Last updated: 3 November 2026 · applies to Pixie 0.17.3 and later
Pixie's pets can chat with you. The replies are written by a small language model that ships with Pixie and runs entirely on your own computer. This page explains how that chat is kept safe, and what it cannot do.
At the start of every chat, and again at least every three hours while you keep chatting, Pixie shows a note that the pet is an AI — a computer program, not a human — and that its replies can be wrong.
Pixie checks each message you send for clear expressions of suicidal thoughts or self-harm, in all ten of its languages. When it finds one:
What this cannot do. The check looks for clear phrases; it can miss what is said indirectly, and it may occasionally show the note when you were joking. Nobody reads your chats — everything stays on your computer — so no one will be alerted. Pixie is a toy, not a crisis service. If you or someone else is in danger, call your local emergency number now.
Every reply follows rules that the user cannot change or switch off: no swearing or slurs, no sexual or explicit content, no hateful content, no help with illegal or harmful acts, no encouragement of self-harm, no impersonation of real people, and no reproduction of copyrighted works. A second filter on your computer also hides any reply that contains clearly explicit or abusive language. These rules make problems rare, not impossible: the model is small and can make mistakes. Pets have fixed personalities; you can choose a name and a tone, but not rewrite their instructions.
Pixie is not directed to children under 13. AI companions may not be suitable for some minors; parents should decide whether the chat is right for their child.
If the chat said something it should not have, please tell us at support@getpixie.app (a screenshot helps). We use these reports to improve the rules in future updates.