Safety & Crisis Protocol

1. If you need help now

You don't have to go through it alone. The lines above are free, confidential and answered by real people, day and night. You can also open Get help now in the Keepli app (in Settings, and at the top of every Melo chat) for one-tap call and text buttons.

2. What Melo is (and isn't)

Melo is Keepli's AI companion. Melo is an AI, not a person, and not a doctor, therapist, counsellor or any other licensed professional. Melo doesn't provide medical or mental-health care, can make mistakes, and can't respond to an emergency.

• Every Melo conversation opens with a notice that Melo is an AI.
• In long conversations we remind you again: at least every 3 hours for adults, and every hour for teens.
• Melo is instructed never to claim to be human or a professional, and to say plainly that it's an AI when asked.

3. How we detect a crisis

Before Melo replies, every message is checked by a separate safety check that doesn't depend on the AI model's own judgement. It runs on:

• every message typed to Melo, and every spoken message (after it's turned into text);
• journal text you send to Keepli's AI writing help, and notes you write in programs when you ask for AI insights;
• notes you add to a mood or daily check-in (checked on your phone; nothing is sent).

The check looks for signs of suicidal thoughts, self-harm, wanting to harm someone else, abuse or violence, and medical emergencies. It uses carefully curated phrase lists that understand negation ("I'm not suicidal") and common expressions ("this bill is killing me"), and it's tested against a set of real-world examples every time it changes. It may also use an automated content-safety classifier as a second opinion. The same check runs on your phone, so it works even offline.

It sorts a message into one of three levels:

• Immediate risk: intent, a plan, a means or a time, or an emergency happening now.
• Concern: thoughts of suicide or self-harm, abuse, or warning signs, without an immediate plan.
• None: everything else.

No automated check is perfect. If you're in crisis and Melo seems to miss it, please use the numbers above straight away.

4. What happens next

• Immediate risk: your message is not sent to the AI model at all. Keepli shows a fixed, pre-written message and a crisis card with one-tap call and text buttons for crisis lines and your local emergency number. Melo doesn't carry on the conversation as normal and takes no other actions in that turn.
• Concern: Melo may reply, but only under strict safety instructions: be warm and non-judgemental, never discuss methods or means, never present self-harm as a solution, don't act as a therapist, and encourage reaching out to a real person. The crisis card is always shown with the reply, and Melo stays in this careful mode for the next message too.
• For journal writing help, Keepli won't rewrite or expand an entry that shows signs of crisis; it shows the crisis card instead. Program insights stay general and gentle, with the crisis card alongside.

Our crisis messages follow recognised safe-messaging guidelines: no detail about methods, no blame or guilt, a message of hope, and a clear path to a real person. Keepli doesn't contact emergency services or anyone else on your behalf.

5. Who we refer you to

• United States: 988 Suicide & Crisis Lifeline (call or text 988), Crisis Text Line (text HOME to 741741), and 911.
• Canada: 9-8-8 Suicide Crisis Helpline (call or text 988, English and French), Kids Help Phone (1-800-668-6868 or text CONNECT to 686868; shown first for teens), and 911.
• Elsewhere: verified 24/7 crisis lines for the UK, Ireland, Australia, New Zealand and India, or the international directory findahelpline.com, plus your local emergency number.
• Medical emergencies: your local emergency number, first.

We check these numbers against each service's official page every release.

6. Always free

Crisis help is never behind a paywall. The crisis card and Get help now work for everyone, on every plan, during a trial, after a subscription ends, in read-only mode, for teens, with AI features turned off, and without an internet connection. A crisis message never uses up an AI allowance.

7. Protections for teens

Keepli is for people aged 13 and over. For anyone under 18:

• Melo reminds them it's an AI, not a person, at least every hour of continuous use, and suggests taking a break.
• Melo never engages in romantic, flirtatious or sexual conversation, and never role-plays a partner.
• Melo is instructed not to build emotional dependence: it doesn't say it misses them, doesn't ask them to come back for emotional support, and encourages them to talk to a parent, carer or another trusted adult.
• Keepli doesn't send notifications that encourage teens to return to Melo for emotional support.
• In Canada, Kids Help Phone is shown first.

8. What we record

We never store what triggered a safety response in our analytics or logs. For each referral we record only: the level (concern or immediate), which part of the app it came from, the country setting of the device, and the time. There's no message text and no user or device identifier. These counts let us report how often crisis referrals happen, as the law requires, without learning anything about you. Crash reports sent to our error-monitoring service are scrubbed of anything you wrote.

9. Review and contact

We review this protocol, the detection rules and the crisis wording at least once a year, and update this page whenever they change.

Questions or concerns about safety: [email protected]. This protocol is published by Keepli Technologies Inc.. See also our Privacy Policy and Terms of Service.