The operating systemis free. Compute is not.
Everything that runs on your own machine costs nothing, because it costs us nothing. The only thing we charge for is inference we have to pay for on your behalf.
Free
The complete operating system, running entirely on your hardware. Not a trial and not feature-gated. This is the product.
- All six surfaces: chat, code, browser, simulation, voice, agents
- A2UI with the full widget library
- Megha 0.6B running locally, offline capable
- Local vector memory over your own documents
- MCP connectivity for local stdio servers
- All three editions: Mac, desktop, and full OS
- Community support
MeghaOS Pro
OptionalOptional. Routes the work you choose to a hosted Frontier model instead of the local model, for tasks a 0.6B model cannot carry.
- Everything in Free, unchanged
- Metered access to hosted Frontier model
- You choose per task; local stays the default
- Four usage tiers from $25 to $200 a month
- Priority queue and support scale with the tier
- Cancel at any time from your account
Enterprise
Fleet deployment with the controls, audit and support a regulated environment needs. Including fully air-gapped, where no hosted routing exists at all.
- Everything in Free
- Fleet deployment with atomic, reversible updates
- Policy as configuration, version-controlled
- Attributable audit trail with export
- Air-gapped installation and offline licensing
- SSO and directory integration
- Named support contact and deployment assistance
Two different things, priced differently.
MeghaOS is the free part, and that is not a promotional position; it is a consequence of the architecture. The agent runs on your metal, on your own hardware, against a model on your own disk. It costs us nothing to run because we are not running it.
Hosted inference is the paid part. A 0.6B model is extraordinary for its size and still not a frontier model. When a task needs one, you can route that task, and only that task, to a hosted Frontier model. That compute is billed to us, so it is billed to you.
The consequence is worth stating directly rather than leaving you to infer it: a request routed to the hosted model leaves your device. That is what hosted means. Local is the default, the choice is per task, and if your requirement is that nothing ever leaves the machine then Free already meets it, as does an air-gapped Enterprise deployment, where the hosted path is not installed at all.
Four sizes of the same thing.
The plans differ by how much hosted usage is included, not by what the software can do. Billing lives in your account; this site never handles a payment.
| Plan | Per month | Included usage | Support | Queue |
|---|---|---|---|---|
| Starter | $25 | Access to fine-tuned frontier models | Community support | Standard speed |
| Pro | $40 | Access to fine-tuned frontier models | Priority email support | High priority queue |
| Ultra | $100 | Access to fine-tuned frontier models | 24/7 priority support | Highest priority queue |
| Max | $200 | Access to fine-tuned frontier models | Dedicated account manager | Enterprise SLA |
Opens login.meghaos.ai, where your account and billing live.
What people ask about pricing.
Is MeghaOS itself free?
Yes, completely, and it stays that way. The operating system, the desktop environment, the Mac app and the local Megha 0.6B model cost nothing because they run on your hardware and cost us nothing to run. There is no paid version of MeghaOS.
Then what am I paying for?
Hosted inference, and only that. A 0.6B model running on your laptop is remarkable for its size but it is not a frontier model. If a task needs one, the paid tiers let you route that specific task to a hosted Frontier model. We pay for that compute, so we charge for it.
Does using a paid tier mean my data leaves my device?
For the requests you route to the hosted model, yes. That is what routing to a hosted model means, and we are not going to pretend otherwise. Local remains the default and you choose per task. If your requirement is that nothing ever leaves the machine, stay on Free or talk to us about an air-gapped Enterprise deployment, where the hosted path does not exist at all.
What does the included amount mean?
Each tier includes a monthly allowance of cloud compute spend. The allowance is metered against actual model usage, so light months do not consume it.
Will the Free tier get worse when I stop paying?
No. Everything listed under Free today stays under Free, whether or not you have ever paid. Cancelling removes hosted routing and nothing else; the machine in front of you keeps working exactly as it did.
Do I need an account?
Not for local work. Running a local model needs no account and no network call; you select it at first launch and it works offline from then on. An account is only needed to buy hosted inference, because something has to meter it.
Is the model open?
Megha 0.6B is distributed under an open Apache 2.0 license for use with Ollama, vLLM, SGLang, and Hugging Face Transformers. Weights and training recipes are openly accessible.
How does Enterprise pricing work?
It is scoped per deployment: fleet size, air-gap requirements, support level. Given where the product is, those conversations are currently design-partner shaped rather than transactional.
Start with free.
The whole operating system, on your own machine, with nothing to enter a card for.