Understanding AI permission

The intended permission wording

To understand my requests and prepare replies, suggestions and account memory, I allow Fil to send the content needed for these features to the AI provider identified in the app: my messages, useful excerpts of conversations, agenda and memory, and images, documents, place information or voice messages used by enabled features. Audio may go to a transcription service and its text may then be analysed.

If my connection uses OpenRouter, requests go through OpenRouter and model execution services, including Jev/TypeSafe decisions when configured. The app must show the actual services and models before asking for permission. A different connection, such as OpenAI, Google Gemini, Mistral, Groq, DeepSeek or a compatible server, must be identified in the same way.

These services may process data outside Europe and have different retention rules. AI permission does not guarantee zero retention or give a blanket permission to train models. I may decline AI features and withdraw permission for new transmissions; this does not recall past processing or automatically delete my account.

An active, timely choice

The intended choices are “Allow AI features” and “Continue without AI”, with access to this information. Nothing should be preselected or inferred from navigation, silence, account creation or provider configuration.

Declining must prevent the affected features from sending content to AI. Available non-AI features may remain usable. Search, location and notifications have their own exchanges and permissions. Confirming a suggestion cannot retroactively authorise analysis that already happened.

Separate permissions

Microphone permission enables recording, not by itself transmission to AI. Selecting a calendar enables import and does not replace disclosure of its use as AI context. Place observations, notifications and waiting-list contacts are not covered by a single undifferentiated AI permission.

Memory can retain information derived from exchanges. Its retention and sharing choices are separate from the initial analysis: information excluded from memory may already have been analysed by the provider. Particularly sensitive information and other people’s data need extra care.

Withdrawal and provider changes

Dedicated permission collection and enforcement before calls still need to be integrated before public distribution. In the private test version reviewed, configuring an AI connection can enable calls. Publishing this wording does not change that behaviour or create a recorded agreement.

Until dedicated controls exist, deleting the AI connection in settings stops new calls through it. Requests already started may finish using their original configuration. New recipients or wider purposes require updated information and, where needed, fresh permission. For erasure or help: contact@heyfil.com.