The Problem
Community chats are where scams spread fastest: a phishing link, an impersonated admin, a too-good offer. By the time someone reports it, people have already clicked. The goal was to warn people in real time, before they click.
| Threat | What it looks like |
|---|---|
| Phishing | Links to pages that steal logins or payments |
| Impersonation | Messages pretending to be someone trusted |
| Malicious links | URLs that lead somewhere harmful |
How It Works
The extension scans the chat page as messages render, checks links with on-device heuristics, scores suspicious patterns with a lightweight classifier, and shows a warning before the click.
Key Decision: Everything Stays on the Device
A safety tool that reads your chats has to earn trust. So nothing is uploaded: the URL checks and the classifier both run locally, and suspicious patterns are flagged without transmitting personal data.
| Approach | Trade-off |
|---|---|
| Send messages to a server to check | More powerful models, but your chats leave your device |
| Check everything on the device | Lighter models, and your chats stay yours |
Protection Engine
| Layer | Job |
|---|---|
| DOM scanning | Reads messages as they appear, with no delay |
| URL heuristics | Checks links against known risky patterns, on the device |
| Lightweight classifier | Scores messages for suspicious patterns |
| Warning | Tells the person before they click |
WA-Guardian is an internal tool, so there's no public build or source to link to.
