Your data, and the AI behind Cadence.

Personalized answers need context. Here is what you share, when it is sent, and what stays under your control.

Résumé and interview onboarding

Your uploaded résumé is converted to text. The extracted text and interview details are saved to your workspace. When you prepare answers, the relevant context is sent through Vercel AI Gateway to generate a personalized answer pack. You can review or edit the text before using AI.

Live questions and answers

When you start listening, your selected audio source streams through AI Gateway to Google’s transcription model. Cadence does not save raw audio. A detected or typed question, your relevant background, and recent question-and-answer context are sent to the selected answer model. If you upload a question screenshot, that image is also sent for the requested analysis.

Models and providers

Cadence currently supports Google Gemini, OpenAI GPT, and Anthropic Claude models through Vercel AI Gateway. The default answer model is Gemini 3.5 Flash-Lite, with Gemini 3.5 Transcribe Live for audio transcription. Model availability may change. Current Gateway metadata identifies these selected services as not using prompts for training, but retention differs across services. Transcription and some realtime services do not have zero-data-retention coverage. We do not promise universal zero retention.

Consent and control

We ask for your permission before enabling AI processing. You can withdraw it from Account, stop listening at any time, edit your context, export your workspace, or delete your account. Microphone permission from your operating system is separate from AI consent. Make sure everyone whose speech is captured is aware and that your use is permitted.

Use your judgment

AI output is a suggestion. It can mishear speech, invent details, or miss context. Cadence asks models to ground personal claims in your supplied experience and labels suggested approaches, but you should always verify the answer yourself.