OSIR · The AI-Native Domain Registrar

AI & Domains

Give Your Agent a Home: A VPS and a Domain of Its Own

2026-08-13 · OSIR Team

Give Your Agent a Home: A VPS and a Domain of Its Own

Summary: The most useful thing about agents like Hermes Agent and OpenClaw is that they keep working when you close the laptop: watching domain expiry, checking DNS after a migration, reporting in every morning on Telegram. That requires a home - a small always-on server - and a name, because an agent you will connect services to should live at something like agent.yourdomain.com, not at a bare IP that changes. This guide covers both halves: the server and the name.

Part 5, the final part of our Agents & Domains miniseries. Previously: Your domain is your agent's name.

Why your laptop is the wrong home

OpenClaw's default mode - running on your own computer - is perfect for trying an agent out. It is the wrong place for the always-on work that makes a personal agent worth having:

  • Scheduled jobs die with the lid. An expiry watchdog that only runs while your laptop is awake is not a watchdog.
  • Messaging needs presence. The magic of texting your agent at midnight only works if something is listening at midnight.
  • Memory wants a fixed address. Hermes Agent is explicitly designed to run continuously on a server, accumulating cross-session memory and self-written skills; its own documentation assumes a modest always-on Linux box - a small VPS is enough to start.

A VPS (virtual private server) gives you that: a machine that is always on, always connected, and entirely yours. Ours start small and scale up; current packages and prices are on the pricing page, and VPS details are here.

Why the agent gets a domain, not just an IP

You could run an agent on a bare IP address. You should not, for reasons that compound over time:

  1. Stability. IPs change when you rebuild, resize, or migrate the server. A name (agent.yourdomain.com) survives all of that with one DNS edit - and your integrations, webhooks, and bookmarks keep working.
  2. TLS. Anything your agent exposes (a status page, a webhook receiver) wants HTTPS, and certificates are issued to names, not to bare IPs in practice.
  3. Identity. As part 4 covered, every emerging agent-discovery standard - agent cards, DNS-based discovery, signed requests - anchors to a domain. Giving your agent a subdomain today means it is one TXT record or well-known file away from participating in whatever standard wins.
  4. Email and notifications. An agent that sends you reports benefits from mail authentication (SPF/DKIM) that only a domain can carry.

The setup is one DNS record: an A record pointing agent.yourdomain.com at your VPS's IP. If your domain is with us, that is a ten-second edit in the dashboard - or, in the spirit of this series, ask your assistant to do it over the MCP connector.

The setup, end to end

Here is the shape of the full setup. Steps 1 and 2 can be done in a single AI conversation - we demonstrated exactly that flow (order, build, DNS) in an earlier walkthrough.

  1. Order and build a small VPS. Pick an entry-level package with a mainstream Linux template (Ubuntu LTS or Debian). Add your SSH key at build time rather than using passwords. Via MCP this is the orderVps + buildVpsInstance pair, both behind the usual confirmation gate since ordering costs money.
  2. Point the subdomain. Create the A record: agent.yourdomain.com to the VPS IP. Propagation for a new record is typically minutes (our DNS guide explains why).
  3. Install your agent. Follow the official install guide for your agent - Hermes Agent and OpenClaw both document server installation, and both are moving fast enough that copying commands from a blog post (including this one) would be a disservice. The install is a normal Linux affair: runtime, agent, API key for your model provider.
  4. Connect messaging. Both agents pair with Telegram, WhatsApp, Discord, or Signal from their configuration. This is your daily interface; the server fades into the background.
  5. Connect its tools. Point the agent at the MCP servers it should use - for domains, ours - and authenticate with the device-login flow described in part 2.

Hardening: the boring part that matters most

An always-on agent with tool access is a high-value target, so treat its home accordingly:

  • SSH keys only, no password login. Disable root password auth the day the server is built.
  • Expose nothing inbound by default. A personal agent mostly makes outbound connections (to messaging APIs, model APIs, MCP servers). If nothing needs to reach it, keep the firewall closed and skip the public status page - no inbound surface, no inbound attacks.
  • Scope its credentials. Give the agent its own API keys, its own registrar session (revocable in one click), and a spending balance you would be comfortable losing. Never hand it a password vault.
  • Watch what it installs. Skills and integrations from community registries are the ecosystem's known weak spot - malicious skills have been found in the wild. Install sparingly, from sources you trust, and read the skill before the agent runs it.
  • Keep the audit trail warm. Ask the agent weekly what it has done with its tools; on our connector getMyAuditLogs makes that a one-line request. An agent reporting on its own actions is a surprisingly effective tripwire.

What you end up with

For the cost of a small VPS and a subdomain on a domain you already own: an assistant that never sleeps, remembers everything, reports to your phone, and holds exactly the keys you gave it - watching your domains, your DNS, and whatever else you wire in. The full loop this series described, running on infrastructure that is yours: the agent has a name, the name is on your domain, and the domain is under your control. That ordering is the point.

Agent hosting FAQ

How big a VPS does an AI agent need?

Small. Personal agents like Hermes Agent are designed to run on entry-level servers - the model inference happens at your AI provider's API, not on your box, so the VPS mostly runs the agent loop, memory, and messaging connections. Start with the smallest package on our pricing page and resize later if needed; rebuilding is free, and the domain name means nothing else has to change.

Should the agent live on a subdomain or its own domain?

A subdomain of a domain you already own (agent.yourdomain.com) is right for a personal agent: zero extra cost, inherits your domain's identity, one A record to set up. A separate domain makes sense when the agent is a public-facing product with its own brand.

Can the agent set up its own home?

Mostly, yes - and we find this genuinely fun: an MCP-connected assistant can order the VPS, build it with an OS and your SSH key, and create the DNS record in one conversation (each paid step behind a confirmation). The agent software install on the server is the one part you still do by hand.

Hermes Agent or OpenClaw for a server install?

Hermes Agent was designed server-first (always-on, persistent memory, self-improving skills), which makes it the natural fit for this setup. OpenClaw began machine-local and is famously flexible. Both are full MCP clients, so the domain tooling in this series works identically with either. Try the one whose community and docs you prefer.

What happens to the agent if I move it to a new server?

This is the domain payoff: build the new VPS, install the agent, restore its data, and update one A record. Every integration pointing at agent.yourdomain.com follows automatically. With a bare IP, every integration breaks.

This concludes the Agents & Domains miniseries: the landscape, the hands-on setup, agentic buying, agent identity, and this post. For everything our platform offers agents, start at AI integration.

Image generated with AI (Higgsfield).