Guard
Scammers call. Guard answers.
Phone fraud works because it takes your time and your attention. Guard takes theirs.
How Guard works, step by step.
Detection runs on your phone first. The cloud is used only when you opt in or when the on-device engines already see real risk.
- 01
Before it rings
On your phoneThe number is checked against a small on-device cache of moderator-reviewed scam numbers. Known scammers can be rejected or silenced before the phone rings.
- 02
Passive monitoring
On your phoneOnce connected, acoustic cues such as speaking rate, relentless pressure and synthetic-sounding speech artifacts are measured locally. No audio leaves the phone.
- 03
Live Guard
Cloud, opt-inYou turn it on from its own consent screen. Speech is transcribed on-device first. If that fails, a few seconds of audio go to a protected server function, only to be turned into text, and are not stored.
- 04
Pattern engines
On your phoneRule packs in 20 languages and a concept engine look for how scam scripts are built: a request for a code, a push to a safe account, urgency. One word never triggers an alarm; the combination does.
- 05
Second opinion
Cloud, opt-inOnly when the on-device engines already see real risk, a language model reviews the transcript through an App Check protected function and returns a structured verdict. If it fails, the on-device verdict stands.
- 06
Action
On your phoneA clear on-screen warning, the Banking Shield, and one tap to hand the call to Guard, block, or report the number.
Guard Hand-Off: the AI that wastes their time.
When you tap Hand-Off, your microphone is muted and Guard talks to the caller for you. Every minute a scammer spends here is a minute they are not spending on a real victim.
Hand-Off starts only when you tap it. After an audible notice to the caller, a short evidence recording may be kept and is deleted within 15 days. It is never used for advertising and never used to retrain AI models.
How Cally Neutralizes a Real Scam Call
Watch our interactive 3-act cinema simulation showing smishing quarantine, banking app lock, and Guard AI voice takeover.
Act 1: Fake Delivery SMS (Smishing)
Scammers start with phishing SMS trying to seize credentials. Cally's on-device NLP neutralizes the domain locally before the user can click.
What Guard looks for.
Turkish, English, German, Spanish, French, Italian, Portuguese, Russian, Ukrainian, Polish, Dutch, Romanian, Arabic, Hindi, Urdu, Indonesian, Vietnamese, Japanese, Korean and Chinese.
Catches paraphrases. Fires only when concepts combine the way a scam script does, and stays quiet for protective advice like never share the code.
Recognises verification codes in messages and calls across many languages.
Flags fake parcel and account-verification texts and the links they carry, using rules the phone evaluates offline. Message bodies never leave the device.
Speaking rate, turn-taking pressure, radio-static effects used in fake police calls, and synthetic-voice artifacts, measured on-device.
Community votes alone are never presented as proof. A number is marked as confirmed only after a trusted moderator reviews it.
With Usage Access granted, Cally warns you at the system level if you open a banking app during a suspicious call. It does not keep or upload your app history.
Bring the people you protect into the same shield.
Where your data goes.
Plain answers, matching our Privacy Policy.
An honest note on accuracy.
We do not publish a detection accuracy percentage, because we have not yet validated one independently, and a number without a method is marketing. Guard's detection is a layered, rule and signal based system with an optional language-model second opinion. Its synthetic-voice indicators are acoustic heuristics, not a certified deepfake detector. No system catches every scam. Never share codes, passwords or card details by phone, whoever is asking.