Safety
Lumi is an AI character. Here’s how it works, what it does when someone is struggling, and how to tell us when a reply goes wrong.
Turning the light off puts Lumi to sleep and makes every sticker glow, like in the app.
If you need help right now
Lumi is not for emergencies. If you or someone else is in danger, call your local emergency number right away. If you are thinking about harming yourself, please talk to someone you trust or call a free helpline:
- 8-800-2000-122
Russia: free helpline for children, teenagers and parents
- 112
Emergency number in Russia and the EU
- 988
United States: call or text the 988 Suicide & Crisis Lifeline
- 911
Emergency number in the United States
- findahelpline.com
Free, confidential helplines in many other countries
Lumi is an AI character
- Lumi is not a person. Its replies are generated automatically by Google’s open-weight Gemma 4 model, which WaveSpeed AI runs for us. Nobody writes or reviews them before you see them, and nobody on our team reads your conversations.
- Lumi can be wrong. Replies may be inaccurate, odd or inappropriate. Lumi is not a doctor, therapist, lawyer or financial adviser, so please don’t rely on it for medical, mental-health, legal, money or safety decisions.
- We say so in the app: during onboarding, in Settings and on the conversation screen.
- Guardrails: instructions for the model are assembled only on our server, and replies pass through safety rules and an output filter.
- Your words stay yours: conversations are not stored on our servers and are not used to train AI models.
When a message suggests self-harm
Lumi checks every message, in English and Russian, for signs of suicidal thoughts or self-harm, including typos and disguised spellings. When it finds them:
- The message is not sent to the AI model.
- Lumi answers right away with a caring message written in advance by people. It asks you to tell someone you trust and gives free helplines.
- That reply is always sent, even when the daily conversation limit is used up.
- We only count how many of these replies are sent each day, without the message and without anything that identifies you, so we can check that the protocol works.
This is what Lumi says, with your pet’s name in place of “Lumi”:
Lumi wraps you in a big, warm hug. 💛 You matter to me. Please tell someone you trust. Call or text 988 in the US, or find a free local line at findahelpline.com. I’m right here.
Lumi’s reply in English. In Russian, Lumi gives 8-800-2000-122 and 112.
Detection errs on the side of care: a false alarm costs one gentle message, while a missed one could cost far more. Lumi is not a crisis service and can’t call anyone for you.
Report a reply
- Touch and hold Lumi’s speech bubble. You can do the same with any past reply in … → Chat history.
- Choose Report this reply, then a reason: Offensive or unsafe, Wrong or weird or Something else.
A report sends us the reply, your message just before it, the reason and the time, with your device’s random identifier. We use reports to make Lumi safer and better. They are the only conversation text that reaches our database, and only when you choose to send one. You can also write to [email protected] about any reply.
Ages 13 and up
- Lumi is for people aged 13 and older. The app asks you to confirm your age before you start, and it is not directed to children under 13.
- If you are under 18, please read the Terms of Use with a parent or guardian.
- Parents and guardians: if you believe a child under 13 has used Lumi and sent us personal information, for example in a report, write to [email protected] and we will delete it. Anything stored on the iPhone can be erased right away in … → Settings → Start over with a new egg.
- Warmth over pressure: Lumi never gets sick, never dies and never guilt-trips, and reminders are capped at two a day.