Is my audio used to train AI models?

No. Canto does not use your recordings or transcripts to train AI models, and never sells them or licenses them to anyone. Your content is processed by AI providers only to deliver the features you use — transcription, summaries, chat, and search. The binding version of this is the Privacy Policy.

What AI processing happens to my content?

When a recording finishes, it goes through a fixed pipeline, all server-side:

1. Transcription. The audio is sent to a specialist speech-to-text provider, which returns the transcript. Canto keeps the provider's raw output alongside your recording so features can be re-run without re-transcribing.
2. AI generation. Automatic titles, summaries, chat answers, and document templates are produced by the Google Gemini API from your transcript.
3. Search indexing. Transcript passages are converted into numerical embeddings — lists of numbers representing meaning — and stored with your data so semantic search and chat retrieval work. Embeddings aren't readable text, but they are derived from your content and are deleted with it.

There is no AI processing of content you haven't recorded or imported, and AI runs when you use a feature or a recording completes — not as background mining of your library.

Do the AI providers train on my content?

Canto's position, from the Privacy Policy: these providers process your content to deliver the service to you, and Canto does not use your recordings or transcripts to train AI models. Where a provider's standard terms would let it train on customer content, Canto is moving onto plans that turn that off, and will tell you the current position for any individual provider on request — email hello@tessa.ba.

One more provider sees AI traffic: Canto uses an EU-hosted tracing service to debug AI quality issues, and those traces may include the prompts and responses of your AI interactions. Like everything else, this is processing to run the service, not training.

Where does AI processing happen?

Your stored data lives in the EU, but the Gemini API is not region-pinned, so AI requests can be processed outside the EEA. Those transfers rely on the European Commission's Standard Contractual Clauses. Details: How is my data stored and secured?

FAQ

Does chatting with AI about a shared recording expose my questions?

No. Your AI chats are private to you — other Space members and the recording's owner can't read them. See Who can see and edit what?

Is AI content generated from my voice itself?

Summaries, titles, and answers are generated from the transcript text, not from your voice audio. The audio is used for transcription and playback.

Can I use Canto without the AI features?

Transcription and automatic notes are core to how Canto works — every completed recording is transcribed and summarised. But chat, search, and templates only process content when you invoke them, and you can simply not use them.

Does deleting a recording delete the AI data too?

Yes. Deleting a recording removes the audio, transcript, raw transcription output, embeddings, and generated AI content. Deleting your account removes everything — see How do I delete my account?