What OpenClaw Actually Costs Per Month
Almost every surprise in an OpenClaw budget comes from the same mistake: treating it as one price. It is three separate things. The software is free, the server it runs on is a fixed monthly cost, and 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 cost controls actually are.
The three things you are paying for
1. The software: free
OpenClaw is open source under the MIT licence. There is no licence fee, no per-seat cost, and no paid edition to unlock.
2. The server: fixed
An agent that answers you at any hour has to be running at any hour. That is a monthly server cost, and it does not change with how much you chat.
3. The AI usage: variable
Every reply consumes tokens from your own provider account. This is the part that moves, and the part most cost estimates leave out entirely.
Part one: the fixed hosting cost
On OneClickClaw this is a single predictable line. Each plan is a dedicated single-tenant VPS in Denmark (EU) with the server work done for you: provisioning, SSL, security updates, snapshots, and monitoring. There is no setup fee and no per-message charge from us: the plan price is the plan price.
| 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 |
Feature-by-feature comparison and current offers are on the OpenClaw hosting pricing page, and the hardware behind each tier is described under managed OpenClaw VPS hosting.
Part two: the variable AI usage cost
Every OneClickClaw plan is BYOK, bring your own key. You connect your own account with Anthropic, OpenAI, Google, xAI or another supported provider, and that provider bills you directly for what your agent uses. We never resell model access and never add a margin to it, 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, not a dodge. Published per-token prices change often enough that any number written into a page like this would be wrong within months. What does not change is what drives the bill, so estimate from the drivers instead.
How many messages per day
The single biggest driver. An agent you message a few times a day costs a fraction of one that answers a busy group chat or runs scheduled work around the clock.
How much context each reply carries
Every reply re-sends the relevant conversation history. Long threads, large pasted documents, and big system prompts multiply the input cost of each message.
Which model you point it at
Cost per million tokens differs by an order of magnitude between a provider's small and flagship models. Choosing the smaller model for routine chat is the fastest way to cut spend.
How much tool and file work it does
Agentic work costs more than chat. A request that reads files, searches, and takes several steps makes several model calls, not one.
How to estimate your own model spend in five minutes
This gives you a defensible number using your provider's current published prices rather than someone else's guess.
Count the messages you will actually send
Be concrete: a personal assistant you use through the day might be 20 to 50 exchanges; a bot answering a group chat could be several hundred. Write the daily number down and multiply by 30.
Estimate the context each exchange carries
A short question with a short reply is small. A reply that has to re-read a long thread, a system prompt, and a pasted document is much larger. Treating a typical exchange as a few thousand input tokens plus a few hundred output tokens is a reasonable starting shape for ordinary chat.
Look up your chosen model's current price
Take the input and output price per million tokens straight from your provider's own pricing page on the day you plan. Small models and flagship models can differ by 10x or more for the same conversation.
Multiply, then double it
Monthly exchanges times tokens per exchange times price per token gives your base figure. Double it as a safety margin for retries, tool steps, and the days you use it more than you expected. If the doubled figure is comfortable, the real one will be.
Set a hard spend limit before you go live
Configure a spending cap and a usage alert on the provider account you are about to connect. This is the only control that can actually stop a runaway bill, and it takes two minutes.
OpenClaw's own documentation on API usage and costs covers the same ground from the software side and is worth reading alongside this.
Four controls that actually reduce the bill
Match the model to the job. Most day-to-day agent traffic does not need a flagship model. Point routine chat at a smaller model and reserve the expensive one for work that genuinely needs it. This usually saves more than everything else combined.
Keep conversations short. Long-running threads re-send their history on every turn, so cost per message climbs the longer a thread runs. Starting a fresh conversation for a new topic is free and immediately cheaper.
Trim the system prompt. Instructions are paid for on every single message. A tight, useful prompt beats a long one that repeats itself.
Choose the plan you need, then leave it. A bigger server does not make model usage cheaper. Pick the tier that fits your workload, take annual billing if you are staying, and put your optimisation effort into the variable bill instead.
The cost people forget: running it yourself
A bare VPS advertised at a few euros a month looks cheaper than a managed plan, and on the invoice it is. The comparison is only fair once you add what the invoice does not include: installing Node and OpenClaw, putting a reverse proxy and a certificate in front of it, keeping the firewall right, applying updates without breaking the agent, arranging backups you have actually tested, noticing when it goes down, and fixing it yourself when it does.
For someone who enjoys that work and already runs servers, DIY is a legitimate choice and we say so plainly on the host OpenClaw without Docker or server setup page, which lays out both paths step by step. For everyone else, the managed price is buying back the hours, not just the hardware.
Whichever way you go, the security work is the same work. The managed OpenClaw VPS security checklist sets out exactly which controls are handled for you here and which stay yours either way.
OpenClaw cost FAQ
- Is OpenClaw itself free?
- Yes. OpenClaw is open-source software under the MIT licence, so there is no licence fee for the software. What costs money is the always-on server it runs on and the AI model usage it consumes. Those are two separate bills.
- Does the OneClickClaw price include AI model usage?
- No, and that is deliberate. Every plan is BYOK: you connect your own Anthropic, OpenAI, Google, xAI, or other provider account, and that provider bills you directly at their own rates. We never resell model usage or add a margin to it, so you always pay the raw provider price and can see it on your own provider dashboard.
- So what is the minimum realistic monthly cost?
- 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 you actually send. A light personal agent can add very little model spend; a heavy always-working agent can add more than the server costs. The estimation method further up this page shows how to work out your own number before you commit.
- Why not just run OpenClaw on a cheap VPS myself?
- You can, and for some people that is the right answer. The cheap VPS line item is only part of the real cost: you also own Node and OpenClaw upgrades, SSL renewal, firewall rules, backups, monitoring, and being the person who fixes it at 2am when the agent stops replying. The managed-versus-DIY breakdown compares both honestly.
- Can I cap what the AI usage costs me?
- Not from OneClickClaw, because we never hold your provider account, but you can cap it at the source. Every major provider lets you set hard spend limits 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.
- What happens to cost if I upgrade the plan later?
- You can move up a tier without rebuilding your setup, and Stripe prorates the change. A bigger plan buys more CPU, RAM, disk, and bandwidth on your dedicated server; it does not change your AI provider bill, which is driven by usage rather than by server size.
Keep exploring
- OpenClaw hosting plans and pricingThe full tier comparison, annual pricing, and what every plan includes.
- Managed OpenClaw VPS hostingThe dedicated EU server behind each plan and what managed covers.
- OpenClaw VPS security checklistWhich security controls we run, and which stay yours.
- Pricing and billing guidesHow billing, upgrades, and invoices work in practice.