Read
Chat with your documents — every format, a whole folder at once
Most tools answer from a handful of retrieved fragments and hope the right one was in there. This one reads short documents whole and navigates long ones by their structure — PDFs, Word files, slides, spreadsheets and ebooks alike — and every answer quotes the passage it rests on.
- Reads
- PDF · Word · PowerPoint · Excel · EPUB
- Answers
- Quotes matched to the source
- 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
One question, the whole pile
The questions worth asking rarely live in one file. They span a reading list, a case file, a term's worth of material — in whatever formats it arrived.
- Mention a folder in your question and every document in it becomes the subject — the paper, the Word draft, the deck and the ebook in one answer.
- It is worked through the way you would: a one-line gist of each document, the ones the question needs picked by what they are about, their chapter maps opened, and the relevant sections read in full — not the ten most similar fragments pulled from the lot.
- Every claim names the document it came from, so where two sources disagree you see both sides and which file said what.
- Under the answer, a list of what it actually read — so “did it even look at my file?” has an answer.
- PDFanalyst-report-2025.pdfSizes the market top-down, from shipments
- DOCXstrategy-memo.docxBuilds a bottom-up estimate from customers
- PPTXkickoff-deck.pptxQuotes one headline figure, no method
- XLSXpricing-2024.xlsxPrice list by tier
- EPUBcrossing-the-chasm.epubAdoption stages, in general
Every format you get sent — and a quote that lands on the spot
A citation is only worth something if you can check it in a click. Each format is quoted at the finest place it actually has.
- PDFs quote the page and open on it with the sentence highlighted; slides quote the slide number; EPUB and Kindle books land on the passage in its chapter; spreadsheets land on the place in the sheet.
- Word documents open in a clean reading view with their headings and are asked about, annotated and quoted like anything else. A quote lands on its place in the document — there is no page number, because the Word layout is not reproduced.
- Selecting a passage to explain, and boxing part of a page to ask about it, work the same way across PDFs, Word files, slides and books.
- Nothing is imported or converted in your folder. The .docx you were sent is the file you read.
A document is read, not retrieved from
The usual approach cuts every file into fragments and answers from the few that look most like the question. It fails quietly when the answer is spread across a document or two sources disagree — the pieces that would have told you were never retrieved.
- Under a stated budget — roughly 40,000 Chinese characters or 25,000 English words, which covers most papers, reports and chapters — the whole text goes into context. Nothing is chunked, nothing is ranked, nothing is left out.
- Above it, the AI reads the contents first, reasons about which chapters the question needs, and reads those sections in full — going back for another if the first did not settle it, the way you would use a table of contents.
- @ several files at once and the budget is handed out smallest first, so a three-page brief next to a four-hundred-page book is still read in full rather than crowded out by it.
- That answers the questions retrieval cannot. “What is the main argument?” “Does the conclusion follow from the data?” “What does the report never mention?” No fragment matches those words, so keyword and embedding search have nothing to return. Reading does.
How you get at the material
You set what it draws on
The whole folder is in reach by default. @ a document and it is read; switch to “only these” and nothing else is consulted. Pin documents to a conversation and they stay in scope turn after turn.
Quotes matched to the source
Each quote is aligned to the exact wording in the document it cites. One that cannot be found there is flagged in the answer, not passed off as checked.
Ask about a selection
Select a passage — or drag a box around a chart or a table — and ask about just that.
Region questions →Follow-ups do not re-pay for the paper
A document that stays in context between turns is cached, so the third question about the same report costs far less than the first.
Scans, once recognised
A scanned PDF has no text to read until it is recognised. Recognise it once and it is asked about, quoted and searched like any other.
Scanned PDFs →Answers become notes
Keep an answer and it lands in your folder as Markdown, with its citations intact.
Coming from a chat-with-documents website: what changes, what stays
Most people who land here have uploaded a few files to a website and asked them something. Here is what doing that on your own folder changes — including the rows where the website wins.
| Topic | A chat-with-documents websiteUpload files, ask in a chat window | In MindMax |
|---|---|---|
| Your documents | Uploaded, one at a time or in a batch. | The folder they already live in, opened in place. Nothing uploaded to read or search it; when you ask, the text the AI needs goes to the model. |
| Formats | Varies by tool; PDF first, often Word and others. | PDF, Word, PowerPoint, Excel, EPUB and Kindle, Markdown and text — and scans once you have them recognised. |
| What an answer draws on | Varies by tool, and it is not always visible which parts of a file were used. | Whole documents when they fit, chapters by structure when they do not. Each answer lists what it read. |
| Many documents at once | Often a collection you search across. | A folder you mention: the documents the question needs are picked by what they are about and read by their structure, and each claim names its document. |
| Checking an answer | Often a reference back to the spot. | A verbatim quote aligned to the source, opening on the page, the slide or the passage; one that does not match is flagged. |
| Keeping the work | Chat history in the service. | Answers saved as Markdown notes with their citations; annotations as files beside the documents. |
| Getting started | Open a browser tab, on any device. | A desktop app for macOS and Windows. |
| Working as a team | Some offer shared workspaces. | One person, one folder. Share the folder through whatever already syncs it. |
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 pageAsk the whole folder — every format, read whole where it fits — and check each answer against its quote.
- ThinkTake a question that needs thinking through to a canvas, where each branch comes back with its reasons.
- WriteKeep an answer as a note, or draft from the same folder in Markdown, and quotes keep their source.
Where it gets hard
- Very large corpora still rely on search
A thousand documents will not fit in context, and no budget makes them. At that scale the honest description is "structure and search, done carefully", and it can still miss things — narrow the folder when the answer matters.
- A folder brings in up to 200 documents
Mentioning a folder takes in its first 200 readable documents, subfolders included. For a bigger library, mention the subfolder that matters, or leave it to the default whole-folder scope and its search.
- Word and Excel are read, not laid out
A Word file comes through as its text, headings and tables, not its page layout, so a quote lands on a place in the document rather than a page number. The old .doc and .xls formats are not read — save them as .docx or .xlsx first.
- Questions with no anchor in the text
Ask something the documents simply do not address and you get an answer about what they do address. Checking the quotes is how you catch this; there is no way to skip that step.
- Scanned material needs recognition first
A PDF with no text layer is invisible to search until you recognise it, and recognition quality sets the ceiling.
Common questions
Does it chunk my documents and search the pieces?
Not when it does not have to. A short document goes into context whole, so nothing is retrieved and nothing is missed. A long one is navigated by its chapters, and a folder by what each document is about; fragment search is used to find wording, not as the default path.
Are my documents uploaded?
No. They stay in the folder they already live in, and opening, reading and searching them happens on your machine. When you ask a question, the text the AI needs for that answer is sent to the model.
Can I chat with a Word document?
Yes. A .docx opens in the reading view with its headings, and you can ask about it, annotate it, select a passage to explain, or ask about it together with the rest of the folder. Quotes land on their place in the document; the Word page layout itself is not reproduced.
Which formats can I ask about?
PDF, Word (.docx), PowerPoint (.pptx), Excel (.xlsx), EPUB and DRM-free Kindle files, Markdown and plain text — and scanned PDFs and images once they have been recognised.
Can I ask about a whole folder at once?
Yes. Mention the folder in your question and its documents — up to 200 — become the subject of that question. The ones the question needs are picked by what they are about and read by their structure, and each claim names the document it came from.
Does it work with documents in Chinese or other languages?
Yes. Documents are read in the language they are written in, and the answer comes in the language you ask in, with quotes kept in the original.
How do I know the answer is not invented?
Every claim carries the verbatim quote it rests on. Each quote is aligned to the exact wording in its source, one that cannot be found there is flagged in the answer, and clicking a quote takes you to the passage.
Does asking cost credits?
Opening, reading and searching your own files never spend credits. Answering a question is AI work and does — or runs on your own API key.
See also
Where the answers point back to the page, and where you mark up what matters.
Book-scale questions, navigated by chapter rather than by fragment.
A card per page for what went unsaid, and a review on the slides.
When you want to find where something is rather than ask what it says.
Why there is nothing to upload or import before you can ask.
From asking the pile to a Word brief you can send, citations intact.