Own your AI
We put a name on a domain we control, an agent behind it, and the machine under your desk. Here is what that means, what it costs, and what we will not do to grow.
Most people do not own the AI they use. They rent access to it, and the rent goes up whenever the provider decides it should. Every question you ask is metered, every document you paste in is a deposit somebody else holds, and every improvement in the model is a price rise you are told is progress.
We think that is the wrong shape for a tool this important.
What we actually built
Three things, in the order you meet them.
A name. yourname@ngari.ai — a real email address that answers you. You email it like you text an assistant and the reply comes back to your own inbox in about a minute. No app, no login, no password to remember. It is free to claim, and the $12 a year version holds the mailbox for you.
A lane. Some inference runs on our hardware, close to you. Some routes to a named third-party provider when the work needs a bigger model — DeepSeek today, with Moonshot/Kimi as a fallback. We name which is which, every time, because "private AI" that secretly ships your data to a vendor is not private, it is a different landlord.
A machine. The sovereign estate: land, power and shell, delivered as community-owned compute. That is the billion-dollar product, and it exists because the only durable version of "own your AI" ends with you owning the thing that runs it.
What it costs, in plain numbers
| | | |---|---| | The address | free | | The mailbox, held for you | $12 a year | | Metered cloud inference | published per-million-token rates in micros | | The workstation | from $8,999 | | The estate | $1,000,000,000 |
We publish the whole ladder because a price you have to ask about is a price that changes based on who is asking.
What we will not do
We will not finance the free lane by turning you into the product. That is the entire business model we exist to replace, and an advertising lane that watched you would not be a new revenue stream — it would be the strongest possible argument that the sovereignty story is marketing. So the ad rules are written down before there is anything to sell: context, never people. No third-party code. A published price, not an auction. And the member whose surface carries a placement gets paid.
We will not pretend the model thinks. It predicts text. It is often right and sometimes confidently wrong, and the governance document says so in its second point, not in a footnote.
And we will not tell you it is perfect. Nobody can promise your mail lands in a specific inbox — the receiving system decides that. What we can promise is that we will remove every mechanical reason to be judged bulk, and tell you what we measured.
The thing you can hold us to
The rules are public and they are amendable. If an autonomous decision affects you, you can contest it and get human review. If you think a rule is wrong, any user may propose an amendment, and it gets a version and a date. That is not a marketing line; it is a paragraph in a document you can read, with a number.
Start with one tap: claim your name. Then email it something you actually need. If it is good, tell somebody. If it is wrong, tell us, and we will fix the copy and the code in that order.