When an organizer runs a session with ORACLE, we process the session's audio to produce transcripts, translations, summaries, and visuals. Attendees who join on their phone are identified only by an anonymous session token — no account, no email, no tracking across events. Questions, poll votes, and reactions are stored against that anonymous token.
A session cannot start until the organizer confirms that recording consent has been obtained for the event, in line with Singapore's Personal Data Protection Act (PDPA). Announcing recording at the venue is the organizer's responsibility; ORACLE enforces the confirmation step.
Raw audio is deleted automatically after a configurable retention window (default: 30 days after the event). Transcripts, translations and generated artifacts are kept for the organizer until the event is deleted. Organizers can export or permanently delete all data for an event at any time from the dashboard, and deletion removes the stored artifacts as well as the database records.
Session audio is streamed to Deepgram (United States) for speech-to-text. Summaries, translations and generated visuals are produced by Google Gemini models, currently relayed through yunwu.ai, a third-party API gateway which therefore also processes the prompts we send. Prompts contain transcript text, and attendee questions where the Ask feature is enabled.
This means processing happens outside Singapore. Under the PDPA that is a cross-border transfer, and we state it plainly here so an organizer can decide before any recording begins. If your organisation cannot accept the gateway in that path, contact us and we will move your workspace to a direct provider relationship.
We do not train models on your event data, and we do not sell data to anyone.
For access, correction, or deletion requests — or any privacy question — contact our Data Protection Officer at [email protected].