Pero ai disclosure

AI Disclosure

Last updated September 2, 2026

Provided for Apple App Review Guideline 5.1.2(i) and for users.

What AI does in Pero

Pero's core loop runs on third-party AI models hosted outside your device:

  1. Speech to text. Audio from the capture button and from voice questions is transcribed by a Whisper model on Cloudflare Workers AI. Audio is not stored.
  2. Triage and properties. The transcript, or the text you typed into the capture field, together with the task titles and notes already on your account, is sent to a language model (Google Gemini, through Cloudflare AI Gateway) to decide whether it is a task, note or quick win and to fill in deadline, project, estimate and similar fields.
  3. Plan, questions, weekly review. The same model assembles the daily 1-3-5 plan, answers questions you ask by voice about your own tasks, and writes a short weekly review.
  4. Search index. Task and note text is embedded by a model on Cloudflare Workers AI so voice questions can find relevant items.

What is sent

Only your content (audio, transcripts, typed captures, task and note text) and a random request ID. Your name, email and sign-in identifiers are never sent to AI providers. Providers are used under API terms that do not allow training on your content. Pero does not train models on your content.

When it starts

The backend refuses AI requests until your account has an AI-processing timestamp. It is set when you pass the microphone step of onboarding, whether you allow the microphone or choose to type instead. There is no in-app switch to turn AI off while keeping the account; to stop processing, delete your account (Settings, Account, Delete account) or email support@pero.day.

Limits

AI output can be wrong

AI output can be wrong. Pero shows a tilde next to fields it guessed and lets you correct every result from the receipt card. Pero does not use AI to make decisions with legal or similarly significant effects on you.

Report a problem

Use the support page or email support@pero.day with the approximate time of the capture. We do not keep the audio, so include what you said if you can.