Your files, and exactly what leaves your Mac

You're reading a book in Resage on your Mac, you highlight a passage, and you ask: "Why does the author contradict herself on page 40?" The app answers, grounded in the text. But what just happened to your book? Did the whole file go up to a server? Did the AI just get trained on your marginalia?

The short answer: no. The whole file never leaves your Mac. Only the few passages needed to answer the question you asked are ever sent, and they are used only to answer it. But "few passages" is doing a lot of work, and you deserve the precise boundary — what goes, what stays, and where the line is drawn.

What stays on your Mac

Everything that makes reading work — the reading itself, search, the read-through that turns a scanned page into searchable text, and the text recognition that does it — runs on your Mac. That last one is worth spelling out, because it's the part most apps gloss over.

When you import a scanned PDF, Resage needs to recognize the text so you can search it and discuss it. That recognition runs on Apple's on-device Vision framework. The page image never leaves your Mac to be read — the recognition happens on your machine, before anything is sent anywhere. Many apps say "on-device" and mean "we send it to our server and then delete it." We mean the pixels stay put.

Search, too, is local. The index of your book lives on your Mac. When you search for a term, the matches are found on your machine, not by a server.

What leaves your Mac

When you ask a question, Resage needs to send something to the model that answers it. That something is the few passages needed to answer your question. Here's how it works, concretely:

1. You select a passage or ask about the page you're on. Resage takes that passage — a few sentences, maybe a paragraph — and, if needed, retrieves a few relevant passages from the rest of the book using local search.
2. Those passages are sent to the model along with your question. The model sees only that context, not the rest of your file.
3. The answer comes back, and the passages are used only to answer that question. We keep a record of the question you asked and a pointer to the part of the book it was about — that record is how the learner model gets built, and it is covered in the section below. What we do not keep is a copy of the passages themselves, and your content is never added to a training set.

There's one exception worth knowing, because it's a case where a bit more goes. If your question is about a figure, diagram, or table that has to be seen rather than read — a chart, a map, a handwritten note — Resage can send an image of that single page so the model can look at it. That's it. One page, not the whole book.

What never leaves your Mac

The whole file. Ever. We don't upload your book to make it searchable, we don't upload it to build a profile, we don't upload it to train a model. Your content is never used for training. Period.

This is a hard boundary, and it's the reason we can say "your files never leave your device" without a footnote the size of a EULA. The only thing that leaves is the minimum context needed to answer the question you asked.

The learner model is different — and that's intentional

Resage keeps a private model of how you learn — what you understand, where you got stuck, which explanations actually landed. That model is what makes day 30 not feel like day 1. It's a core part of the product, and it's not a memory gimmick.

That learner model lives on one page you can open, read, and edit at any time. Facts it is confident about persist; guesses fade if they aren't borne out. You can delete individual items, reset it entirely, or turn memory off.

But here's the honest caveat: unlike your files, that profile is stored on Resage's servers, tied to your account. We say this plainly because it's the one part of the system that is not local. It's made of short written notes about your reading — not copies of your book's text. It points back to the discussion a note came from, rather than keeping the passages themselves. And you can read it, edit it, or switch it off at any time in Settings.

Why do we do this? Because the learner model is meant to follow you across devices. iPhone is coming next, and when you pick up your phone to continue a book, the model should be there with you — not stuck on your Mac. That's a trade-off we made deliberately, and we want you to know about it.

The precise boundary, in one paragraph

Here's the whole thing, in terms you can hold us to:

• Stays on your Mac: your files (always), search, text recognition, and the read-through that makes scanned books searchable. All of it runs locally.
• Leaves your Mac, only when you ask a question: the few passages needed to answer that question, used only to answer it. The whole file never goes.
• Leaves your Mac, only if your question is about something visual: a single page image, so the model can look at a figure or table.
• Never leaves, ever: your content as training data. We don't train on what you read.
• Stored on our servers, but not your files: the learner model — short notes about your reading, editable and deletable, tied to your account so it follows you.

That's the boundary. No vague promises, no fine print that swallows the headline. If you read a book in Resage, the book stays on your Mac. The AI reads alongside you, but it reads what you ask it to read — not everything you own.

Why this matters for a reading app

A general-purpose assistant that you paste a PDF into is a different beast. You're handing it the whole document, and you have no idea what it does with it after the chat closes. Resage is a reader first: you read the actual page in its original layout, and the AI stays in a bar at the bottom until you call it. That's not just a design choice — it's a privacy choice. The AI doesn't need to see your whole book to be useful. It needs to see the passage you're struggling with.

And because every answer is grounded in the actual text in front of you, the AI flags uncertainty instead of making things up. That matters most with classics and primary sources, where the model's general memory is most likely to be wrong. It says "the book says X" only when it can point to the passage. That's the same discipline as the privacy boundary: we tell you exactly what we're relying on.

The limits, stated plainly

We're not claiming to be more private than every other tool on the market. Some apps keep everything local, including the model. Resage doesn't — discussions need a connection, because the answer comes from a model in the cloud. That's a real trade-off, and we're not hiding it.

What we are claiming is that the boundary is precise, and that we've told you exactly where it is. Your files stay on your Mac. The only thing that goes is what's needed to answer the question you asked. And the one part of the system that isn't local — the learner model — is something you can read, edit, and delete.

That's the kind of privacy that survives scrutiny. Not because we make grand promises, but because we can tell you, byte by byte, what leaves your Mac and what doesn't.