, with full root. The difference is the setup: OpenClaw arrives installed with its gateway registered as a service, on hardware sized for it, so you skip the install and go straight to adding your key and your channels.

O1 Starter already meets OpenClaw’s own minimum of 2 vCPU and 4 GB of RAM, so even the smallest plan gives the gateway the room the project asks for. Start there, and move up to O2 Everyday or O3 Multi when you give the assistant more to do.

Not to install it. Ordering provisions a VPS with OpenClaw pre-installed and its gateway registered as a service, so it comes back up on its own after a reboot. Claiming it is a one-time step over SSH: read your gateway token and approve your device. Connecting your channels and your model key follows OpenClaw’s own onboarding, and an engineer can point you right if you get stuck.

The base plans are a real KVM VPS with full root that you run: OS updates, OpenClaw updates and housekeeping are in your hands. Hardware, network and the virtualisation layer are ours. If you want our engineers on the box (updates, monitoring, hardening), add a paid Server Management tier at checkout.

Yes, at VM level: we take a daily full-disk backup of your instance on a best-effort basis, and it is what we use to bring the whole server back if something goes badly wrong. Restoring an individual file or database is yours to do on a self-managed box, so keep your own copies of anything you cannot lose. The “Offsite Backups for Data” option at checkout adds offsite copies. And since OpenClaw keeps its memory as plain files, your own exports are easy.

Your instance is single-tenant: your own box, your own gateway, in the datacenter region you choose at checkout (India, Germany or the US), and OpenClaw keeps its memory as plain files you can read yourself, so nothing about it is opaque to you. Your keys are yours, and we do not read your conversations. It is worth knowing exactly where that line sits: we operate the host, so the disk is technically reachable by us, and OpenClaw’s own documentation says resisting a compromised host is not one of its design goals. So what you get is a machine nobody else shares, plus a policy we hold ourselves to. If your threat model needs more than that, talk to us before you order and we will tell you whether this fits.

Treat it the way the project itself does: OpenClaw’s docs say to treat inbound messages as untrusted input, and by default the agent can act on the machine it runs on. That is what makes it useful and it is also the risk. Give it its own box (you are), be deliberate about which channels and which skills you connect, and ask us if you are unsure about one. We would rather have that conversation than have you find out later.

It is the full hosting price for the term you choose. Promotional discounts apply to your first term. If your billing address is in India, 18% GST is added at checkout.

Both are open-source AI assistants that live on a server of your own and answer through your chat apps, on the same self-managed KVM line. OpenClaw is built around one gateway for around two dozen chat channels, with voice, a live canvas and companion apps. Hermes, from Nous Research, turns what it works through into skills it writes and refines itself, and connects to more than forty model providers. If you are unsure which suits what you have in mind, ask an engineer.