FTAN / browser inference

FTAN-2.5

Detect clean or offensive text with a compact binary classifier running privately on your device. No server-side Python or API call is involved.

Text to analyze

Offensive threshold 50%

Preparing the browser model...

Results will appear here.

Label	Confidence	Status

Limitations: offensiveness depends on context, intent, quotation, sarcasm, and surrounding conversation. This experimental model should not be used alone for high-stakes enforcement. See the model card. The first load downloads a 65 MB quantized ONNX model and caches it in your browser.