Skip to main content

Reading and annotating

This guide covers the reading surface: opening documents, annotating them, and asking questions that come back with something you can check.

Before you start​

You need a folder open in MindMax with at least one document in it. If you have not done that yet, see installing MindMax.

Opening a document​

Documents appear in the file tree as they are found in the folder. Opening one shows the document itself — not a text extraction of it — so pages, figures and layout are what you would see in any other reader.

For formats that are not page-based, like Markdown and Word, there is a text surface with the same annotation and question tools.

Annotating​

Select a passage to highlight it or comment on it. Annotations are anchored to the passage, so they survive closing and reopening the file. They are kept in the hidden .mindmax folder inside your knowledge base — on your disk, not on a server — and the document itself is never modified.

Highlights and comments on a PDF, listed in the Comments panel.

Use annotations for the things you want to find again — not as a substitute for notes. A highlight tells you something here mattered; a comment tells you why, and the why is what you will actually need in a month.

All of them are listed in the Comments panel, for this document or across every document in the knowledge base. From there the AI can answer your questions in the thread, check a doubtful claim, or collect the annotations into a note — see reviewing and revising with AI for what the panel can do.

Having the whole document annotated​

Instead of annotating by hand, you can pick a lens — guide, critique or study — and have the AI work through the whole document, leaving notes beside the passages and a write-up at the end. To review a draft for its author, start a review instead; the comments you keep can go back as a Word file. See having a document annotated by AI.

Asking about a region​

Some things on a page are not text: charts, tables with broken layout, diagrams, scanned figures. Drag a box around the region and ask about just that. The question carries the image along with the surrounding context, so you do not have to describe what you are looking at.

A boxed figure goes into the question with its document and page attached.

This is usually the right tool when a paper stops making sense at a figure.

It is the same box in every reader — PDFs, EPUB and Kindle books, PowerPoint slides, images, and your own Markdown notes — and the region knows where it came from, so the answer can take you back to the page, chapter or slide it was drawn on.

Two things are worth knowing about how it behaves:

  • The text under the box goes with the question. Where the region has a real text layer, numbers, formulas and table values are taken from that text rather than read off the picture. Where there is none — a scanned page, a pure graphic — the answer says that a figure was read off the image and may be wrong.
  • You can keep the result. A region and the answer it produced can be saved into your daily note in one step. The picture is copied into your folder at that point, so deleting the conversation later does not leave a broken image behind.

Turning a region into a note​

When you release the box it freezes, with a toolbar: Ask AI, To note, Copy and Reselect. To note turns the region into ordinary Markdown saved beside the document it was cut from — chapter-seven.pdf becomes chapter-seven - table.md, with the picture kept in a folder next to it:

The region toolbar and its To note menu.
You boxedYou get
A tableA real Markdown table — rows, columns, numbers and units in the original order
A chartThe plotted data as a table, axis estimates marked, plus a short reading
An equationLaTeX, with the equation number kept
A dense page or slideA faithful outline, key sentences verbatim
A diagram or handwritingStructured notes, arrows spelled out as relations

The first menu item works out which of those it is, so you do not have to choose. The text layer under the box is given to the converter, which is why a table boxed in a PDF comes out with the document's own numbers rather than numbers read off the picture — and why this beats converting a screenshot of the same page.

Converting only ever creates a new file; nothing you already had is edited. Copy takes the region as an image and spends nothing.

Up to four regions can travel with a single question, which is how you compare two figures in one go.

Asking about the document​

Select a passage and choose Ask AI or Explain, or ask about the document in the assistant panel beside it. The answer carries citation marks: hover one to read the quoted sentence with the text around it, click it to move the reader there with the sentence highlighted.

Two habits worth forming:

  • Check the claims that matter, not all of them. Pick the two your work will rest on and click through. That is 90 seconds and it changes what the answer is worth.
  • Ask what the document does not establish. It tends to surface the caveats that a summary smooths away.

Scanned documents​

A scanned PDF has no text layer, so it has to be recognised before it can be searched, quoted or annotated. This is something you start deliberately — see recognising scanned PDFs.

Reading as text​

A PDF can also be read as one clean reflowed column with page markers — useful on a small screen or a dense two-column layout. Highlights, comments and questions work the same way there, and anything you mark lands back on the printed page.

Common problems​

A quote reads oddly or has words running together. Usually a multi-column or figure-heavy page. Click through to the source — the page will settle it. Layouts like this are the hardest case for any extractor.

Annotations seem to be in the wrong place after an update. Report it. Anchoring is meant to survive re-extraction, so this is a bug rather than expected behaviour.

A question about a long book gives a shallow answer. Try asking about a specific chapter or section first. Narrowing the question usually beats rephrasing it.

Next​

Last updated on