How Paper handles
your study materials.

Each source you upload (PDF, video, audio, web page) gets its text extracted, split into passages, and indexed so chat can find relevant excerpts on demand. This is retrieval-augmented generation (RAG): when you ask a question, Paper retrieves the most relevant passages from your own content and sends them to an AI model to craft an answer with citations.

Paper currently uses chat models hosted by OpenAI, xAI, and Google. The user can pick which one from chat settings; Grok (xAI) is the default. Audio transcription and the embeddings used for retrieval go through OpenAI.

We evaluate providers continuously and may add or replace them. Any third party that processes user content is listed in the Subprocessors section below.

Embeddings are numerical representations of your passages that let the system find semantically similar content. We store them in our own managed database alongside the source they came from, never with a third-party embedding-host.

Embeddings inherit the same access controls as the source they were generated from. If you delete a source, its embeddings are deleted with it.

No. Paper doesn't train any models of our own. The AI providers we call (OpenAI, xAI, Google) use API endpoints that, by default, do not use API inputs for training. We have not opted into any training feedback loops, and we don't send your content through any provider feature that would change that default.

Some AI providers may retain API traffic for a limited period, often up to 30 days, for abuse monitoring or security purposes. We do not use provider-side prompt storage as a long-term archive.

Paper is designed so each chat can only retrieve from the sources attached to that page and accessible to that user. Chats don't reach content from pages you haven't been given access to.

When a page is shared, everyone with access sees the same set of attached sources - that's the point of sharing. A shared page still won't reach into your other private pages or into other users' workspaces. Each chat turn only ever runs against the sources of the page it's on.

You can narrow further by picking a page range on a PDF or a timestamp window on a video. The retriever filters by those bounds before any passages reach the model.

Per-user access is enforced at the database layer, not just in the application code. Queries are scoped to the authenticated user, which adds an additional layer of protection if application logic fails. Workspace members only see what has been explicitly shared into that workspace.

• In transit: TLS 1.2+ on every connection between your browser and Paper, and between Paper and the AI provider APIs we call.
• At rest: Disk-level encryption on the database and file storage. Sensitive integration tokens (OAuth credentials for calendar connections, for example) get an additional field-level encryption layer before being written.
• Passwords: Passwords are handled by our authentication provider and stored as secure hashes. Paper's application code does not store plaintext passwords.

Paper's core database and file storage are hosted through Supabase on AWS infrastructure. We don't replicate user data to any third-party analytics warehouse.

Deleting a source or page removes it from the active workspace immediately. We keep point-in-time backups for 30 days so we can restore your account if something goes wrong - after 30 days, the backup is purged and the data is unrecoverable.

When you delete an account, the same 30-day backup window applies. AI-generated content (flashcards, summaries) and embeddings are deleted together with the source they came from.

Email/password or Google sign-in. Sessions are stored as encrypted cookies that JavaScript on the page can't read, so a malicious script can't lift your session.

Not by default yet. If your workflow requires 2FA, email us at hello@paper.ac and we'll prioritise it.

Yes. Each chat page lets you toggle individual sources on and off, restrict a PDF to a specific page range, or restrict a video to a timestamp window. The AI only ever sees what's selected.

Notes export as Markdown from any page menu. Uploaded source files can be re-downloaded from the source view. Account deletion is in Settings → Account; if you need a one-click full-account export, email us and we'll arrange it.

Not yet. Paper is currently in beta, and we don't yet have formal third-party audits or certifications. We apply industry-standard security controls in practice, and we'll publish audit reports as we obtain them.

If your workflow requires a current attestation report, email hello@paper.ac and we'll let you know what we can share.

Not yet. We don't currently sign BAAs and Paper is not a HIPAA-compliant environment. If you have a specific use case, get in touch.

The handling of personal data, your rights to access / correct / delete, and our retention windows are described in our privacy policy. If you need a Data Processing Addendum, email us.