What Hermes Agent Actually Costs Per Month
Almost every surprise in a Hermes budget comes from treating it as one price. It is three separate things. The software is free and open source. The always-on machine it runs on is a fixed monthly cost. The AI model usage is a variable bill you pay your own provider directly. This guide separates them, shows you how to estimate the variable part before you commit, and tells you where the real cost controls sit.
The three things you are actually paying for
The software
Free
Hermes Agent is open source. No licence, no seat fee, no per-agent charge from the project itself.
The server
From EUR 14.99/mo
A dedicated machine that stays awake so your agent can act on a schedule and message you first. Fixed and predictable.
The model usage
Variable, paid to your provider
Billed to your own API key at the provider's published rate. It never passes through us, so we never mark it up.
Part one: the fixed hosting cost
This is the part you can know exactly, in advance, forever. Every plan is a dedicated VPS in Denmark, inside the EU, with provisioning, SSL, security updates, snapshots, and monitoring done for you. There is no setup fee and no per-message charge from us. The plan price is the plan price, and it is the same whether you run Hermes Agent or OpenClaw on it.
| Plan | Server | Monthly | Annual (per month) | Billed yearly |
|---|---|---|---|---|
| Starter | 2 vCPU, 4 GB RAM, 15 GB NVMe | EUR 14.99 | EUR 12.49 | EUR 149.88 |
| Hobbyist | 4 vCPU, 8 GB RAM, 25 GB NVMe | EUR 25.99 | EUR 20.99 | EUR 251.88 |
| Pro | 8 vCPU, 16 GB RAM, 40 GB NVMe | EUR 49.99 | EUR 39.99 | EUR 479.88 |
The full feature comparison and current offers live on the hosting pricing page, and the Hermes product itself is described on the managed Hermes Agent hosting page.
Part two: the variable AI usage cost
Every plan is BYOK, bring your own key. You connect your own provider account and that provider bills you directly for what your agent uses. We never resell model access and never add a margin, which means two things: you pay the raw provider rate, and we genuinely cannot quote your model bill for you, because it depends on your usage and your provider's current prices.
That is an honest limitation rather than a dodge. Published per-token prices move often enough that any figure written into a page like this would be wrong within months. What does not move is what drives the bill, so estimate from the drivers instead.
One thing is worth calling out for Hermes in particular. A proactive agent is designed to start conversations. That is the point of it, and it is also why a Hermes bill can look different from a request-and-reply chatbot bill even on identical hardware. Budget for the messages you did not ask for, not only the ones you did.
How often the agent speaks
The biggest driver by far, and the one people underestimate with Hermes specifically. An agent that messages you proactively is, by design, sending messages you did not ask for. Useful, but every one of them is a model call.
How much context each reply carries
Every reply re-sends the relevant history. Long-running threads, large pasted documents, and a heavy system prompt multiply the input cost of each individual message, quietly.
Which model you point it at
Cost per million tokens differs by an order of magnitude between a provider's small and flagship models. Routing routine chat to the smaller one and keeping the flagship for real reasoning is the fastest cut available.
How much tool and file work it does
Agentic work costs more than chat. A request that reads files, searches, and takes several steps is several model calls, not one. Skills that run on a schedule multiply that by however often they run.
How to estimate your own number, in four steps
- 1
Count the messages, not the hours
Write down how many exchanges you realistically expect per day, then add the scheduled work: a morning briefing, a nightly summary, anything on a cron. Scheduled items are the ones people forget, and they run whether you read them or not.
- 2
Look up your provider's current rate
Take the input and output price per million tokens for the exact model you plan to use, from your provider's own pricing page today. Do not use a number from a blog post, including ours.
- 3
Assume more input than you think
A single reply usually re-sends a chunk of conversation history plus your system prompt. Input tokens routinely outweigh output tokens several times over, which is why long threads get expensive quietly rather than loudly.
- 4
Set the cap before you go live
Put a hard spend limit and an alert on the API key in your provider dashboard. Then run the agent for a week and read the real number. One week of actual data beats any estimate on any page.
The cheap VPS comparison, honestly
You can run Hermes Agent on a bare VPS for a few euros a month, and for some people that is genuinely the right answer. If you are comfortable on a command line and you enjoy this part, go and do it. We would rather tell you that than sell you something you do not need.
The honest comparison is not our price against the VPS line item. It is our price against the VPS line item plus the work: Node and Hermes upgrades, SSL renewal, firewall rules, backups, monitoring, and being the person who wakes up when the agent stops replying. That work has a cost even when it does not appear on an invoice.
Where a managed plan earns its money is the day something breaks and it is not your problem. Where it does not is if you would have enjoyed fixing it anyway.
Hermes Agent cost FAQ
- Is Hermes Agent itself free?
- Yes. Hermes Agent is open-source software from Nous Research, so there is no licence fee to use it. What costs money is the always-on machine it runs on and the AI model usage it consumes. Those are two separate bills from two separate places, and conflating them is where most budget surprises come from.
- Does the OneClickClaw price include AI model usage?
- No, and that is deliberate. Every plan is BYOK: you connect your own Anthropic, OpenAI, or other provider account, and that provider bills you directly at their published rates. We never resell model usage and never add a margin to it, so you pay the raw provider price and can audit it on your own provider dashboard rather than taking our word for it.
- What is the minimum realistic monthly cost for a Hermes agent?
- The floor is the hosting plan, EUR 14.99 per month on Starter (EUR 12.49 per month billed annually), plus whatever your AI provider charges for the messages your agent actually sends. A quiet personal agent adds very little on top. An agent that answers a busy channel all day can add more than the server costs. The estimation method further up this page is how you work out your own number before you commit to anything.
- Does Hermes cost more to host than OpenClaw?
- No. The plans, the hardware, and the prices are identical for both. A Hermes agent and an OpenClaw agent on the same tier get the same dedicated server in the same Danish data centre. What differs between the two is how each project works, not what we charge for the machine underneath.
- Why does an agent need an always-on server at all?
- Because a Hermes agent is supposed to reach you first, not wait for you to open an app. Scheduled work, channel listeners, and proactive messages only happen while the process is running. On a laptop that closes, sleeps, or changes network, the agent goes quiet without telling you. That is the specific problem a dedicated server solves.
- Can I cap what the AI usage costs me?
- Not from us, because we never hold your provider account, but you can cap it at the source and you should. Every major provider lets you set a hard spend limit and usage alerts on the API key you bring. Setting one before you go live is the single most effective cost control available to you, and it takes about two minutes.
- What happens to the bill if I upgrade the plan later?
- You move up a tier without rebuilding anything, and Stripe prorates the change. A larger plan buys more CPU, RAM, disk, and bandwidth on your dedicated server. It does not change your AI provider bill, which is driven by how much your agent talks rather than by how big the machine is.
Keep exploring
- Managed Hermes Agent hostingWhat you get, how fast it goes live, and what the plans include.
- Hermes Agent VPS hostingManaged, plain VPS, or your own machine. What each one actually asks of you.
- Hosting plans and pricingThe full tier comparison, annual pricing, and what every plan includes.
- OpenClaw cost guideThe same breakdown for the other agent we host, if you are weighing both.