What Agnos doesn't do

What this version of Agnos does not do, so you know before you buy.

Scanned documents

Agnos does not read text out of pictures. A PDF needs a text layer, and a scan or a photo of a page has none. If your documents are scans, this version is not for you.

Some file formats

Old Office files (.doc, .xls, .ppt), Rich Text and OpenDocument files must first be saved as .docx, .xlsx or .pptx. Password-protected files, images, audio, video, archives, programs and binary data files are refused. See Supported files.

Windows only

Agnos runs on 64-bit Windows 10 and Windows 11, and there is no build for any other system. The installer is not code-signed yet, so Windows warns about it the first time; see Troubleshooting.

Parts of a file

Some content is not read, such as Word comments and page headers, text in charts and SmartArt, pictures, and PDF comments. Hidden sheets, hidden slides and speaker notes are skipped unless you turn them on. Each file's warnings say what was left out; the full list is in Supported files.

How documents come out

• Spreadsheets become rows of text. Formulas, formatting, merged cells and column types are not kept.
• Only the Sections split keeps a table or code block whole. Other splits cut wherever the count says.
• Word pieces carry their headings, not page numbers. A PDF page reference can be off by one where a sentence crosses a page break.

Duplicates and splits

• Text rewritten in other words, translated or summarised is not matched as a repeat.
• Text that differs in its numbers is never a repeat, so retyping 1,200 as 1200 is not caught.
• Keep one copy, counting near matches keeps the first copy, not the longest, so a later copy's extra sentence can be lost.
• Above 100,000 distinct pieces, an export matches identical text only, and its manifest says so.
• A validation score from a split is slightly optimistic: related text written differently can land on both sides.

Review and the rules

• The rules are rules of thumb. A removal means a rule matched, not that the text was worthless, and a kept piece is not guaranteed correct, current or safe to use.
• Review lists removals, not repairs inside a line. If cleaning makes more than 500 removals in one file, all are counted but only the first 500 are listed.
• The Privacy settings match patterns. They miss things a person would catch, including names in ordinary prose, so do not rely on them alone before sharing a dataset.
• Values carried as data beside the text, such as kept JSON fields or a row's values in Records mode, are never cleaned or hidden.

No encryption

Your documents and the cleaned text sit in %LOCALAPPDATA%\Agnos with your Windows account's permissions. Agnos adds no encryption, and anyone who can run programs as you can read them. If they are sensitive, use Windows disk encryption; see Your files and privacy.