How much does OpenClaw cost? (It's free — here's what isn't)
OpenClaw is free. Compare hosting, model and connected-tool costs with a dated VPS price, worked monthly budgets and what an existing ChatGPT account can cover.
OpenClaw itself costs $0. Running it adds the computer, model access, and any paid tools you connect. An existing computer avoids a new hardware purchase; a VPS adds monthly rent; a ChatGPT account can cover supported model calls within its Codex allowance. At Everpod, a managed pod is $29/mo with up to $10 of model usage included each month. The cheapest setup depends on what you already own and whether you want to maintain the computer yourself.
The OpenClaw license is free
OpenClaw’s MIT license permits using and modifying the software without a license fee. A hosting subscription pays for a service around that software, not for permission to use OpenClaw.
Separate the computer from the model
Installing OpenClaw on a Mac does not make a cloud model run locally. The computer runs your agent and its tools; the model can still be a separately billed service. Running the model locally is a different hardware decision, covered in the local-model versus cloud comparison.
| Cost | What you pay for | What changes the bill |
|---|---|---|
Computer | Hardware and power, or hosting | Memory, storage, backups and optional services |
Model | API usage or a subscription allowance | Model choice, context size, task length and retries |
Connected tools | Paid search, voice or other services you enable | Each service’s own price and usage limits |
What a self-hosted budget looks like
As a dated price reference, Hetzner’s June 2026 price table lists the CX23 at $6.49/month in Germany or Finland, excluding IPv4 and VAT. It is a server price, not a complete OpenClaw bill or a guarantee that the plan is in stock. Add any backup, address and storage charges from the quote you actually receive.
If your model usage then comes to $10 in a month, those two items total $16.49 before the excluded charges and other paid tools. With $40 of model usage, they total $46.49. Those are arithmetic examples, not estimates of typical usage: the server price alone cannot tell you what an autonomous agent will cost.
For a computer you already own, count electricity and any additional hardware you buy, not a second purchase of the same machine. OpenClaw’s hardware guidance recommends at least 2 GB of RAM for headroom and notes that browser automation needs more resources than the basic gateway. Size the machine for the work you intend to run.
Can an existing ChatGPT subscription remove the API bill?
It can cover supported agent model calls through OpenClaw’s ChatGPT sign-in. OpenAI lists Free and paid plans on its Codex pricing page, with different allowances. Staying within an account’s included allowance avoids a separate API charge for those calls; it does not buy unlimited agent work or credit for every OpenAI API.
With API billing, measure a representative task before scheduling it to repeat. A long conversation can send substantial context on several model calls, so one task is not one billable request: one agent job, metered call by call, took 11 calls on one model and 21 on another. Check the provider’s spending controls: a budget alert is not necessarily a hard stop. A fallback using an API key can also turn a subscription limit into metered spending.
What managed hosting includes
Everpod’s managed OpenClaw hosting gives your agent a private, always-on computer, set up, secured and backed up, with model usage included. The standard pod is $29/mo with up to $10 of model usage each month. You can also connect your own model account; charges on that account remain separate.
The comparison is not just a monthly server price. Self-hosting leaves setup, updates, backup recovery and security upkeep with you. If that is work you want to do, compare the machines in the Mac Mini, VPS and managed-pod guide. If you want the agent without taking on that maintenance, that is what the managed service is for.