Manual

How it works

1. Drop files on the page or press "Choose files". PDFs, photos (PNG, JPEG, WebP and others) and text (TXT, Markdown) are accepted. Photos and text become pages at once.
2. Every page is a card on the desk. Select cards, rotate them, delete them, drag them to another place or into another file. Actions for a whole file are in the "File" row: merge, split, numbers, stamp, compress, read text, metadata, format.
3. Every action becomes a step in the list on the right. A step can be changed or removed; the result is computed again. Ctrl/⌘ Z undoes the last action, all the way back to the start.
4. The desk shows the result as it will be. Nothing happens to the originals.
5. "Save result" builds new files and hands them over as an ordinary download.

To redact something, fill a form or place a signature, open the page: double-click a card or press Enter.

No file at hand? Press "Try a sample" on the empty desk and pick one. A short list of things to try appears next to it; you can hide it, and it is never shown for your own files.

• The big pack - 63 pages, 12.4 MB: everything at once: blank pages, splitting, scans, a form, redaction, compression.
• Phone scan - 5 pages, 1.3 MB: turn the pages that lie wrong, drop the blank one, read the text.
• Contract - 10 pages, 0.05 MB: name it from its requisites, number the pages, stamp it.
• A stack of invoices - 8 pages, 0.05 MB: split into eight invoices and name each one.
• Form - 2 pages, 0.03 MB: fill the fields, place a date and a signature.
• Photos in a PDF - 6 pages, 1.8 MB: compress six big pictures.
• Photos as separate files - 6 PNG files, 1.8 MB: put six pictures into one PDF.
• Personal data - 3 pages, 0.61 MB: redact text and a scan, check that nothing is left under the black.
• English contract - 6 pages, 0.06 MB: text in two languages: a name, export to Word, numbers.

Every name, company, code, account, phone and address in the samples is invented: the companies are called "Демо…", the company codes start with 99, the bank does not exist. A sample is loaded from this same site only when you pick it, and works offline after that.

What is guaranteed and what is not

Redaction

Open a page, choose "Redact" and draw over the area. When you save, the tool does this:

• reads the page content and removes every character that touches the area. The rest of the line stays where it was;
• overwrites the pixels of images under the area with black and stores the image again (the old image does not get into the file);
• removes links, comments and form fields that lie on the area;
• removes lines and shapes that are entirely inside the area (underlines, a cell fill, text turned into outlines);
• creates no recognised-text (OCR) layer in this area;
• draws the black box.

Then comes the check. The finished file is opened again, its text is extracted, and no character may lie inside an area. In the browser the area is also drawn without the black box: it must be a flat background. You see the outcome as one line: "Checked: there is no text under the 3 redacted areas".

When a page becomes a picture

Some pages cannot be edited this way: a font without a width table, content inside a nested block, an image in a rare format (JBIG2, JPEG 2000, CCITT), visible graphics under the area. Then the whole page is turned into a 200 dpi picture with the area painted over. Such a page has no text at all any more, and the result says so plainly and names the reason. The "pixelate" look always works this way.

What the check does not see

Lines that only cross the area (a table border) stay: they are not text. If more than 12% of the pixels under the area are not background, the page becomes a picture. The check does not look for the same text elsewhere in the document.

Fill and sign

If a PDF has form fields, they become inputs on the open page: text, check boxes, radio buttons, lists. The fields can stay editable or be locked: then the values become part of the page.

If there are no fields, place text, a date or a check mark anywhere: choose the tool and click on the page. An item can be moved, resized and removed.

A signature can be drawn with the mouse or a finger, typed, or taken from a picture (a white background becomes transparent). It is stored in the browser only if you tick "Remember". This is a visual signature. It has no legal force of a qualified electronic signature.

Text recognition

The "Read text (OCR)" step reads the text on scans and lays it as an invisible layer over the picture: you can search and select text in the file. Pages that already have text are left alone. The "text", Word, Markdown and EPUB formats run recognition on scans by themselves.

The engine and the language model (Ukrainian and English, about 7.5 MB) are loaded from this same site on the first recognition, with a progress line. The more accurate model is about 25 MB. After that everything works offline.

Say it in words

In the line at the top you can write what to do: "merge these, drop blank pages, number from page 3, compress for email". The sentence is read by an ordinary table of rules in your browser, in Ukrainian and English. There is no model and no network here. First you see how you were understood, and only then you confirm. What the tool did not understand is shown separately and is not guessed.

Dictation is off until you turn it on. It uses the speech recognition of the browser itself. Chrome, Edge and Safari may send the voice recording to Google, Microsoft or Apple servers. Files go nowhere.

Recipes

A recipe is a saved list of steps. Press "Save as a recipe", give it a name, and next time the same set of actions runs with one click on other files. Recipes live in this browser. A recipe can be saved to a small JSON file and passed to a colleague. Built-in recipes and all operations are on the "Operations and recipes" page.

Offline and as an app

On the first visit the browser stores the site itself: pages, code, fonts and libraries, about 6.5 MB in all. After that the site opens and works without a network. The recognition data (7.5 MB) and the extractor for Word and Markdown (5 MB) are stored on first use.

In Chrome and Edge the site can be installed as an app: the "Install" button in the address bar. Such an app can open PDFs straight from the file manager. In Safari on a Mac: "File → Add to Dock".

Keys

For scripts and agents: window.ihatepdf

The same operations are available from code on the page. This suits an agent that drives a browser, and your own scripts in the console. There is no server behind it: everything runs in this tab.

// what it can do
ihatepdf.operations()      // [{ op, title, about, params: [{ name, type, default, choices }] }]
ihatepdf.recipes()         // built-in recipes and the ones saved in this browser

// a sentence -> steps. no model, no network
ihatepdf.parse("merge these, number from page 3")
// { steps: [{ op: "merge", params: {} }, { op: "number", params: { from: 3 } }], unknown: [], notes: [] }

// a plan + files -> files. the desk on the screen is not touched
const { files, report } = await ihatepdf.apply(
  [{ op: "merge" }, { op: "delete", params: { pages: "blank" } }, { op: "compress", params: { targetMB: 5 } }],
  [fileA, fileB])          // File, Blob, or { name, bytes } / { name, text }
files[0].name; files[0].blob; files[0].size
report.notes               // what happened: compressed, pictures, ocr ...
report.redaction           // the proof after a redact step

// the desk a person sees
await ihatepdf.desk.addFiles([file])
ihatepdf.desk.addStep("rotate", { pages: "1-2", angle: 90 })
ihatepdf.desk.setPlan({ steps: [{ op: "ocr" }] })
ihatepdf.desk.state()      // { files: [{ name, format, pages }], steps: ["..."], errors: [], selected }
await ihatepdf.desk.save() // { files, report } without a download dialog
ihatepdf.desk.undo(); ihatepdf.desk.redo(); ihatepdf.desk.clear()
ihatepdf.samples()         // the built-in samples of made-up documents: [{ id, name, pages, bytes, files, actions }]
await ihatepdf.desk.loadSample("invoices") // one of them by id; no id is the big pack, and ihatepdf.sample() says what is on which of its pages

The plan

A plan is plain JSON: { "steps": [{ "op": "rotate", "params": { "pages": "1-3", "angle": 90 } }] }. Steps run in order. ihatepdf.operations() returns the list of operations with their parameters; the same list is on the "Operations and recipes" page.

• doc - the number of a file on the desk, from 1. Without it the step acts on every file.
• pages - "all", "odd", "even", "blank", "first", "last", "last 3" or ranges "1-3, 7, 9-".
• Coordinates for redact and annotate are in points (1/72 inch) from the top-left corner of the page as the reader sees it.
• An error carries a code field, for example noSuchPage, password, empty.

Limits

• Very large files need memory. A scan of hundreds of megabytes may not open on a phone.
• Compression shrinks JPEG and simple raster images. If that is not enough for the size you set, pages become pictures and the result says so.
• Conversion to Word, Markdown and EPUB keeps text and headings, not the layout.
• After merging, splitting or deleting pages the bookmarks of a document are not carried over.
• A filled form with Cyrillic is about 400 KB larger: a font is embedded so the fields stay editable.
• A damaged PDF may not open. The page says so and sends the file nowhere.