Ember is an AI friend for adults, in the Messages app, in private beta. What makes it worth a second read is not the product, it is the document.
It publishes a Safety & Crisis Protocol, dated and stated to be published in accordance with California Senate Bill 243. That makes it the first listing here carrying a document because a law required it rather than because a lawyer suggested it. The protocol says an automated classifier screens inbound messages for expressions of crisis or suicidal ideation, and that on a positive detection the reply leads with 988, the Crisis Text Line and 911 before any AI-generated response. It then says the classifier is not perfect, that it may miss distress and flag things that are not a crisis, and that Ember cannot call for help, cannot reach a person and cannot intervene. It also discloses that certain safety signals are routed to trained members of the team for review, which means a person may read the thread.
There is no privacy policy at any path we could find, so the training question, the retention question and the question about who carries the messages are all unanswered. A companion product with a legally mandated crisis document and no privacy policy is an odd shape, and it is the shape as published.
The number is not a number. It changes on every page load, so there is nothing stable to record and nothing stable to save in your contacts.