Privacy Policy

Last updated: August 9, 2026

This explains what Pocket Search collects, how AI processes your research, and the choices you have. For the agreement itself, see the Terms of Service.

What Pocket Search is

Pocket Search is a deep-research tool. Ask a hard question and it works the way a careful researcher would: it searches the open web, reads the primary sources itself, weighs claims against each other, and writes a cited brief you can trace back to where each fact came from. It keeps researching until the answer is genuinely ready, not until the first plausible sentence, and it shows its work so you can judge it.

A research system does the searching, reading, and writing, with a language model at its core. That is what makes it fast and thorough: it can read dozens of sources in a single pass and surface conflicts you might not have found. Like any research, a brief can be incomplete, outdated, or wrong, so verify what matters before you rely on it.

How AI processing works

When you ask a question, we send it, along with the web pages the search finds, to the third-party language-model provider that powers the model you're using. That provider processes your question and those sources to draft the brief. This means your question and the content of the pages it reads leave our server and reach that provider as part of normal operation. We don't control what a provider retains on its side; their handling is governed by their own terms.

What we collect

• Your access credential. Your access key or admin key proves you may use the service. It never reaches your browser in raw form; what you get back is a signed session, not the key.
• Account basics. Your license id, how many research credits you have, and basic technical logs (enough to keep the service running and debug it).
• The free trial, if you use it. Start a trial and we keep a small trial identity plus a salted hash of your IP and browser fingerprint (never the raw values), so we can enforce the one-trial-per-visitor limit and stop automated sign-ups. A trial needs no email.
• Your email, if you join the waitlist. Give us your email to request a key and we keep it on file until you're sent the key (or you ask us to remove it), and we use a third-party email service to deliver it.
• Your questions, only as needed. Your question and the sources read are processed server-side to produce the brief (see "How AI processing works" above).
• No ads, no tracking, no selling. We don't run advertising or cross-site tracking, and we don't sell your data.

How long we keep things

Your conversations and reports are stored in your browser. Server-side retention depends on the features you use:

• A recent-answer cache, up to 24 hours. We briefly hold your most recent answer so you can reopen it after a dropped connection or a restart. It expires on its own.
• Multi-device sync, only if you turn it on. If you enable sync, your reports are stored on our server so your other devices can see them, and stay until you delete them. Leave sync off and your reports live only in your browser, with no server copy beyond the 24-hour cache.

Temporary file attachments

When file uploads are available and you choose to attach a PDF, Markdown or plain-text file, we store it temporarily on our server for text extraction. Original files are removed after extraction. Extracted text is available to new research requests for 72 hours from upload; a run already using it may retain it for up to two hours longer. Adding files or asking follow-up questions does not extend an older file's deadline. Cleanup runs automatically while the service is running. You can also delete temporary files from attachment management. Only PDF, Markdown and plain-text files are accepted; scanned PDFs aren't OCR'd, so an image-only PDF yields no readable text.

Relevant excerpts are sent to our AI providers. Web-assisted research may send document-derived search terms to external search services; this is not an offline or documents-only mode. Provider retention follows their own terms. Filenames and quotations in generated answers remain in your chat history and may sync with your reports. Deleting a file stops further access but cannot recall text already read by a model or quoted in an answer.

What we share, and the Limited Use commitment

Your data is shared only where needed to run the service: with the language-model provider that drafts your brief, and with the search providers that find sources. We commit to the following limits:

• With the email service that sends your key. If you join the waitlist, your email is sent to our email provider (currently Brevo) so the key can be delivered.
• No advertising. Your data is never used for personalized, re-targeted, or interest-based ads.
• No selling. We don't sell your data, share it with data brokers, or use it for creditworthiness.
• Human access is limited. People don't read your questions or reports in the ordinary course of running the service. We may look only for security, legal compliance, or where you've asked us to help with a specific issue.
• Only for the service. Anything collected is used to provide or improve Pocket Search, not an unrelated purpose.

Security

Traffic to and from Pocket Search uses encrypted connections (HTTPS / WSS), with TLS terminated at the network edge. No security measure is perfect, but we keep the surface small: minimal data, short retention, and no trackers.

Your choices

• Keep reports private. Leave multi-device sync off and your reports stay in your browser only.
• Delete your history. You can clear conversations in the app at any time; clearing sync removes the server copies too.
• Avoid sending what you wouldn't want processed. Because a question and the sources read are sent to a language-model provider, don't research anything sensitive enough that you wouldn't want it processed that way.

Changes to this policy

We may update this policy from time to time. If we do, continuing to use Pocket Search means you accept the revised policy.

See also the Terms of Service. Questions? Reach us at [email protected].