The live assistant keeps text and captions in page memory, offers a local text download and does not save conversations for FASS model training.
Quick answer
Use Download conversation before resetting or leaving the page to save the available messages and voice captions as a text file. The live assistant does not save its transcript or raw audio in the FASS database; OpenAI processing has separate terms.
The live Ask FASS assistant keeps its conversation in page memory. FASS does not save its transcript or raw audio to the application database or use these conversations to train or fine-tune models.
Explicitly saved trip plans and voluntarily shared answer-feedback excerpts are separate exceptions to page-only chat. A saved plan contains structured trip details and public shortlist references, until plan deletion or account erasure; it contains no transcript or audio. Answer feedback contains category/topic labels by default. Only an edited excerpt you deliberately preview and confirm, up to 500 characters, is retained with feedback. Feedback is excluded from reports after 90 days and scheduled for deletion by bounded hourly maintenance. These records are included in account export/erasure, are not sent to an AI provider, and do not train models. Applied trip details can separately be sent as AI context after consent, including explicitly supplied learner ages. See trip planning and answer feedback.
Choose the header's Download conversation control before resetting or leaving the page. It creates a plain-text file on your device from the available messages and voice captions. It does not upload the file or create a server conversation record. Captions can be incomplete or wrong; this is not an audio recording. Keep a downloaded file private if it includes personal details.
New conversation, reloading, account changes and withdrawing AI consent clear the in-page history. Closing the panel ends voice but keeps that page's text history available. Clearing FASS history does not delete a file you downloaded.
A labelled Draft reply is temporary text while a supported answer is being generated. It joins the conversation and becomes available to download only after the complete reply is validated and its stream finishes correctly. Stopping or failing the reply discards that draft. Draft text is still AI-provider processing; being temporary does not mean it stays on your device.
With AI & Chatbot consent, OpenAI processes submitted messages, voice captions and connected microphone audio. The assistant can also send the current page route and language, relevant public search/help results and only the account or school records you are authorised to use and which are needed for your request. Avoid passwords, card details and unnecessary health or family information.
OpenAI's processing and retention follow its service/account terms. Page-only history does not mean that content never leaves your device or that provider retention is zero. See the Privacy Policy and AI Transparency.
Older Ask AI help chat can keep browser history for 24 hours; directory finders, saved searches, provider messages and school chatbots have separate rules. This guide describes the new live assistant only.
Memory-only history is separate from necessary operational records. After assistant consent/start, an HTTP-only pseudonymous guest cookie lasts up to 90 days; keyed network identifiers help enforce quotas. Usage/credit receipts contain dates, duration, counts and charges, not full transcripts or audio. Closed usage and inactive quota records are kept up to 90 days; daily network-quota identifiers up to two days. Security flags contain keyed hashes, categories, counts and expiry, not raw IP or messages; they expire after 24 hours and are removed by maintenance. Action previews expire after ten minutes, are scrubbed after execution and are purged after 24 hours. Account export/erasure covers account-linked operational records; required financial and minimal audit records have separate retention. Server security checks inspect submitted text/transcripts in memory. See the Privacy and Cookie policies for details.
Navigation reliability diagnostics are separate from conversation storage. The current server process keeps fixed offer/attempt/outcome counts for up to 30 minutes without request text, routes, filters or account/IP identifiers in those aggregates. Short-lived server-issued reporting receipts are valid for ten minutes and use keyed owner/action digests and fixed lifecycle information, not raw identities or URLs. They are bounded in memory and pruned during reporting or diagnostic reads; restart clears them. A successful browser report is not proof that you finished a task or resolved a support request. These diagnostics do not train an AI model.
After AI consent, signed-in recommendations can automatically use minimal saved name, self-reported activity/ability and piste comfort, up to eight current public favourites per kind and limited completed-booking aggregates. These are not verified skill or personal ski time, and your current request takes precedence. Guests, demos and support impersonation receive no real-account preferences. See saved preferences.
A private preview uses a browser-bound access cookie for at most 72 hours; authorised QA demos have a separate at-most-three-hour workspace cookie. Neither cookie alone gives AI consent or extra data rights. Grant metadata contains hashes and dates, not raw link tokens, and terminal grants are removed after 24 hours by maintenance. These are not stored conversations or a self-service temporary-account feature.
Contextual walkthroughs on Account Support are separate from the live assistant transcript. An offer stores fixed guide information and a keyed fingerprint, not your raw draft. Accepting help may retain the unchanged support draft and attempted-step information for up to 24 hours; ending the session clears its draft and preview content sooner. Minimal assistance and outcome records remain for up to 30 days and are included in account export and erasure.
An optional AI explanation requires your existing AI consent and a deliberate send after disclosure. It uses a bounded excerpt of the selected owned request and recent customer-visible replies, excluding internal/private notes and attachments. The explanation is not saved as a FASS model conversation or used for training. If you review and send a handoff summary, that summary becomes a normal reply on your support ticket and follows the ticket's retention rules.
Still need help with this article?
Send the article URL and the step that needs more explanation.