Is OpenClaw Free? The Real Costs in a Business 2026 (license €0, but…)
Is OpenClaw free? Yes, the license is €0 (open source under MIT). But a working agent costs money: the model (tokens), the server (a VPS from about €5 a month), and your time on boundaries and maintenance. We break three typical setups down into real numbers.
OpenClaw is open-source software under the MIT license, so the license itself costs €0. The real cost of running it is three lines: the model (tokens) billed by usage, the server (a VPS from about €5 a month), and your time on boundaries and maintenance. "Free" applies to the code, not to a working agent.
Quick answer
Yes, OpenClaw (people also type "open claw") is free as a license. It is open source under the MIT license: you download, run, and modify it with no license fee, no subscription, and no per-seat limit (OpenClaw documentation, 2026). The code is on GitHub and always will be.
The cost begins where the agent begins to work. It breaks down into three lines:
- license: €0 - the open-source OpenClaw code, no fee for the software,
- server (VPS): from about €5 a month - the machine the agent runs on around the clock,
- model (tokens): by usage - every request and every step the agent takes is a model call you pay the provider for,
- time on boundaries and maintenance: the line that never appears in a price list and is often the most expensive - setting boundaries, escalation, and an audit trail, then watching it after updates.
Self-host open claw and you pay those last three lines with your own budget and your own evening. Buy it as a service and you pay someone so you do not have to. Below are the numbers for both.
Why the license really is €0
Because OpenClaw is released under the MIT license, one of the most permissive open-source licenses. You can download it, install it, change the code, and use it commercially, with no license fee, no "free plan with limits," and no per-seat charge (OpenClaw documentation and independent reviews, 2026). This is not a trial that expires.
So "OpenClaw is free" is both true and misleading. True, because you genuinely will not pay for the program. Misleading, because the program is not the same as a working agent. A real-world analogy: a car handed to you for free is still fuel, insurance, servicing, and your time behind the wheel. The OpenClaw license is a gift. The rest is running it.
The server: where "free" ends at the VPS bill
An agent that is meant to work while you are away from the keyboard has to run somewhere around the clock. That is a server, usually a VPS. You can host it on your own hardware ("free" then means electricity and your time on admin) or rent it.
A concrete price, today: at Hetzner the cheapest sensible VPS for a personal agent (CX23, shared) costs about €5.49 a month net, and the Arm variant (CAX11) about €5.99 (Hetzner pricing after the 15 June 2026 standardization). For a business setup with guaranteed performance (a dedicated vCPU, CCX13), the price list starts at about €42.99 a month.
One honest note: this is not a price set in stone. Hetzner raised its cloud prices sharply in mid-June 2026, some plans by more than 100% (Hetzner announcement, June 2026). The server is cheap, but it is not zero and it is not static. When you budget, treat it like an electricity bill: small, but every month.
The model (tokens): why a free tool still generates cost
This is the line most "how to run a free agent" guides skip. OpenClaw does not think on its own. Every time the agent understands, plans, or executes a step, it sends a request to a language model (Claude, GPT, a local model). The model is billed per token, and that is the real, variable cost of the "free" tool.
OpenClaw is not tied to one model provider. You can plug in a US model (Anthropic, OpenAI), a Chinese one (DeepSeek, Kimi, GLM, MiniMax), or your own local model. The price gap between them is huge, and that gap, not the license, is what decides your bill.
Providers bill input tokens (everything you send: the instruction, history, documents) and output tokens (what the model writes back) separately, and output is usually several times more expensive. Concrete rates per million tokens (input / output), order of magnitude as of July 2026:
| Model | Provider | Per 1M tokens (input / output) |
|---|---|---|
| DeepSeek V4-Flash | DeepSeek (China) | ~$0.14 / $0.28 |
| MiniMax M2 | MiniMax (China) | ~$0.26 / $1.00 |
| Kimi K2.6 | Moonshot (China) | ~$0.95 / $4.00 |
| GLM-5.2 | Zhipu / Z.ai (China) | ~$1.40 / $4.40 |
| Claude Haiku 4.5 | Anthropic (US) | $1 / $5 |
| GPT-5.4 | OpenAI (US) | $2.50 / $15 |
| Claude Sonnet 5 | Anthropic (US) | $3 / $15 |
| Opus 4.8 | Anthropic (US) | $5 / $25 |
| GPT-5.5 | OpenAI (US) | $5 / $30 |
Sources: Anthropic, OpenAI, DeepSeek, Moonshot, Z.ai, and MiniMax pricing pages, June-July 2026. Provider rates change fast and are often promotional, so treat the table as an order of magnitude and check the current price at the source.
Note the spread: between the cheapest open model and the top flagship, output cost differs by more than 100x ($0.28 versus $30 per million tokens) for the same unit of work. Cheaper models handle simple, repetitive steps; you keep the flagship for the work that genuinely needs it. This is the single biggest cost lever you have.
One honest GDPR note: a cheaper model from a non-EU provider means data leaves the European Economic Area. For personal or company data that is a separate question to settle (a data-processing agreement, where processing happens), not just price. The safest option is a local model on your own server, covered below.
What does that mean per task? A single request is fractions of a cent. But one agent task is usually not one request; it is a dozen-plus model calls in a loop. So you calculate the cost of a task on its real number of steps, not on a single request.
Monthly model cost =
number of tasks per month
x average model calls per task
x average tokens per call
x the provider's token priceSubstitute your numbers, not ours. At small, personal usage this is often a few dollars a month. For a business agent working all day it is tens to hundreds, depending on model and volume. You can cut it with a local model (for example via Ollama), which brings token cost to zero, at the price of quality and heavier hardware.
What OpenClaw costs in three typical setups
This is not a market price list, just a way to read the budget. The key column is the last one, "time on boundaries and maintenance." It usually decides whether the "free" agent is actually worth it.
| Setup | License | Server (VPS) | Model (tokens) | Time on boundaries and maintenance |
|---|---|---|---|---|
| Hobby (local, single user) | €0 (MIT) | €0 (own computer) or free tier | cents per request, or €0 with a local model | your evening when something breaks |
| Personal operator on a VPS | €0 (MIT) | from ~€5.49/mo (Hetzner CX23) | a few to a few tens of € per month by model and task count | a few hours to set up plus watching it after updates |
| Business with boundaries | €0 (MIT) | from ~€42.99/mo (dedicated vCPU) and up | tens to hundreds of € per month by volume and model | a real line: boundaries, escalation, trail, on-call for failures |
For comparison: the fully managed version (OpenClaw Cloud) costs $49 a month and takes the server, configuration, and maintenance off your plate (getopenclaw.ai, 2026). That is exactly the price of trading "your time" for "someone else's time."
The real bill for a "free" agent
Sum it into one formula you substitute your own numbers into. Because "€0 license" is true, just not the whole truth.
Real monthly cost of "free" OpenClaw =
€0 (license)
+ server (a VPS or electricity for your own hardware)
+ tokens (tasks x steps x model price)
+ your hours on boundaries and maintenance x your rateThe first three lines are usually a few tens of euros. The fourth line is the one nobody puts in the "free" headline. If your hour is worth €35 and the agent takes two hours a week in fixes and watching, that is about €280 a month, at a zero license. That is why the honest answer is: OpenClaw is free, but being its administrator is not.
When NOT to self-host OpenClaw
Honestly: there are situations where standing up open claw with your own hands is a bad purchase, however free the license is. In those cases the managed version, or an implementation from someone, comes out cheaper, not more expensive.
- You have no production owner. If no one in the company reacts when the server goes down at two in the morning or the model provider changes its API, the agent will eat time instead of saving it. Then $49 for the managed version, or maintenance from a vendor, is cheaper than your on-call shift.
- The process touches personal data and money without defined boundaries. An agent that sends emails to customers or changes data in a system needs boundaries, escalation, and a trail before it goes to production. That is not part of the install; it is separate, serious work (and a separate GDPR topic).
- You want to save money, and your time counts. Once you count your time on learning, configuration, and watching it, a "free" self-hosted setup often costs more than a finished service. The software is free; your attention is not.
If any of these fits you, self-hosted OpenClaw is not the cheapest route, just the most labor-intensive. For more on what actually decides a business rollout, see our guide to OpenClaw in a company, and for how OpenClaw compares with our variant, the Hermes vs OpenClaw comparison.
How to start
- Book a free process scan and show one process you would like to hand to an agent.
- Prepare: how many times a month the process happens, which systems are in the path, where exceptions appear, and who on your side reacts when something breaks.
- After the call you get a recommendation: self-host it (and learn how), order it as a personal agent, or an honest "stay on managed for now."
Are you a developer who wants to build and self-host agents the engineering way? That is exactly the content of our AI-Native Course: boundaries, trail, escalation, and token costs, not just installation. Prefer a personal agent that is set up and maintained without your time on watching it? That is what the personal AI agent service does.
Book a free process scan | AI-Native Course | Personal AI agent | Pricing