You wrote it down. Now find it.

Search everything you've written — except the tablet

There are months of thinking on that tablet. Meeting notes, half-worked plans, a page you kept extending downward until it was three screens tall. Then you need one specific thing — the number a client quoted, the idea you sketched in March — and there is no way to search ink.

So you try the tablet's convert-to-text. On careful print it does okay. On the notes you actually take, fast and cursive and abbreviated, it hands back word salad, because it reads character by character and your handwriting isn't characters, it's writing. reMarkable's own converter lands around 75 to 85 percent on real notes. The missing chunk is usually the words that mattered.

Meanwhile your AI can summarize anything you paste into it. Except the one place you do your actual thinking. That's the gap remarkdown closes.

Geometry-aware OCR

Most OCR photographs the whole page and hopes. We start with the geometry. Before reading a single word, we look at how your page is laid out: where the lines sit, where the columns run, where the blank space falls between them. We cut the page into regions along that whitespace, then read each region with an AI vision model that uses the surrounding words to make sense of the messy ones.

• Cuts follow your whitespace. Each cut snaps to the whitest row it can find, so it tends to land in the gap between lines rather than through your writing.
• Side by side stays separate. A diagram next to a paragraph, or two columns of notes, become their own regions instead of bleeding into one.
• Every region is read in context. The vision model works out a scribbled word from the words around it, like you would. On cursive and fast handwriting, that beats the tablet's character-by-character conversion.
• Edits re-read only what changed. Each region is fingerprinted by its content. Change one paragraph and only that region gets transcribed again — and only new regions spend credits.
• Sketches aren't forced into words. A diagram becomes a short description in the text, and the real drawing is always one tap away.
• Your ink is untouched. When your AI opens a handwritten page, it still gets the actual image.

Want the full picture of what your AI can see and do? How it works →

Use the whole page. We keep up.

reMarkable lets a page be as big as your thinking. Pull the canvas down and keep writing. Sprawl a mind-map sideways. It's one of the best things about the tablet, and people use it hard.

It's also where ordinary OCR gives up. To read a giant page, it renders the whole thing as one image, so the bigger your canvas, the smaller your words, until the model is squinting at a blur.

Geometry-aware OCR doesn't shrink your page. It walks it. Every region is read at full clarity, however far the canvas goes, so a sprawling page transcribes as cleanly as a small one. Use the canvas the way reMarkable meant you to — we keep up.

Better, not magic

A vision model reading one clean region at a time beats the on-device converter on real handwriting, and it wins biggest where the tablet struggles most: cursive, speed, abbreviation. Region-by-region reading also means a dense page never gets flattened into a single blurry image.

It isn't magic, and we won't pretend it is. A genuinely illegible word stays a guess. A sketch is described, not redrawn. Equations and complex tables get no special treatment yet. If a word was unreadable on paper, it will be unreadable to the model too — we'd rather tell you that than oversell it.

Nothing gets left behind

Transcription runs in the background. A page that fails the first time isn't abandoned — we retry it on a schedule until it lands. Nothing needs babysitting.

What it costs

Transcription is part of Pro ($9/month) with Document Reading turned on. It runs on credits, and every Pro account starts with 250 to transcribe with. You only spend credits on new regions of a page — edit a page later and only the changed region costs anything. Reading your notes and full-text search don't spend credits; only transcription does. It runs on your recent notes by default, and you can widen the window or switch it off in Preferences. If credits run out, it pauses rather than surprising you.

Straight answer on your data

For every paired user we keep an index of document names and folders, so you can browse and search titles. We do not store the contents of your notes by default.

The text and page images inside your documents are stored only on Pro with Document Reading on, and only for the one folder you connect — that's what makes transcription and search possible. It's covered by the Document Reading terms, and it's deleted the moment you turn reading off or downgrade. Nothing outside that folder is ever read or stored.

Common questions

Why is reMarkable's built-in handwriting conversion inaccurate?

The on-device convert-to-text is tuned for tidy print. On cursive, cramped, or fast handwriting it drops words and guesses wrong, because it reads character by character and there's no context to fall back on. In practice people see roughly 75 to 85 percent accuracy on real notes, which is enough to be frustrating when you're searching for the one line that mattered.

How does remarkdown's OCR work?

When you connect a folder, we render each handwritten page as an image and look at its geometry first: where the blank space runs between lines and columns. We cut the page into regions along that whitespace, with each cut snapping to the whitest row so it tends to fall in a line-gap rather than through your writing. Each region then goes to a general-purpose AI vision model, which reads it in context, using the surrounding words to work out the messy ones. It's not a stroke-based handwriting engine and it doesn't run on the tablet; it reads a rendered image of your page. Regions are fingerprinted by content, so editing a page later only re-reads what changed, and only new regions use credits. Your original ink is never modified.

Does it change my original handwriting?

No. Your ink stays exactly as you drew it. When your AI reads a handwritten page it still gets the original image; the transcription only exists to make the page searchable.

What about my big, extended canvas pages?

Those are exactly where geometry-aware OCR helps most. Ordinary OCR renders a giant page as one image, so the bigger your canvas the smaller your words, until the model is squinting at a blur. We walk the page region by region instead, so a page you extended down three screens transcribes as cleanly as a small one.

Is it perfect?

No, and we won't pretend it is. A vision model reads messy writing far better than the on-device converter, but truly illegible words are still a guess, and a sketch is described rather than redrawn. It's good enough to make search work and to let your AI understand what a page says.

What does it cost?

Transcription is part of Pro ($9/month) with Document Reading turned on. It runs on credits, and every Pro account starts with 250 to transcribe with. You only spend credits on new regions of a page — edit a page later and only the changed region costs anything. Reading your notes and full-text search don't spend credits; only transcription does.

Do you store the transcribed text?

For the one folder you connect on Pro, yes — the page images and their transcribed text are stored so search and reading work, covered by the Document Reading terms. For everyone else we only store an index of document names and folders, never the contents. Turn reading off and the synced copy is deleted.