What our AI is — and what it isn't.

A warm, non-judgmental conversational companion trained to listen, validate, gently reflect, and suggest small grounded steps. It works in 10 languages and adapts its tone to your emotional state.

It is not a clinician, a substitute for therapy, or a crisis service. It cannot diagnose, prescribe, or provide medical advice. For clinical care, please use our verified human therapists or your local mental-health professional.

We use Google Gemini and OpenAI GPT family models routed through our gateway. Models may change as better ones become available; the principles on this page stay constant.

The companion writes short third-person notes about you — facts, themes, goals, relationships you share — so it can be a continuous presence instead of a stranger each time. Every memory is visible to you on the Me page and can be deleted individually or all at once.

We never train external models on your data. Your conversations are not used to improve any AI model belonging to us or any provider.

A separate classifier scans each message for signs of self-harm or imminent risk. If it triggers, the conversation immediately surfaces helpline numbers in your country and a one-tap crisis button. The classifier never blocks you from talking — it only adds support on top.

Language models can be wrong, biased, or inconsistent. The companion may occasionally say something inaccurate or culturally tone-deaf. When you notice it, please flag the message — that feedback goes to our care team, not to model training.

Our care team reviews flagged conversations and crisis events. They never browse private chats opportunistically. All access is logged and audited.