Read
Search inside every PDF, book and note in your folder
Someone made that argument, somewhere in the forty things you read last month — a paper, a book, or a note you wrote about one of them. Your notes app searches the notes, your reader searches the one file that is open, and neither finds it. Here one search covers the whole folder, and a result opens on the sentence, not just the file.
- Searches
- PDF, EPUB & Kindle, Word, PowerPoint, Excel, notes
- Credits
- None for searching; only when the AI re-picks
- 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 search, every format, landing on the sentence
A list of file names is a second search you now have to do by hand, inside a four-hundred-page book. Here a result is a place.
- Papers, books, slides, spreadsheets and your own notes come back in one list — the notes next to the documents they are about.
- Click a result and the document opens there with the sentence highlighted: the page in a PDF, the chapter in an EPUB, the slide in a deck, the line in a note.
- Each file is one card — under relevant content, topped with a line on what the file is about — so forty hits in one book do not bury the single hit in your notes. Expand a passage in place when the excerpt is not enough to judge.
- A paragraph repeated through a document — a running header, a fixed opening line — is shown once, not once per page.
- PDFbehaviour-change-review.pdfHighHabits depend less on willpower than on the context in which a behaviour is repeated.p. 12
- EPUBA Book on Habits.epubMediumChange the room, and the cue that starts the habit is gone with it.Ch. 3
- Notehabits.mdMediumMy summary: context does the work, not resolve.§ Environment
You do not have to decide which search you need
Whether you want the exact words or something close depends on results you have not seen yet. So both run as you type, in two sections, and neither is a mode you switch on.
- Literal matches show every place the words you typed appear, highlighted. Relevant content ranks passages by how much of what you asked they cover — they need not contain your exact phrase, or every word, or keep the order — and marks each one High, Medium or Background.
- The shape of the query decides which section leads: a question or a sentence puts relevant content first, a single word or short phrase puts literal matches first.
- A section with nothing in it says so in one line, so an empty result is never mistaken for an empty folder.
- When a sentence-length query turns up nothing that clearly matches, the AI takes another pass on its own: it searches deeper, reads the candidates and keeps the ones that fit what you meant. The panel says it did this, and Undo brings the original results back. You can also ask for it with Let AI re-pick.
When you know exactly what you are looking for
Sometimes the job is not finding a passage but finding all of them — every mention of a term, a name, a figure.
- The results bar counts them: how many matches, in how many files, across the folder. Sort by best match, file name, or when a file was modified or created; match case applies to literal matches.
- Put a phrase in quotes, or turn on Literal matches only, and you get exact matches and nothing else.
- Narrow with operators, combined freely: path: for part of the folder, file: and tag:, line:() and section:() to require words on one line or in one section, and [key:value] for a note's properties. path: and tag: suggest as you type. If you have used Obsidian's search they will look familiar; OR, negation and regular expressions are not supported.
- Bookmark a search and it sits with your bookmarked files, one click from running again.
The rest of the search surface
Every format together
PDFs, EPUB and Kindle books, Word, PowerPoint, Excel and Markdown are searched by their text. An image joins once it has been read, by its description.
Supported formats →The same ruler the AI uses
The High and Medium labels are the same judgement the AI applies when it searches your folder. Files you have hidden from the AI still show up here, labelled, because hiding them was about the AI, not about you.
New files, found on the next search
There is no indexing step to wait for and nothing to keep up to date. Drop a PDF into the folder or edit a note, and the next search sees it.
Filters that stay out of the way
Show only highly relevant results, sort either section, or open more options. Defaults are what most searches need, so you rarely touch them.
Coming from your notes app's search and a reader's Find
Most people look things up one tool at a time. Here is what changes — and the rows where a system-wide search is still the better tool.
| Topic | A notes app's search, plus Find in a PDF readerHow most people look things up today | In MindMax |
|---|---|---|
| Where it looks | The notes app searches notes; the reader searches the one file that is open. | The whole folder at once — notes, papers, books, slides and spreadsheets in one list. |
| What a result is | A file name, or a jump to the next match you then page through. | The passage, with the sentence highlighted once you open it. |
| When you only remember the idea | Find needs the exact characters. | Relevant content ranks passages by what they share with your words, and the AI can re-pick by meaning. |
| Counting mentions | One count for the open file. | Matches and files counted across the folder, narrowed with path: or quotes. |
| Speed on a huge library | A system-wide search keeps an index of your whole disk and answers instantly. | Text is read as you search. Thousands of large files take longer, and past a certain size only part of the folder is searched in depth. |
| Beyond the folder | A system-wide search also covers mail, apps and every other folder. | Only the folder you opened. |
Where it gets hard
- Ranked by wording, not by a vector index
No embeddings sit behind this panel. Relevant content ranks passages by the words they share with your query, so a question phrased very differently from your files, or asked in another language, can come back thin. That is where the AI re-pick earns its few seconds — but it chooses among passages the wording turned up, so text that shares no words with your question stays out of reach. Try the words your source would use.
- A large library is searched in part, and says so
Past a certain size, only part of the folder is examined in depth on any one query, and the results say so. The honest reading of an empty result there is "not found in what was searched"; a more specific wording or a path: clause is the fix.
- Scans are invisible until recognised
A PDF with no text layer cannot be searched at all. The panel tells you how many files are in that state, so a silence is explained rather than misleading, but the recognition step is one you start.
- Not instant on thousands of files
There is no index you build or maintain, which is what removes the waiting and the staleness. The cost lands here: the text is read as you search, so a very large folder answers more slowly than a lookup in a prebuilt database, and the first search after opening it is the slowest.
Common questions
Does it search inside PDFs and EPUBs, or only the file names?
Inside them. PDFs, EPUB and Kindle books, Word, PowerPoint, Excel and Markdown are all searched by their text, and a result opens the document at the passage with the sentence highlighted, rather than just naming the file.
Do I have to build an index first?
No. There is no indexing step to wait for and nothing to keep up to date — add a file or edit a note and the next search sees it. The trade is speed: a very large folder is read as you search, which is slower than a database lookup.
Is this semantic search?
Not in the vector-database sense. Relevant content ranks passages by the words they share with your query, without requiring the exact phrase. Judging by meaning is the AI re-pick: it reads the candidates and keeps the ones that fit, and runs on its own when a sentence-length query finds nothing that clearly matches. No embeddings are involved either way.
How is searching different from asking the AI?
Search hands you the passages and you judge them; asking gets you an answer with citations back to the page. Underneath it is the same search — when the AI looks through your folder it uses the same ranking you see here.
Can I narrow a search with operators like path: and tag:?
Yes: path:, file:, tag:, line:(), section:() and [key:value], all combined with AND, plus quoted phrases for exact matches. A query with an operator or quotes shows literal matches only. OR, negation and regular expressions are not supported.
Why can it not find a file I know is there?
Usually because that file has no readable text yet — a scan awaiting recognition, or something still being read in — and the panel says how many files are in that state. In a very large library, the other reason is that only part of it was searched in depth, which the results also say.
Does searching cost credits?
No. Searching and opening your own files is local work. Only the AI re-pick spends credits, including when it runs on its own, because that is a model reading the results.
See also
When you want the answer rather than the passages — same material, with citations back to the page.
Why there is no index to build, and what that trade costs.
What to do about the files search cannot see yet.
Why retrieval is the fallback here rather than the default.
What you give up without embeddings, and what you get back.
Exactly which file types are read and searched.
Operators, exact matches, the AI re-pick and why a file does not come up.