ToxiFilter
Real-time tone & toxicity analysis next to any text field. On-device ML.
Real-time tone analysis as you type — green, yellow, red indicator
On-device ML model (LiteRT.js) — 100% offline, no data sent to servers
Pre-submit warning modal for toxic content
Adjustable sensitivity levels
Curated lexicon detector catches profanity the ML model misses
100% free, no premium, no ads, no tracking
ToxiFilter analyzes the tone and toxicity of your text in real-time, right next to any text field. A green, yellow, or red indicator tells you if your message might come across as hostile or toxic — before you hit send.
How it works
ToxiFilter combines two detection systems:
-
ML sentiment model — An AverageWordVec text classifier (752KB) running on-device via LiteRT.js. It analyzes the overall sentiment of your text and outputs a positive/negative probability score.
-
Lexicon detector — A curated list of profanity, insults, and hostile phrases. This catches specific language the sentiment model might miss.
The tone analyzer fuses both signals into a simple traffic-light indicator:
| Level | Condition |
|---|---|
| 🔴 Red | High toxicity detected |
| 🟡 Yellow | Some concerning tone |
| 🟢 Green | Text reads as neutral or positive |
Privacy
ToxiFilter runs entirely on-device. The ML model is bundled with the extension (752KB) and inference happens in your browser via LiteRT.js (WebGPU/WASM). No text is ever sent to any server. No analytics, no tracking, no telemetry.
Free forever
ToxiFilter is 100% free. All features are available to everyone. There are no premium tiers, no locked features, no usage limits. If you enjoy it, you can buy me a coffee — but there’s zero pressure.
How would you rate ToxiFilter?
Click a star to vote — no account needed.
Frequently asked questions
Is ToxiFilter really free?
Yes, 100% free. All features are available to everyone. No premium tiers, no locked features, no usage limits. If you enjoy it, you can buy me a coffee — but there's zero pressure.
Does ToxiFilter send my text to a server?
No. ToxiFilter runs entirely on-device using LiteRT.js (TensorFlow Lite for the web). The ML model is bundled with the extension — nothing is downloaded or uploaded.
How does the toxicity detection work?
ToxiFilter uses a combination of a sentiment analysis ML model (AverageWordVec text classifier) and a curated lexicon detector. The ML model catches general negativity, while the lexicon catches specific profanity, insults, and hostile phrases the model might miss.
What do the colors mean?
Green = your text reads as neutral or positive. Yellow = some potentially concerning tone detected. Red = high likelihood of hostile or toxic content. You can adjust the sensitivity in the popup to make the thresholds more or less strict.
Will ToxiFilter slow down my browser?
No. The ML model is tiny (752KB) and inference takes milliseconds. Analysis is debounced — it only runs when you pause typing, not on every keystroke.
Can I use ToxiFilter on any website?
Yes. ToxiFilter works on any website with text fields — social media, email, forums, chat apps. It detects textarea, input, and contenteditable fields automatically.