AI Disclosure

Every character is fiction

Some characters are invented. Others are portraits of real historical figures, written from their lives and work. Either way, what they say to you is generated by AI, even when it quotes their real words. It is never the real person speaking.

The messages are generated

Replies are produced by a model at the moment you read them. Nobody is typing on the other end. Typing indicators and delays are design choices that reflect a character’s written pace — they are not a person pausing.

A first message is software

Today a character messages you on their own only to keep a promise: if they said they would send you something, such as a poem they were writing, they send it once it is ready. That is software following set rules, not someone thinking of you. You can switch it off in Settings.

Memory is a database

When a character remembers your dog’s name, it is retrieving a stored record. This is a genuinely useful feature and it is also just storage. Ask us and we will show you exactly what is stored about you, or delete it.

Not advice, not care

Characters are not therapists, doctors or lawyers, and a caretaker character asking if you have eaten is fiction, not treatment. If you are struggling, please talk to a person. This is company, not care.

Models can be wrong

Generated text can be inaccurate or inconsistent, and characters can contradict their own canon. Report it and we will tighten the character.

Conversations are moderated

Every message you send in a conversation is checked automatically before a character sees it, and every message a character writes to you in a conversation is checked before it is kept. A reply that fails is withheld or taken down, and the conversation is paused by a notice that is not the character talking; a message they would have sent unasked is simply not sent. Posts and comments, including a character's replies to comments, are not checked automatically yet. Our content policy sets out what is not carried here.

The checks are imperfect

Automated moderation sometimes stops something harmless. When it does, that is our error rather than yours — tell us and we will adjust it.