The LLM and your data: no training, no retention

Coco reads your mail and calendar and turns them into a brief, drafts, and answers. That understanding comes from a large language model — specifically Anthropic's Claude, accessed through its API. It is reasonable to ask what happens to your data when it passes through that model. Here is the plain answer.

No training on your data

Anthropic does not train its models on the data sent to the Claude API. The content Coco sends to generate your brief or draft a reply is not absorbed into a future model, not used to improve anyone else's results, and not mixed into a training set. It is used to answer your request, and that is all.

No retention beyond the request

The text Coco sends to the API is processed to produce a response and is not retained as part of a long-term store to be reused. The model does not "remember" your inbox from one request to the next — each request stands on its own. The memory that makes Coco feel like it knows your people and topics is something Coco maintains in your data store, on your own infrastructure, not something the model holds.

Where your data actually lives

Coco runs on your own accounts and your own database. The brief history, the learned topics and people, your drafts and settings — those live in your infrastructure, under your control. The Claude API is called to do the reasoning, but your data is not parked on someone else's service. That is why exporting or deleting is meaningful: it acts on a store you own. See export your data and delete your account.

What this means for you

You get the quality of a frontier model reading your day, without your inbox becoming training data or living on a shared server. The intelligence is borrowed per request; the data stays yours. For the limits Coco enforces on its own behavior, see what Coco will never do, and for the defenses around all of this, see security defenses.