Desktop audio transcription, screenshot recognition, and document extraction happen locally. Generation sends the question, enabled excerpts, session context, and necessary recent conversation to your configured provider. Original files, raw audio, and screenshots are not included in that request. Preparation, rewriting, mock practice, and review may each make model requests.
Confirmed desktop materials, session context, and records you save use operating-system encryption, or remain in memory if it is unavailable. Terminology rules and exported files are plain text. Answers are not automatically archived; raw audio is not saved. Closing a screenshot clears the app's memory copy, not your system clipboard.
Web uploads follow a different path: after confirmation, a file goes to the server for extraction. Review and explicitly save the text to your account before cross-browser readback. Hosted AI remains off. Product and download pages do not collect résumés, interview records, API keys, or payment details, or load third-party analytics or fonts. Do not email secrets or keys.