Not a white box on top — the words and images are removed from the file itself. Everything runs in your browser. Nothing leaves your computer.
Really deleted. Kiri uses PDF redaction: the text operators and image data are removed from the page's content stream. After deleting, type the word into the search box at the top of the editor — it won't be found. Open the downloaded file in any viewer and try Ctrl+F or select-all-copy; it's gone there too.
No. There is no server. The PDF engine (MuPDF, compiled to WebAssembly) runs in your browser and your file never leaves the tab. You can turn off Wi-Fi after the page loads and everything keeps working.
PDF has no concept of reflow — each word is drawn at a fixed position. Removing one doesn't move the others. Use the Write tool to put new text in the gap.
Scans are pictures, with no text layer. Use the Erase tool: drag a box over the area and the pixels underneath are blanked out in the image itself.
No limits. Very large files (hundreds of MB) depend on your device's memory; Kiri only renders the pages you can see, so long documents are fine.
Free, no account, no watermark. The code is open source under the AGPL-3.0 license. There's no catch — it costs nothing to run because there's no server.