🔥 Roast Detector
AI-powered toxicity detection with a side of sass
Paste any message and an on-device machine learning model scores how harsh or toxic it sounds — a fun way to check whether that comeback, tweet draft or group-chat reply reads as playful banter or an actual roast. The TensorFlow.js toxicity model runs entirely in your browser, so the text you test is never uploaded.
Frequently asked questions
How does the roast detector work?
It runs a pre-trained TensorFlow.js toxicity model directly in your browser. The model scores your text across categories like insults and general toxicity, and those scores combine into a single roast level. The model downloads once, then all analysis happens on your device.
Is this a serious content moderation tool?
No — it is for entertainment. The underlying model is a real ML toxicity classifier, but it misses sarcasm and context, can flag harmless text, and can miss genuinely hurtful phrasing. Use it for laughs with friends, not for moderating a community or judging real people.
Is the text I enter kept private?
Yes. The model runs locally via TensorFlow.js, so the messages you paste are analyzed on your device and never sent to a server.
Why did it score my message wrong?
ML toxicity models judge word patterns, not intent. Friendly trash talk between friends can score high, while a subtle dig can score low. Treat surprising scores as part of the fun rather than a verdict.