How Much Does It Really Cost to Run OpenClaw or Hermes?

OpenClaw and Hermes are free, open-source software, but actually running one costs anywhere from near $0 to $500 or more a month, and the number that blows up is not the server, it is the AI tokens. A cheap VPS is $5 to $15 a month; the tokens behind an active agent run from $50 into the hundreds. The only way to make that predictable is a plan where the AI is included at a flat rate with no token meter. Here is the honest breakdown.

The costs nobody separates

Running an OpenClaw-style agent (OpenClaw itself, or the Hermes agent from Nous Research) has separate costs, and most “how much does OpenClaw cost” articles blur them together:

Cost	Typical range	Notes
The software	$0	OpenClaw and Hermes are free, MIT-licensed
The server	$0 to $24/mo	Free on a laptop or Oracle free tier; ~$5 to $24 on a VPS
A managed host	$6 to $49/mo	Removes server work, but usually the AI is still not included
The AI (tokens)	$10 to $500+/mo	Bring-your-own-key, variable, the real bill
WhatsApp messages	per message from Oct 2026	Only if your agent runs on WhatsApp

The first two are small and fixed. The fourth is the one that surprises people.

The server is the small, fixed number

Self-hosting is genuinely cheap. The software is free and open-source↗, and you can run it for close to nothing:

• $0 on an old laptop, a Mac Mini, or a free-tier cloud instance (Oracle’s always-free tier).
• Roughly $5 to $15 a month on a basic VPS (Hostinger around $6, OVHcloud around $4, AWS Lightsail $12 to $24, DigitalOcean around $24).
• $6 to $49 a month for a one-click managed host that patches and updates the box for you.

This is the number the “I run OpenClaw for $6 a month” videos quote. It is real, and it is not what hurts.

The AI is the number that explodes

Almost every self-host and managed-host setup is bring-your-own-key: you plug in an OpenAI, Anthropic, or OpenRouter key and pay per token, on top of the server. That line is not small and it is not stable.

Real operators, not vendors, report the range. Light, carefully routed use lands around $10 a month. But active agents regularly run much higher: one widely-shared writeup was titled “I tried the free AI agent, it cost me $500,” with a VPS at $23 to $70 and the model behind it at $50 to $500 or more. Reddit threads are full of people cutting bills from $420 down to $168, or asking for help before they spend “$100 every few days.” At the extreme, the project’s own founder reportedly ran up around $1.3 million in API tokens in a single month. The point of the extreme is not that you will hit it, it is that nobody who did meant to.

Why the token bill is unpredictable by design

An OpenClaw or Hermes agent is token-hungry for structural reasons, not because you are doing something wrong:

• It re-sends a large tool-definition and system-prompt payload on every single request, often 15,000 to 20,000 tokens before you have typed anything, because it has to re-describe every skill it can use.
• Scheduled jobs, heartbeats, and agent loops fire whether or not you are watching.
• Long documents and growing context multiply the tokens per turn.
• Premium models running around the clock compound all of the above.

So the cost grows in lockstep with how useful the agent is, which is the opposite of what you want, and you do not learn the number until the invoice arrives. It is the same unpredictable-bill problem that pushes companies off per-token pricing everywhere else.

And from October 2026, WhatsApp adds a meter

If your agent lives on WhatsApp, there is now a fourth cost. From October 1, 2026, Meta charges per message for the service replies your assistant sends inside the 24-hour window, which used to be free, and the charge applies to third-party AI assistants but not to Meta’s own. We cover it in WhatsApp’s new service-message charge. Telegram has no per-message charge, which is one more reason it is the better channel for a personal agent.

The honest comparison

Priced against the real options:

Setup	Monthly	Predictable?
Self-host VPS + your own key, light use	~$15 to $30	No, tokens vary
Self-host VPS + your own key, active use	$100 to $500+	No
Managed host + your own key	$50 to $100+	No, AI usually not included
Tessera Founder (agent + AI, flat)	55 EUR / 55 USD	Yes

Every row except the last leaves the AI bill floating.

What the cost-cutting advice misses

Search “openclaw cost reduction” and you get the same playbook everywhere: switch to a cheaper model, add caching, prune context, disable heartbeats, use a router. All of it works, and all of it is worth doing. But notice what it does not do: it lowers a number you still cannot predict. You are optimizing a meter, not removing it. Next month a longer project, a busier week, or a model you left on premium quietly moves the bill again. For a solo founder who wants to know what they pay, “cheaper but still variable” is not the same as “known.” That is the gap none of the cost guides close, because most of them are written by the hosts and model vendors who bill you per token.

How to make it flat: the AI included, no token meter

The fix operators actually want is not a cheaper model, it is a plan where the variable line is gone. Tessera Founder hosts a private Hermes instance (the open-source, OpenClaw-style agent) and includes the AI itself at a flat 55 EUR (or 55 USD) a month, with no token meter. Your assistant can work around the clock; the only limit is pace, 20 requests per minute, which is plenty for a one or two person company.

We connect Telegram on day one through the official Bot API, with no per-message charge, and set up WhatsApp through the official Meta API if you want it. It is the same OpenClaw alternative argument, seen from the cost side: the agent is free, the server is cheap, and the one line you cannot predict, the AI, becomes a fixed number. For a one or two person company that just wants the assistant working, that predictability is usually worth more than shaving a few dollars off a bill that still moves.

FAQ

Is OpenClaw free?

The software is free and open-source (MIT license). The costs come from the server you run it on and the AI tokens it consumes. So the honest answer is that OpenClaw is free to install and not free to run.

How much does OpenClaw cost per month?

Anywhere from near $0 to $500 or more. A cheap VPS is $5 to $15 a month, or free on a laptop; a managed host is $6 to $49. The variable part is the AI: light use is around $10 a month, but active agents commonly reach $100 to $500 or beyond.

Why do OpenClaw token costs get so high?

The agent re-sends heavy tool-definition and context on every request, sometimes 15,000 to 20,000 tokens before your first word, and scheduled loops run on their own. Combined with premium models and long documents, the token count grows fast, and it scales with how much the assistant works.

Can I reduce OpenClaw costs?

Yes, partly: route simple tasks to cheaper models, prune context, and turn off unnecessary scheduled jobs. But that lowers a bill you still cannot predict. A flat plan that includes the AI removes the variable line entirely.

More cost questions

Is a managed OpenClaw host cheaper than self-hosting?

Not necessarily. A managed host removes the server work, but most still expect you to bring your own AI key, so the variable token bill stays. The cheaper-in-practice option for a predictable total is a plan that includes the AI, not just the hosting.

Does WhatsApp add to the cost?

From October 1, 2026, yes. Meta charges per message for service replies sent by third-party AI assistants on WhatsApp. Telegram has no such charge, so running your agent on Telegram avoids it.

What is the cheapest predictable way to run an agent?

If you want a known monthly number, a managed plan with the AI included, like Tessera Founder at a flat 55 EUR or 55 USD, is more predictable than self-hosting with a bring-your-own-key token bill that can swing from $10 to $500 month to month.