Hermes Agent is an open-source AI agent (from Nous Research, MIT-licensed) that plugs into Claude Code and gives it hands: shell access, file operations, and tool integrations, all running on infrastructure you control. Hostinger offers it two ways — a managed plan with zero server admin, or a one-click VPS template for full control. This guide covers both paths, connecting it to Claude Code, and running your first task.
Want the full picture first — what Hermes is, what it costs, how it differs from OpenClaw? Read the hub: Hostinger Hermes Agent: What It Is and What It Costs.
What you need before you start
- A Hostinger account with the AI Automation plan or a VPS — the plan is available through our link and provisions in minutes.
- A model provider API key. Hermes Agent thinks with a third-party model, so generate an API key in your model provider’s dashboard before you begin.
- Claude Code installed locally (optional for the basic setup, required for the integration section below).
- SSH client (Terminal, PuTTY, or Windows Terminal) if you take the VPS path.
Before you start: pick your path
- Path A — Managed AI Automation plan: Hostinger provisions and maintains the Hermes Agent environment. You configure the agent, not the server.
- Path B — VPS one-click template: Hermes Agent pre-installed on your own VPS. Full control over models, tools, and networking — and full responsibility for updates.
Choose managed if you want to use an agent this week. Choose the VPS if you plan to customize deeply, run multiple agents, or integrate with infrastructure you already host.
Path A: Managed setup (about 15 minutes)
- Order the AI Automation plan. One plan covers Hermes Agent, OpenClaw, n8n, and Paperclip — switchable from the control panel. Complete checkout and wait for provisioning.
- Open the Hermes panel. In hPanel, find the AI Automation section and open the Hermes Agent dashboard. Note the connection details: endpoint URL and any auth token shown.
- Add your model API key. Hermes Agent needs a language model to think with. Paste an API key from your model provider in the settings. Model usage is billed by the provider, not by Hostinger — light use typically costs a few dollars a month.
- Verify it responds. Most panels include a test prompt or status indicator. Send “summarize what you are and what tools you have” — a coherent answer means the agent core is live.
Path B: VPS one-click template (about 30 minutes)
- Deploy the VPS with the Hermes template. Select the Hermes Agent template during VPS ordering. Deployment takes a few minutes.
- Save the credentials. Server IP and root password arrive in the deployment summary or welcome email — store them in a password manager.
- SSH in once to confirm the agent service is running and to change default passwords. The template docs list the service name and the web UI port.
- Configure the model key in the settings file or web UI, same as Path A step 3.
- Secure the endpoints. Put the agent’s web UI and API behind HTTPS (reverse proxy + a domain A record), and firewall everything except the ports you actually use. An agent with shell access must never sit on the open internet unprotected.
Connect it to Claude Code
Hermes Agent’s standout trick is pairing with Claude Code — Anthropic’s agentic coding tool — so your coding assistant can reach beyond the local machine:
- Install Claude Code on your local machine per Anthropic’s docs, if you have not already.
- Add Hermes as a tool endpoint. In your Claude Code configuration, register the Hermes Agent endpoint URL and auth token from your panel (Path A) or server (Path B). Hermes exposes its tools — shell, files, integrations — as callable functions.
- Test with a safe command. Ask Claude Code to “ask Hermes for the server’s disk usage.” A correct answer proves the whole chain: Claude Code → Hermes → your Hostinger environment → back.
Your first real task
Try something with a verifiable result: “Check my website’s SSL certificate expiry and disk usage, and report both.” It exercises tool use, produces checkable facts, and shows you the agent’s working style. Review what it did before granting broader permissions — the first tasks are as much about calibrating your trust as about the output.
Good second tasks once the basics work: “Watch this log file and alert me if the word ERROR appears,” or “Every Monday, summarize the merged pull requests in this repo and post the summary to our channel.” Each one teaches you a new capability — monitoring, scheduling, integrations — while doing something genuinely useful.
Safety basics (do not skip)
- Least privilege first. Grant only the tools the current task needs. Expand permissions deliberately, not by default.
- Human approval for irreversible actions. File deletion, deployments, sending messages — anything you cannot undo should ask first.
- Keys in the vault. API keys and tokens live in the agent’s credential store, never in prompts or chat logs.
- Separate concerns. If you run both Hermes and OpenClaw, give them separate credentials and separate scopes — one compromised agent should not mean two.
- Keep it patched. Managed plan: automatic. VPS: update the template on a schedule; agent frameworks get security fixes often.
Frequently asked questions
What is the difference between Hermes Agent and OpenClaw?
Both are AI agents you can host on Hostinger. OpenClaw is positioned as a personal assistant (messaging-first, everyday tasks); Hermes Agent is a developer-leaning open-source agent from Nous Research with a standout Claude Code integration. Our hub pages compare them in detail: Hermes Agent hub and OpenClaw hub.
Do I need Claude Code to use Hermes Agent?
No. Claude Code integration is a highlight, but Hermes Agent works standalone through its own web UI and API. Connect Claude Code only if you want your coding assistant to reach the agent’s tools.
Does the managed plan include the AI model costs?
No — same as OpenClaw. Hostinger provides the hosting and the agent software; you supply a model provider API key and pay that provider for usage. Expect a few dollars a month for light personal use.
Can I switch from the managed plan to a VPS later?
Yes. Export your agent configuration (prompts, tool settings, credentials list — not the secrets themselves) from the managed panel, deploy the VPS template, and re-enter them. Plan for an hour; the model key and channel tokens transfer as new paste-ins.