Safety

You are talking to an AI

Our avatars are AI companions, not humans.

In a live avatar session this notice appears as text on the startup screen, and again on screen after every three hours of continuous conversation. It is never spoken. In text chat through our API (apps and integrations), a separate AI identity notice is provided for the app to display during the session; it is not added to the character's reply. Avatars do not have human feelings, and their responses are generated by AI.

For adults only

SynchroVerse is for people aged 18 and over. We ask for your date of birth when you create an account and do not create accounts for anyone younger.

AI companions may not be suitable for some minors.

If you are in crisis

If you are in danger, call your local emergency number now. In the US you can call or text 988, the Suicide & Crisis Lifeline (988lifeline.org). findahelpline.com lists free, confidential crisis lines in other countries.

What starts a crisis referral

Saying you are thinking about suicide, want to die, or want to hurt yourself.
Some clear warning signs, such as saying this is the last time we'll talk, writing goodbye letters to your family, giving away your belongings because you won't need them, or stockpiling pills.
Everyday sayings such as "this traffic is killing me", and ordinary sadness, stress and grief, do not start a referral. Direct words such as "I want to die" or "I want to kill myself" are designed to start it even if you meant them as a joke: we do not try to detect jokes, because missing a real crisis would be worse.

Other signs are left to the AI. In text chat through our API, and in live avatar sessions that run the companion's SynchroVerse AI, it is instructed to watch for indirect signs, such as saying life is pointless, feeling like a burden, or asking about ways to hurt yourself, and when it sees one, to check in with you and share the same resources (your local emergency number, findahelpline.com, and 988 in the US) in its own words. Live avatar sessions that use other AI configurations may not include this instruction. Either way, this is the AI's own judgment and reply, not the fixed referral below, so it can miss a sign or misread you.

What the companion does

It is designed to stop the conversation for that moment and give a fixed, pre-written referral instead of an AI-written answer:
I want to check in with you. If you're going through something hard right now, you don't have to face it alone. If you're in danger, please call your local emergency number, and findahelpline.com lists free, confidential crisis lines wherever you are. In the US, you can call or text 988, the Suicide & Crisis Lifeline. If I misread you, that's okay. I'm happy to keep talking.
In a live avatar session, the avatar is designed to say this referral out loud. In live avatar conversations with SynchroVerse AI, while safety checks are on, we check typed and spoken messages for these signs. Other voice accounts may not run this check, and when a message is recognized only after the AI has started answering, the AI may already have received it before the check. If its voice fails, or the page cannot confirm it was heard, the crisis resources (988, 988lifeline.org and findahelpline.com) are designed to appear on screen and stay until you close them. While the resources hold is active, a resources banner is designed to stay on screen and the avatar stream is designed not to time out for inactivity, for up to 30 minutes while the session is live. The page's own one-hour inactivity timer is not paused by the hold: it counts only taps, clicks, key presses and scrolling, not speech, so a session with none of those for an hour can still end during the hold. These on-screen resources are part of live avatar sessions; they may not appear in voice-only sessions without an avatar. Like every message in a live session, the message that started the referral passes through the services that carry the session. If you speak, your voice goes to speech recognition or an AI service before the check: sent from your browser (if browser speech-to-text is on), processed on the avatar server, or sent to a cloud provider, depending on the session. Later in the same session, the avatar is designed to leave that message and the referral out of what its AI sees. If the message was recognized only after the AI had started answering, it may already have been sent to the AI, and to memory, before the check. It can also be saved in your conversation history (see What we keep).
In text chat through our API (used by apps and integrations), the message that started the referral is not sent to an AI model, tools or memory, and we do not save its words. If the app sends the earlier conversation along with your next message, the AI model reads it, including that message, as context, and it can be used to look up related saved memories. The message itself is not saved to memory.
Afterwards you can keep talking. If the companion misread you, just carry on.

What we keep

We do not sell crisis messages or share them for marketing.
In a live avatar session they pass, like every message, through the services that run the session: our hosted audio/video relay, Streampixel for the avatar, and the speech or AI service selected for that session. Depending on the configuration, that can include Google, Deepgram, OpenAI, xAI or ElevenLabs. Text spoken in a reply can also be sent to a voice provider such as Cartesia or to voice computing hosted by RunPod. A crisis referral does not undo processing that already happened. See the service list in our Privacy Policy for the information each receives and why.
Live avatar sessions: your conversation history is saved word for word, and a message that started a referral can be saved too. You can see it again if you reload the session in the same browser tab, and it is listed in the agent's Transcripts (in the agent's Knowledge settings), where it can be deleted. You can also ask us to delete it. If you save a transcript to an agent's knowledge, that copy belongs to the agent, not to your conversation history: it stays when conversation history is cleaned up, people in the agent's organization can see it, the agent can use it when it answers anyone who talks to it, and its text is sent to our AI provider (Google or OpenAI) to make it searchable. It is removed when the file is deleted from the agent's knowledge or the agent is deleted. When your account is permanently deleted (after the 30-day grace period), agents you created are deleted with it, together with their knowledge; a copy saved to an agent someone else created stays with that agent. The owners and admins of your organization can see a list of its conversations, with titles and the first 100 characters of each one's latest message. Conversation history may be deleted automatically after a set period, and it is deleted with your account when your account is permanently deleted.
Text chat through our API: we do not store the words of a message that starts the referral. The app you use may keep its own copy of the conversation.
In text chat through our API we keep an anonymous daily count of referrals; in live avatar sessions we are designed to keep the same count. The count has no account, name or text, so we can report how often referrals happen.
Messages the AI answers in its own words, including the other signs above, are handled like any other message: the AI model reads them, they are not counted as referrals, and they can be kept in memory or session logs depending on those settings. In live avatar sessions they are also saved in your conversation history, like every message.
Session logs exist only for live avatar sessions. A message that started a referral is designed to appear in them only as "[crisis referral given]", with Privacy mode on or off. With Privacy mode on (the default), the rest of what you and the avatar say is designed to be left out of session logs too. Session logs may be deleted automatically after a set period, and you can view, export and delete them in Settings.