II.5 Privacy and data lifecycle
Read this before uploading anything that matters
A web app is somebody else's computer. That is not a flaw; it is how a web app works. What matters is knowing what happens to your file on that computer. This chapter reports what the product's code and its own pages state, marks what the book cannot know, and then gives you habits that work either way.
It is not a legal document. The authoritative texts are the site's policies (the .app site carries a policies section) and its terms; see also Appendix H.
The life of a file
flowchart LR
A[You upload] --> B[Stored for the task]
B --> C[Worker processes]
C --> D[Result stored]
D --> E{Delete button?}
E -->|Yes| F[Removed now]
E -->|No| G[Removed automatically after about 2 days]
- Upload. Your file travels to the Documentize service and is held in its storage for the task.
- Processing. A worker for that app reads it and writes new files. Your original is never modified.
- Result. The results wait for you to download. Across the app chapters the consistent statement is that results are kept on the server for about two days.
- Removal. You can remove files earlier with the Delete files button (the Merger chapter notes the site states "up to two days" and that you can delete them yourself). In the API the equivalent is an immediate-remove call for a task.
A continuation (see II.4) holds unprocessed input files and options for two days as well, so they are on the server until they expire or you discard them.
What the apps' own FAQs say
Different apps state their own retention. Read the FAQ on the app you use. As reported in the chapters:
| App | The site's statement (summarised) |
|---|---|
| Most classic apps | Results kept about two days; delete button available |
| Ask PDF | PDFs and knowledge base kept about a day, then deleted |
| PDF Knowledge Base | Kept about a day, deleted automatically, "not shared with anyone" |
| Checklist | Documents removed once the task completes |
| Cheat Sheet | "Deleted immediately after generation" |
| Collage | Images processed in memory and deleted right after the PDF is generated; the finished PDF stays about two days |
| Chat (create a PDF) | Uploaded files and generated documents removed automatically after processing |
| Resume Analyzer | Files processed only for analysis and removed automatically afterwards |
These statements differ in wording, and some pages differ from the usual two-day rule. Treat "about two days" as the upper edge for downloads, and do not use any of these as storage. Download what you need.
What the AI apps send to models
From a reading of the workers (see Trust and verification):
- Text, not images. The text of your pages, or of retrieved passages, plus your question or option choices, is sent to a language model. Page images used for highlighting citations are rendered by Documentize and are not analysed by the model.
- Long documents are split into parts and sent as several requests.
- Scans are recognised first in apps that do that (Abstracts); the recognised text is what goes on.
- Embeddings (numeric fingerprints of passages, used for search) come from the same model endpoint.
- Which vendor runs the model is a server setting that the book cannot state. Ask the operator if it matters to you.
- Repository apps (Part XI) read your code; for a private repository you invite the
arise-projectGitHub account first, and the content is read by Documentize's services and, where an app writes text, a language model. Treat the repository as shared.
What stays out of the result
Some secrets are explicitly handled. Sign uses your .pfx certificate and its password to sign; the chapter says Documentize does not issue certificates and does not keep yours, and that the certificate is not part of what you download. Protector says there is no way to read a password back from the result and that Documentize does not keep it for you. These are statements from the app pages; take them as such.
Password-protected and confidential documents: safe-handling rules
- Decide before you upload. If you may not share the document with a processing service, do not upload it. Check your organisation's rules.
- Send less. Cut the file down first. Splitter or Delete Page can reduce what leaves your hands.
- Know what protection does. A Protector password hides the file from people who lack it; it does not stop a processing service that you gave the password to.
- Unlock is a two-way door. After Unlock, the copy has no password. Re-protect it, and delete the open copy.
- Delete when done. Press Delete files as soon as you have downloaded.
- Covering is not redacting. A black box over text can leave the text underneath. Redactor is not available; the chapter explains the manual route and how to verify with a text search.
- Remove hidden data before sharing. Metadata can carry names and dates: inspect with Extract Metadata; Recovery has sanitize options (read what its report says it could and could not prove).
- Other people's data. CVs, medical or legal files carry duties to the people in them, not only to you.
- Download links and ids. Treat a result link or a continuation id as a key. The continuation id is, per the code, random and unguessable, and whoever holds it may continue or discard.
- Keep the original. Nothing here replaces your own backups.
When the file must never leave your network
Use a channel that does not send it away:
| Channel | Where processing happens |
|---|---|
| Web app, MCP server, REST API (cloud) | Documentize servers |
| CLI plugins | Your machine; the engine runs locally |
| Self-host | Your server only |
See Which channel when and Self-host. The self-host chapter says its AI features and app list may differ from the cloud; verify before relying on it.