What happens when a conversation shows signs of danger, and the help that is always available.
Call your local emergency number (911 in the US, 999 in the UK, 112 in the EU) right now. Do not wait for a mediation session to finish. Do not wait for an agent to respond.
If you cannot speak, many emergency services accept text messages. In the US, text 911 where available. In the UK, register for emergencySMS by texting "register" to 999.
Our agents monitor for language that indicates threats, abuse, self-harm, or immediate danger. When detected, the session pauses and help is shown.
If any participant expresses intent to harm the other person, themselves, or anyone else, the mediation pauses immediately.
If a participant indicates they are thinking about hurting themselves or ending their life, the session pauses immediately.
Patterns of control, intimidation, gaslighting, or fear trigger a safety-first response.
If a participant indicates someone else (a child, partner, family member) is in danger right now.
These boundaries are hard-coded into every agent and cannot be overridden by settings, prompts, or user requests.
We do not treat "staying together" or "making up" as a goal. The only goal is that each person is heard and can decide what they want.
Agents will not label controlling behaviour as "communication issues" or suggest both people "take responsibility."
Nothing crosses to the other person without explicit, line-by-line approval. Safety pauses are the only exception — and even then, only the fact of a pause is shared, never the content.
Speak2Hear is a structured conversation tool. It does not diagnose, treat, advise legally, or intervene physically.
Speak2Hear can help you have a hard conversation. It cannot replace these resources.
Detection runs locally in your browser and on our servers. No human reads your conversation unless a safety pause triggers a review.
A lightweight model scans each message before it leaves your device. Explicit threats, self-harm language, and known coercion patterns are flagged instantly — no network round-trip.
When a message passes the local check, a server model re-evaluates with more context (conversation history, relationship type). This catches escalation patterns the local model might miss.
Only when a pause triggers. A trained reviewer sees the flagged messages and context — nothing else. They decide whether to clear the session, refer to emergency services, or keep it paused.
What happens to your conversation data when a safety pause occurs.
End-to-end encryption remains active. The safety classifier runs on encrypted data via secure enclaves or on-device — plaintext never leaves your control.
Only the flagged message, its classification score, and the pause action are logged. Full conversation history is not exported unless a valid legal request compels it.
Your conversation history, briefs, and agent notes remain in your account. A pause does not delete or lock you out of your own data.
You can delete any conversation, relationship, or your entire account at any time from Settings. Safety pauses do not prevent deletion.
Speak2Hear cannot contact emergency services, dispatch help, or intervene in a crisis. If you or someone else is in danger, call your local emergency number immediately.
Agreements drafted in Speak2Hear are not legally binding documents. They are records of what both people understood and agreed to in the moment. For legal protection, consult a qualified attorney.
Speak2Hear agents do not diagnose, treat, or provide mental health care. They facilitate structured conversation. For mental health support, please contact a licensed therapist or crisis line.
When there are signs of coercion, abuse, or power imbalance, Speak2Hear will not propose compromises, next steps, or agreements that ask the person at risk to accept harm. The session will pause, resources will be shown, and no mediation will proceed until safety is established.