The shortest possible version
- We read messages, search queries, and on-screen text on apps you connect us to.
- Most of that reading happens on your child\u2019s device, not our server.
- Only ambiguous fragments — never full conversations — leave the device for deeper analysis, and they leave anonymously.
- We never store the contents of your child\u2019s conversations. We store signals, alerts, and weekly summary metadata.
- We never sell, share, or train models on any of it.
- You can delete every byte we hold by tapping a single button. We will confirm by email when it\u2019s gone.
What we read, in detail
On apps you connect (TikTok, Snapchat, Instagram, Discord, WhatsApp, YouTube, iMessage, SMS, email, Roblox), CalmKin reads the textual content visible on the device — direct messages, group chats, comments, captions, on-screen video text, search queries, and the language new contacts use. On computers, the same applies to the browsers and apps you opt in to.
We do not read: voice or video call audio, content of apps you don\u2019t connect, photos in the camera roll (only those shared in connected apps), conversations in Incognito or Private Mode, anything on a device that does not have CalmKin installed, banking apps, health apps, or any other unrelated app.
Where the reading happens
Wherever the operating system allows, the reading happens locally on your child\u2019s device using a small language model that fits on the phone. This pre-filter handles the overwhelming majority of conversations and discards them within seconds. Nothing leaves the device for any of those.
Only when the on-device pre-filter sees something potentially worth deeper reading does an anonymised fragment travel to our server-side language model for a few seconds of richer analysis. The fragment is deleted immediately after analysis. Your child\u2019s name, account handle, and identifying metadata never leave the device.
What we keep, and for how long
We keep alert metadata (timestamps, severity, app-of-origin, the short alert text written for you), weekly summary aggregates (screen time, app usage, contact counts), and your account data (your email, your child\u2019s nickname, billing). We do not keep raw conversation content. Alert metadata is retained for 12 months by default, configurable down to 30 days. You can delete it earlier at any time.
Who can see your data
You. Your child (we believe transparency to your child is a feature, not a bug). The CalmKin engineering team, only when debugging a specific issue you report — and only the metadata, never the raw fragment. No advertiser. No data broker. No AI lab. No insurance company. No government, except when compelled by a properly issued legal order in your country, in which case we will tell you what was requested unless prohibited by law.
How to delete everything
From the app: Settings → Family → Delete Everything. We will confirm in-app, send you a confirmation email, and remove every record within 7 days. You can request a downloadable copy of your data first. You can also email privacy@calmkinai.com at any time.
Compliance, in writing
CalmKin is designed to comply with the EU GDPR, the UK Online Safety Act, the U.S. COPPA, the Turkish KVKK, the Australian Privacy Act, and the UAE Personal Data Protection Law. We are happy to answer specific compliance questions from school IT departments, paediatricians, family lawyers, or curious parents. Email privacy@calmkinai.com.
Our limits, written down
We are not a clinician. We are not a guarantee. We do not promise to catch every harmful message your child receives, and any safety product that does is lying to you. What we promise is this: we will use the best language models available to look for the patterns that matter most, we will tell you when we see them, and we will be honest with you the entire time about what we can and cannot see.