Security model
Explained in "what happens if something goes wrong" mode, not brochure mode.
There's no file to steal
The usual way to compromise a document-processing service is to attack the server where files are stored: a misconfigured database, an exposed backup, an employee with too much access. In almost every tool that failure point doesn't exist here because that server doesn't exist: your PDF doesn't travel over the network toward this site, so there's nothing on our end to intercept, leak or steal. The exception is PDF to Word, which sends the file to our server only to convert it and does not keep it once the .docx is delivered.
This isn't a promise about configuration — "we have good access policies" — it's a consequence of how the site is built: in those tools the file stays in your browser's memory from start to finish.
Memory-only processing
When you load a PDF, the browser keeps it in RAM for as long as the operation runs. It doesn't get written to a temporary disk file that could be forgotten about, and it isn't saved to any local database unless you explicitly ask to download the result. Close the tab and that file disappears from memory like any other data the browser no longer needs.
Downloads go through a blob, not a server
The resulting file — the merged, compressed or signed PDF — gets built as a Blob right there in the tab and offered for download through a link that points at that local memory, generated by the browser itself. It never passes through an internet URL and never sits hosted on a server waiting to be downloaded: the link stops working the moment you close the tab.
What this site can see, and what it cannot
It can see: which tool was used and how many times, anonymously and in aggregate, so we know what works and what doesn't. It can see: whatever you voluntarily type into the contact form.
It cannot see, under any circumstance: the content of your files, their names, their metadata, or a hash that could identify them. There's no hidden telemetry sending fragments of your documents "to improve the service". If that ever changed, it would be a visible change on this very page, not fine print somewhere else.
The named exception: the AI features
Two optional features break the rule above on purpose: summarising a PDF and chatting with its content. For a language model to answer, it needs to read the text, so the text extracted from the document — not the binary file, not the images — does get sent to the larebelion ecosystem's AI hub. They're flagged as exactly that before you use them, and the other tools don't depend on them or trigger them on their own.
No hidden third-party dependencies
This site doesn't load fonts from a third-party CDN, analytics scripts, tracking pixels, or libraries served from outside this domain. Everything the browser needs to process your PDFs ships with the page itself the first time you load it. Fewer requests to outside servers means fewer places that could, in theory, observe that you're using this tool.