Frequently asked questions

It reads the text visible on screen in five apps only: WhatsApp, Instagram, Telegram, Snapchat and TikTok. It uses Android’s Accessibility Service to do this. It does not read notifications, contacts, photos, location, files or any other app.

Only an alert summary: the risk level (medium or high), the categories, a confidence score, a short note listing the categories, and the name of the app where the signal was found. Message text, screenshots and the recent-message buffer are never sent. Setup also sends a one-time pairing token and the phone’s device name, and the app downloads model files and checks for updates. See what the AI sends.

No. Analysis happens on the child’s phone, and message text is not sent to our servers in any form. Parents see alerts, not conversations.

There is no automatic deletion yet, so alerts and account data stay until you delete them. Revoking a device removes the alerts it created, and deleting a child removes the profile, alerts and devices. A formal retention policy will be published before general release. Until then, write to support@safecircle.tech with questions.

Go to Settings › Profile, open the Danger zone, click Delete account, enter your password and confirm. This is permanent and removes all child profiles, alerts and data. The parent app has the same option under Settings › Danger zone.

The risk level, the child, the app where the signal was found, the time, the categories, a confidence score, and an AI Reasoning note. For now that note is a short summary such as *Detected: grooming*. It does not quote the conversation. See Understanding alerts.

Only medium and high risk. Lower results are discarded on the phone and never reach you. The thresholds are fixed today, and you cannot set them per category yet.

It is an assistant, not a verdict. Accuracy on real children’s conversations has not been measured. Our published figures come from AI-written test chats: about 94% of risky chats of the kinds the model was trained on were caught, and about 63 to 68% of text from models it had never seen. These figures are for the AI model, not for the basic keyword mode that new installs start with.

Yes. Harmless chats can be flagged, and real risks can be missed. Before you act, ask your child what happened. You can report a problem in the dashboard under More › Report a bug.

Yes. Each app is checked at most once every 5 minutes, using the text on screen at that moment, so a short exchange between checks can be missed. Basic keyword mode misses more than the AI model does.

The detection was built and tested for English and Spanish. Other languages should be treated as unsupported. The child app’s screens are in English for now.

Seven categories: grooming, bullying, sexual content, isolation, requests for personal information, moving a chat to another platform, and threats. Self-harm and scams are not separate categories today. If you are worried about your child’s safety now, go to Help now.

The child app runs on Android 7.0 or later and watches five apps: WhatsApp, Instagram, Telegram, Snapchat and TikTok. There is no iPhone or iPad app. The dashboard works in any modern browser, and the parent app is for Android.

Create the child in the dashboard. It shows a QR code and a 6-digit code. In the child app, scan the QR code or type the code. Codes last 5 minutes and work once, and a new code cancels older unused ones. The tap-to-pair option with NFC is not active yet.

Install SafeCircle Child from Google Play, read the privacy note, pair the phone, then turn on SafeCircle Protection in Android’s Accessibility settings. The status screen should show MONITORING ACTIVE. During the beta, the app is only available to invited testers. The full guide is in Setting up the child app.

Analysis works without an internet connection, because it runs on the phone. Alerts need a connection to reach you, and an alert that cannot be sent when it is created is not sent later.

The Accessibility Service is off, so nothing is being read. Open the child app, tap the status, then Open Settings, and turn on SafeCircle Protection. Android can switch this off after a reinstall or an update, so check it after each update.

Status is based on the device’s last check-in. A device shows Offline after 10 minutes without a check-in, and this can lag behind what the phone is doing. Open the child app’s status screen to confirm MONITORING ACTIVE.

We have measured how fast the AI model runs on phones, but we have not measured real battery or heat impact, and we do not publish battery estimates. In benchmarks on 60 Android phones and tablets, the model took about 3 to 19 seconds to produce its first output, depending on the phone tier, and peaked at about 1.1 to 1.9 GB of memory. Checks are limited to once every 5 minutes per app. If battery saver stops the app, set its battery use to Unrestricted. See the on-device notes.

Yes, and we recommend telling them before you set it up. The child app has a home screen icon named SafeCircle Child. Its accessibility switch appears in Android settings as SafeCircle Protection, and Android shows a warning before it is turned on. During setup the child must accept a privacy note. The child can see their own status and settings, but not alerts or message history.

Yes. The child can turn off the Accessibility permission or uninstall the app, and either stops monitoring. The app has no extra protection against this, and we have no alert for it today. When the permission is off, the child app shows PERMISSION MISSING. The dashboard status can lag, so check the child app’s status screen too.

Nothing changes automatically. The app and the account have no age setting, and monitoring stays on until you change it. We have not built an age-based approach. Whether to continue, and how, is a decision for your family, so revisit it with your teen before any change.

Revoke the device under Children, which removes the alerts it created. To stop monitoring for good, also uninstall the child app. You can also delete the child profile, which removes its alerts and devices.

No. The child app does not ask for access to contacts, location, SMS, the microphone, photos or files, and it does not record calls. Its camera permission is only used to scan the pairing QR code.

Not yet. There is no iOS app, beta or release, and no date has been announced. You can join the waitlist to hear when there is news.

The free plan covers one child and one device. A one-month full trial needs no card. The Family plan costs €4.99 a month or €39.99 a year. See pricing for the current details.

Only if you choose to subscribe. You decide at the end of the trial, and nothing is charged while billing has not started.

Schools use the same child app, with the same Android requirement and the same five apps. Schools are invoiced separately, priced by the number of students, and you request a pilot to get a quote. See for schools and request a pilot.

No. The design is the same as for families: staff see alert summaries, not message text. Ask the school which people can see alerts and how long they are kept before a pilot starts.

Some parts are public: the documentation site, the Orbit browser code, and the Horizon and Vigil model weights on Hugging Face. The child and parent Android apps are not published as source code. Check GitHub for the current list.

Yes. The Horizon and Vigil weights are public at huggingface.co/safecircleai. Read the licence on that page before you use them.

Vigil is a research preview. It is not in the app yet, and it has only been tested on synthetic chats. Horizon is the model that runs in the app today.

No. The models were trained on AI-generated conversations, and message text is never uploaded, so it cannot be used for training. The model improves only when a new model release is downloaded to the phone.