Redacting text seems to actually work. However, editing existing text results in both the original text and the edited version being shown in the PDF after download.
(The page downloads mupdf (WASM) for rendering the PDF. When "downloading" (= saving) the PDF, the page first checks whether the allowed three downloads have been reached via a POST request (no PDF data uploaded), then it downloads PyIodide and some Python wheels (pdfrw, defusedxml) before creating the PDF file.
I believe AGPL'd software cannot be sold without a license unless there is full disclosure of the source code.
If that is the case, the OP is likely in violation of MuPDF's AGPL license if he is selling and distributing binaries without contacting sales.
It is a clear AGPL license violation without contacting the MuPDF sales team.
If the OP doesn't want to contact the sales team, it should be open source.
Yes, PDF data is never uploaded to the servers. It's the entire reason I created the project, after seeing the all the main results you see when you search on Google upload your data to their servers.