Read
Box a chart, table or equation and ask AI about just that
What you are stuck on is rarely a whole document — it is one chart, one row of a table, one equation. Pointing at it beats describing it, which is why everyone screenshots. But a screenshot is only pixels: the numbers get guessed and the answer cannot point back to the page.
- Works in
- PDF · EPUB · Slides · Images · Notes
- Per question
- Up to 4 regions
- Runs on
- macOS (Apple silicon) · Windows
- Price
- Free app · AI plans from $10/mo, or your own key
- AI
- Included, or your own API key
A region is not a screenshot
Paste a picture into a chat and the model reads the axis as best it can, invents a plausible number when it cannot, and leaves you a paragraph you cannot check. A box drawn in MindMax carries four things, and each fixes one of those failures.
- The image, for what only an image can carry: layout, shapes, arrows, what is next to what.
- The verbatim text underneath the box, taken from the document rather than from the picture. Numbers, formulas and table values follow that text. Where there is none — a scanned page, labels drawn into a diagram — the answer has to say it read the figure off the image and may have it wrong.
- Where it came from — the file and the page, chapter or slide — so the answer can cite the spot and take you back to it.
- The document itself, added to the question, so "and what does the paragraph above it say?" has something to read.
- 1The imageLayout, shapes, what sits next to what
- 2The text under the box“Accuracy rose 42.7% over baseline”
- 3Where it came frompaper.pdf · page 7
- 4The documentAdded to the question, for the follow-up
The same box, on everything you read
It is one action, available wherever a document is open — not a separate feature per file type. What the region knows about itself adapts to the format.
PDFs
The classic case: a chart, a table whose layout the extractor mangled, a figure in a paper outside your field. The answer cites the page and clicking it goes there.
EPUB and Kindle books
A diagram in the middle of chapter nine. The region remembers the chapter it belongs to, so the answer lands back in the right place in a 400-page book.
PowerPoint slides
Half of what a slide means is the arrangement. Box the part of the page you mean and the question knows which slide it is on.
Images and screenshots
One panel of a multi-part figure, one column of a screenshotted table, a legend too small to read. Works at any zoom level.
Your own notes
A table or a rendered diagram inside a Markdown file you wrote can be boxed and asked about like anything else.
Or skip the question and keep the thing itself
The table you cannot copy out of a PDF, the chart whose numbers you were about to retype, the equation you need in your own notes — box it and it leaves as ordinary Markdown.
- The text layer under the box is handed to the converter, so words and numbers come from the document rather than from reading pixels. That is what a screenshot of the same page cannot give you.
- The note is written beside the document it was cut from — chapter-seven.pdf gets chapter-seven - table.md, with the picture kept in a folder next to it.
- A table becomes a real Markdown table, a chart its data plus a short reading, an equation LaTeX, a dense page a faithful outline. You do not have to pick — the first menu item works out which it is.
- Nothing you already had is touched. A conversion only ever creates a new file.
The answer knows where it came from
A pasted screenshot is a dead end: once the conversation moves on, nothing connects the answer back to the document.
- Clicking the thumbnail in the conversation reopens the document at that page or chapter, with the passage highlighted.
- The region stays attached to the conversation, so a follow-up about the same figure does not start from nothing.
- Quotes in the answer are checked against the source text, and one that does not match is flagged at the end of the answer rather than passed off as a quote.
From a confusing figure to something you keep
The point of the box is rarely the answer on its own. It is that the explanation ends up next to the work.
Box it
Drag around the chart, the equation, the row of the table, the panel — at whatever zoom level you are reading at. The box freezes and waits; if it is slightly off, redraw it without starting over.
Ask in your own words
"What is this actually measuring?" is a fine question when the thing you are pointing at is attached to the question.
Check the answer against the page
Quotes link back to the passage. Where there was no text under the box, the answer says so rather than asserting a number with a straight face.
Keep the explanation
Save the region and the answer it produced into your daily note in one step. The picture is copied into your own folder, so deleting the conversation later does not break the note.
Or keep the thing itself
Convert the region instead and the table, the chart data or the equation lands as a Markdown file beside the document — which is what you wanted when the answer was never the point.
Or take it somewhere else
Copy it as an image at no cost, or drop the region onto a canvas as an item of its own to expand and compare.
One part of a loop: read, think, write
Your notes, and the two hundred PDFs you read to write them, sit in one folder — where a single AI works across both. Tools you would otherwise stitch together each hold one half of that.
- ReadThis pageBox the figure, table or equation you are stuck on in whatever you are reading, and ask about just that.
- ThinkDrop the region onto a canvas as a card of its own, and compare it with a figure from another paper.
- WriteThe explanation goes into your daily note and a converted table lands as Markdown beside the document — ready to quote.
Where it gets hard
- A region with no text layer is only a picture
Scanned pages and pure graphics have nothing underneath to check the image against. The answer is marked as read off the image — which is honest, not accurate. When you need the exact words on a scanned page, recognition is the right tool and it is a separate, auditable step.
- Conversion is transcription with rules, not magic
It writes what is visible and what the text layer says, marks unreadable marks as illegible and labels anything inferred as inferred. A chart whose numbers were never printed anywhere gives you estimates read off the axes, marked as estimates — check them before the table goes into your own argument.
- Handwriting and tiny axis labels are where it goes wrong
And it goes wrong confidently. Scribbled notes, crowded tables and dense charts are worth checking against the original before anything gets built on top of the answer.
- A picture is a picture, and pictures cost credits
Asking about a region spends credits in a way that reading and highlighting your own files does not. Four regions in one question is four pictures.
- Some model providers cannot see images at all
If you have connected your own endpoint that does not accept pictures, the region is converted to text first and the question is answered from that — which is weaker, and the answer will be working from a transcription rather than from the figure.
- The box is not a crop tool
It captures what is on screen in the reader. A figure that spans a page break is two regions, not one.
Common questions
How is this different from screenshotting a chart into a chat window?
A screenshot is only pixels. A region drawn in MindMax carries three more things: the verbatim text sitting under the box, the place it came from — file, page, chapter — and the document itself, added to the question so follow-ups still work. The answer can therefore quote the source and link back to the page rather than describing a picture.
Will it read numbers off the image?
Not when there is text under the box. Where a region has a real text layer, the model is told that numbers, formulas and table values follow the text and the image is only for layout, shapes and spatial relationships. Where there is no text layer — a scanned page, a pure graphic — it must say that a figure was read off the picture and may be wrong.
Which documents can I box a region in?
PDFs, EPUB and Kindle books, PowerPoint slides, images, and your own Markdown notes. It is the same action everywhere, and what the region knows about itself adapts to the format: a page number in a PDF, a chapter in a book, a slide number in a deck.
Can I keep the answer?
Yes. A region and the answer it produced can be saved into your daily note in one step. The picture is copied into your own folder at that point, so deleting the conversation later does not leave a broken image in the note.
Can I get a table out of a PDF without retyping it?
Box it and convert it. The table comes back as a real Markdown table with rows, columns, numbers and units in the original order, written to a file beside the document it came from. The text layer under the box is handed to the converter, so the words come from the document rather than from reading pixels. Charts convert to their data plus a short reading, equations to LaTeX, and a dense page to a faithful outline.
Do I have to say what kind of thing I boxed?
No. The first item in the convert menu lets the model work out whether the region is a table, a chart, a formula, an outline or a set of notes. Choosing the kind yourself is there for when it guesses wrong.
Can I ask a follow-up about the same region?
Yes. The region stays attached to the conversation, so the next question does not start from nothing — and up to four regions can travel with a single question when you are comparing two figures.
See also
Where most boxes get drawn, and the annotation passes that mark up the rest of the page.
When the region is the whole file — and how a boxed table becomes a Markdown table.
Boxing a diagram inside a 400-page book and getting back to it afterwards.
The format that is half picture and half argument.
Where a region becomes an item of its own, to expand or compare against another.
The workflow where papers outside your field break on a figure rather than on prose.