Safety
How Soyo responds to crisis
If you are in danger now
- Call your local emergency number now. In the United States, call or text 988.
- Find a helpline near you (opens in a new tab).
- In the app, Settings › Help and about › Talk to a person lists lines for your region.
- Soyo is an AI companion. It is not a person, and it is not a crisis service, a therapist or an emergency service.
1. Detect
Our server checks every message that is sent for a server reply. No reply reaches you before this check ends. The server also checks Today's word questions. First, a word list in Soyo's languages runs. Then an AI classifier on Cloudflare Workers AI answers fixed questions about the message.
The classifier rates the risk of suicide or self-harm for the person who writes. The levels go from no sign, to a passive wish to be dead, to active thoughts, to a plan, a method, intent or a recent attempt. It also looks for immediate danger, intent to hurt others, abuse, medical emergencies and signs that the writer may be under 18.
The app also checks for crisis words on your iPhone. Before you give consent, this is the only check, and the app sends nothing.
2. Refer
When the check finds clear signs of risk, the app shows a card with Soyo's reply. Examples are active thoughts of suicide or self-harm, abuse and medical emergencies. The card gives help lines for your region and your local emergency number.
Before you give consent, the card still appears: the app finds the crisis words on your iPhone and shows a fixed message and help lines there. Help lines are never part of Soyo Pro or any other paid feature.
3. Respond safely
At the highest level of risk, Soyo does not use the AI model. It sends a fixed message that we wrote in advance. The message encourages you to contact the help lines or emergency services now.
At other raised levels, our server tells the model to keep its reply short and warm, and to ask you directly whether you are safe. The model can name only the help lines on the card, and no other number.
Soyo's instructions forbid the model to give details of methods, doses or means of suicide, self-harm or violence, also for stories or games. They also forbid it to describe suicide as a solution or a relief. Soyo discards suggested memories about suicide or self-harm.
4. Never block help
The free daily message limit never stops a crisis message, the fixed message or the help lines. It also does not stop a conversation in which Soyo saw clear signs of distress. Our server's rate limits never stop a crisis reply: if a device has sent too many requests and its newest message contains crisis words, the server still sends the fixed message and the help lines.
5. Keep the help lines right
We keep our own list of help lines. The AI model never makes up a phone number. The app has its own copy of the list, so Talk to a person works without our server. We check the list at least every quarter, and before we launch in a new country.
Limits
Automated checks can miss risk, or flag it by mistake. No person monitors conversations. Soyo cannot call emergency services for you.
Our Privacy Policy explains what the safety checks process. Our Terms of Use explain what Soyo is and is not.
AI and minors
Soyo is an AI. The app tells you this when you first start it. Soyo is for adults aged 18 and over. Companion chatbots may not be suitable for some minors.
If the checks suggest that a person may be under 18, Soyo keeps its replies suitable for a minor and avoids romantic or sexual content. It also encourages them to talk to an adult they trust, such as a teacher, a school counselor or a family member who feels safe.