What you are buying, and where it stops.

Plain English, and specific where it costs us to be. The usage limits below are not a policy someone wrote. They are the constants the inference gateway enforces on every request, and they are listed here because finding out about a cap by hitting it is a bad way to learn your software has one.

Who we are

Kato is operated by A10X (Armando Alvarez). Contact for anything on this page, including billing: iam@armando.mx. These terms govern the Kato macOS app and the Kato Pro subscription. By installing the app or subscribing, you agree to them.

The app, and what a subscription adds

Kato is a macOS app that reads on-screen text and, with permission, transcribes calls on-device, so it can keep track of what you have committed to. What it reads, stores and sends is described on the privacy page, which is the authoritative document for data handling.

Kato Pro is hosted inference. The judgment layer runs on a large language model, and a subscription means we supply and pay for that model access through our own gateway, so you never configure an API key. You can instead bring your own Anthropic key and turn on Use my own Anthropic key instead of Kato Pro in Kato's Settings menu. The subscription stays, because Kato stays paused without a license. You then also pay Anthropic for those calls, and the limits in the next section don't apply to them. The app keeps its own limit on your key: 400 calls a day.

The usage limits, stated exactly

Every Kato Pro request passes through our gateway, which enforces the following. These are per licence, on UTC days and UTC calendar months.

• 300 model calls per day.
• 800,000 input tokens per day and 4,000,000 input tokens per calendar month.
• 400,000 output tokens per day.
• 20 calls per minute, a burst limit that clears after sixty seconds.
• 3 devices in a rolling 30-day window. Devices already in the window keep working; a new fourth device is what gets refused.
• Model and request shape: the gateway serves claude-sonnet-5 and claude-haiku-4-5, clamps max_tokens to 4096, and rejects request bodies over 256 KB.

One more, shared by everyone: if all licences together pass 50,000,000 input tokens in a UTC day, the gateway pauses until we turn it back on.

What actually happens when you reach one. The cloud-backed features pause until the limit resets: daily limits at the next UTC midnight, the monthly one on the first of the next month. The app doesn't yet tell you which limit it was, so email us and we'll look it up. It does not silently downgrade you to a weaker model, and it does not queue the work. Everything that does not need the model keeps running, and a pause deletes nothing by itself. A call that ends during a pause stays unread: no summary, no promises, and Kato doesn't read it again on its own. Retry on its card reads it after the limit resets, only for your latest call and while the app stays open. The 7-day transcript sweep doesn't wait for that, so a long pause can empty a call's transcript before anything read it, unless you've turned the sweep off as /privacy explains.

For scale, from the usage log on the Mac of the person who builds Kato, which doesn't separate hosted calls from his own key: in September 2026 he used 5,870,489 input tokens, passing 4,000,000 on September 24, and reached 300 calls on 3 of 29 logged days. In October, as of the 31st, he used 10,638,619, passing 4,000,000 on October 14, and reached 300 calls on 7 of 29 logged days. He never reached 800,000 input tokens in a day. On the hosted brain that volume pauses the cloud-backed features from mid-month until the 1st: about 17 days in October. To read your own number, run kato tokens ~/kato --days 31 and add its input and cache-read columns, because the gateway counts cached input too. If your work looks like that, you should know before you pay. We may raise these limits; we will not lower them without telling subscribers first.

Price, trial, renewal and cancellation

The price and the trial length are whatever the checkout page showed you, and the billing portal shows your plan and its next charge. We deliberately do not restate them here, because a number typed onto a page is a number that eventually disagrees with the one you were actually charged.

Billing is handled by Stripe; we never see or store your card details. The subscription renews automatically until cancelled. Cancel yourself at any time from the billing portal (the link is in your welcome email), and access continues to the end of the period you already paid for. If a trial is running, cancelling before it ends means you are not charged.

Refunds. If Kato did not do what this site said it would, email us and we will refund you. We would rather return the money than argue about it. Any refund, whole or partial, ends your access to the hosted brain, and we cancel the subscription by hand when we refund you.

What we do not promise

• Kato's output can be wrong. It is a language model reading your screen. It will occasionally mis-transcribe a call, attribute a commitment to the wrong person, or invent detail. Everything it produces is a draft for you to check, and you remain responsible for what you send, sign or act on.
• No guaranteed uptime. The gateway depends on providers we do not control. We do not offer an SLA, and outages are not grounds for anything beyond a refund if you ask.
• The underlying models will change. We move to newer models as they ship, which can change how Kato words things.
• Not professional advice. Kato is not a lawyer, accountant or doctor, and reading your screen does not make its suggestions advice.

Telling the others on a call is on you. Kato listens on any call in a meeting app or a browser once your mic is in use, and it doesn't tell the others. Telling them, and following the recording laws where you and they are, is your responsibility.

To the extent the law allows, our total liability for any claim relating to Kato is limited to what you paid us in the twelve months before it arose. Nothing here limits liability that cannot legally be limited.

Fair use

The licence is for you, on your own Macs. Do not resell access to the gateway, share a licence across an organisation, or route unrelated traffic through it. If we see a licence being used as a general-purpose API key, we will contact you before doing anything about it.

Your data

Your journal and memory live on your Mac; we do not hold a server-side copy. Kato empties a call's transcript after 7 days unless you change that setting, and keeps only the newest 50 calls, while the copy sent to the brain stays encrypted on your Mac. Sealing a client erases its transcripts and saved copies for good, Time Machine copies aside, and its summaries and promises stay. The privacy page has the full list of each. The privacy page sets out exactly what leaves the device, when, and how to verify it with the app's own egress ledger. If you cancel, your data stays where it always was, on your Mac, and you can export it with kato export ~/kato.

Changes

If we change these terms in a way that affects what you are paying for, we will email subscribers before it takes effect. Everything else gets a new date at the bottom of this page.

Governing law

These terms are governed by the laws of Mexico, and any dispute belongs to the courts of Mexico City, after you have emailed us and given us a chance to fix it.

Last updated October 31, 2026. Every limit on this page is the constant the gateway enforces at this date. If this page and the software ever disagree, that is a bug. Tell us and we will fix whichever one is wrong.