What Coco will never do

Knowing what Coco can touch is only half the picture. Just as important are the lines it will not cross. These are not settings you have to turn on — they are how Coco is built.

Never send anything without your explicit yes

Every reply waits for your one tap. Nothing leaves your account on its own — no auto-replies, no "helpfully" sending while you are away, no exceptions. The send button is yours.

Never add, move, or delete an event you did not ask for

Coco only makes the specific calendar change you told it to make. It does not reorganize your schedule, does not move your meetings, and never deletes an event.

Never delete or empty your email

Coco does not trash, archive away, or empty your mail — and it never even asked Google for the permission to. Your inbox is safe from Coco by construction, not just by policy.

Never let anyone else in

Only the address you list can sign in to your Coco. Leave it blank and the door stays shut to everyone but you. Coco runs on your own accounts, not a shared service that others could reach.

Never share your inbox or train an AI on it

Your mail lives on your own accounts, not a pooled Coco service. It is never sold, never shared, and never used to train a model. The AI that powers Coco runs through Anthropic's API, which does not train on what is sent to it. No one else ever sees your mail. For the detail, see the LLM and your data.

Why this matters

An assistant that reads your inbox is only worth using if its limits are clear and real. These are not aspirations — they reflect the permissions Coco asked for (and the ones it deliberately did not), and the defenses that enforce them. See what Coco can do for the matching list of permissions, and security defenses for how the lines are held even against a malicious email.