Verbatae

Search a folder of documents by what they say. Every result is the passage itself, word for word, with its file and page.

Choose a folder, let Verbatae read it, then search. It does not interpret or summarize.

It needs no network, account or cloud, and nothing you scan is sent anywhere unless you set it up to. Your files are read in place and never moved, renamed, edited or deleted. Verbatae does write one thing beside them: a hidden folder with its copies of email attachments, of messages taken out of mailbox files and of pictures taken out of presentations. Erasing the case removes it.

How this differs from your computer’s file search

File search finds a file by name or a word you remember. Verbatae finds what the documents say, and shows where.

The passage, not a file list.

Ask in plain words. “letters where someone warns about the budget” finds passages that say it even when the words differ.

It reads scanned pages, on this computer, and marks those passages scan.

Every result can be checked. Check it against the file reopens the file (or, for a mailbox message or an email attachment, the mailbox or email) and says unchanged, changed or gone.

It names what it could not read, instead of returning nothing.

Getting started

1. Set up once. Three short screens say what it will look at and where it keeps its records. Nothing is read until you choose a folder.
2. Make a case. New case…, type a name, press Return or Enter. A case is Verbatae’s records about one folder.
3. Choose the folder and scan. Before reading, it shows the file count, a rough time and what it will leave out. Press Scan. The bar shows how much has been read, a large scanned file shows its page, and a time left appears once it holds steady.
4. Stop. Press Stop at any time. Saved work is kept, and it says how many documents were read but not yet searchable; the next scan finishes them. Stopping inside a large scanned file takes seconds; that file is not kept and is read again from the start next time.
5. Search. Press Search this case, type in the box at top left, press Return or Enter. Click a result to read it in place, see its file and page, and check it.

Folders are read when you press Scan, not watched; after adding documents, scan again.

Software folders

Folders of generated material are left out by default: programming environments, downloaded libraries (node_modules), build output, tool caches and project history (.git). They can hold tens of thousands of files and add hours to a scan. A folder is left out only when its contents show it was generated; one merely named build or env is read. Before you scan, each is listed by name with the reason and its file count, so every skipped folder is listed.

To include one: on the Scan screen, open Software folders left out and check Include. The count and estimate update after a moment, and the case remembers your choice; uncheck to leave it out again. A folder inside an included one is listed separately and stays out unless checked. The estimate shows the added time before you scan. From .git only the readable files are read.

Advanced: a .kbignore file in the folder with the line !venv/ includes that folder too.

The search screen

Checkable results. The passage as written, your words highlighted, the full file path, the page, and a fingerprint of the file as read.

Meaning and wording together. Each search matches both, so a quoted phrase and a loose question both work. Weaker matches sit below; check Strong only to hide them.

Filters. Kind of document, year, sender, recipient or a date range. Save results as a set, export them or put them on a timeline.

A record of what was read. Verbatae keeps a list of every document it read, with a fingerprint of each file. Contents lists every searchable document and every unreadable one.

Review. Documents whose wording suggests legal privilege (such as “attorney-client”) are held out of the case until a person decides. Review lists them with the unreadable ones; open each and release it or keep it held.

People, places and events

People & things lists the people, organizations and places named in the text, each with its sentence, file and page, ordered by how often it appears. That is a count, not a judgment.

Same person? Verbatae lists names that look alike (R. Smith, Robert Smith), and shows why. Nothing merges until you confirm; each decision, and any withdrawal, is kept with who and when.

Find this name everywhere. Add a person with other spellings and one-word forms, and get every mention with its sentence, file and page. Names you add are kept apart from those it finds.

Conversations. Related emails show as one conversation in date order, with only what each message adds. It says when messages were joined only by matching subject and people.

Events. A date in the same sentence as a place is an event, shown with that sentence. A date with no place stays on the timeline. A date the words don’t pin to one day is listed separately, with the reason.

Rules. Write a rule in plain English, such as “a payment should have an approval.” Verbatae shows the rule it will run; once a named person confirms it, it can run the rule and list what it looked for and did not find. It draws no conclusions.

Receipts and ending a case

A receipt for each scan: software, machine, time and a title you can type before scanning. It holds nothing about your documents, so it can be handed over. The rest (counts, files not scanned) stays locked inside the case, readable only with the case. You can also name a person who watched the scan (not the person scanning); the name is kept in that locked record, as typed: “Scanned by A, witnessed by B.”

Ending a case. Cases lists the cases on this computer. End of engagement is a slow, guarded erase behind a typed confirmation. It destroys the keys that unlock your case, removes the record of what Ask sent to an outside model, and writes a closing record. It also lists files written to this computer’s temporary folders and caches since the case began; the closing record says no one reviewed that list, or names the person who did. Close the case only after the closing record is written.

Names are as typed. Verbatae does not check identity. We recommend entering each name with the firm’s staff ID, and letting the computer’s own sign-in (Mac or Windows) control who uses it.

Limits of this version

It quotes; it does not compose. No summaries or answers of its own. Ask turns a description into a search, using a built-in phrase reader with no language model. You can link a model (on this computer, your network or a subscription service) in the Model setting; it is off by default. Its output is shown separately from your documents’ text, and is yours to check. Every result is still your own text.

Names are found by their form. One-word names, and organizations without an ending like Inc or LLC, are often missed unless added with Find this name everywhere or the organization list. A date with no year is listed separately, not guessed.

Reads: PDF (including scanned), Word (.docx), Excel (.xlsx, .xlsm), PowerPoint (.pptx, .pptm: slide text, speaker notes, tables, and pictures of a useful size), page images (PNG, JPEG, TIFF and similar), plain and text-like files (notes, logs, Markdown, CSV), web pages, e-books (.epub) and email (.eml, Outlook .msg). Not yet: .xls, .ppt, Outlook .pst. Anything unread is named on Contents.

Scanned pages are read by the computer’s own text recognition (Apple’s on a Mac, Windows’ on a PC), and nothing is sent. Recognition can misread a character, so check figures and names against the page. A photograph of a page counts too. Printed text reads well, even tilted or upside down; handwriting does not.

Complex PDF layouts (multi-column, heavily typeset) may come back out of order; those passages carry a note.

A date is the date the message gives. A header of 1970 or 1980-01-01, which some mail programs write when they have no date, is shown as is. A message with no date is counted as undated and falls in no date window.

The first search after opening a case takes a few seconds; then searches are near-instant.

Requires an Apple silicon Mac (M1 or later) with macOS 14 or newer, or a PC with Windows 11 (64-bit); 16 GB of memory (32 GB recommended); 200 GB free (500 GB recommended) where the case is kept; and about 2 GB for the app.

Questions

Does any data leave my computer? Not unless you set it up to. There is no account, telemetry or cloud. Only an outside timestamp address or an Ask model address can reach out, and only if you set one.

Does “no results” mean it isn’t there? It means nothing readable matched. It cannot see unread formats, handwriting, text too faint or blurred, documents held for review, or skipped folders. Contents lists what is searchable and what is not.

Where is my case kept? In the folder chosen at setup (by default Verbatae, in Documents), under the case name. Your files stay where they are.