Comment Vibe reads the tone of your comment as you type — on LinkedIn, X, YouTube, Reddit, anywhere — and flags it right next to the box. Harsh or toxic? You get a kinder rewrite in one click.
no account · no cloud · your comments never leave the browser
↑ simulated preview — the installed extension runs real on-device AI (Gemini Nano / Firefox AI runtime)
Works on any site with a comment field — social networks, forums, docs, review forms. No setup per site.
A small badge appears while you type: 😊 positive, 😐 neutral, 😕 negative or 🚫 toxic — with a one-line why.
Sounds harsher than you meant? Click the badge and copy a kinder rewrite that keeps your point.
“Great write-up, thanks for sharing! The section on caching finally made it click for me.”
“The update ships on Tuesday. Release notes are linked in the docs.”
“This is a terrible idea and it will never work in practice.”
“Only an idiot would post something like this. Delete your account.”
Every analysis runs locally — Chrome's built-in Gemini Nano model, or Firefox's on-device AI runtime. There is no server, no account, no analytics, and it keeps working offline once the model is downloaded. The code is open source (MIT).
Chrome uses the Prompt API (Gemini Nano) — a generative model, so you get rewrites, streaming and translation. Firefox's brand-new WebExtensions AI API exposes on-device classifiers, so the Firefox build focuses on the tone badge. Comment Vibe is one of the first extensions shipping on both.
| Feature | Chrome 138+ | Firefox 134+ |
|---|---|---|
| Tone badge (positive / neutral / negative / toxic) | ✓ | ✓ |
| “Why” explanation | ✓ model-written | ✓ classifier + confidence |
| Kinder rewrite suggestion | ✓ | — |
| Streaming (badge colors in early) | ✓ | — |
| Understands non-English comments | ✓ | ✓ ~100 languages |
| Output in your comment's language | ✓ | — (English) |
| Runs 100% on-device | ✓ | ✓ |
No. Analysis happens inside your browser using its built-in on-device model. There is no backend, no telemetry, and the extension keeps working offline once the model is downloaded. You can audit the code — it ships as raw, unminified JavaScript.
Chrome 138+ on desktop and hardware eligible for Gemini Nano. Chrome downloads the model automatically the first time it's used — no flags needed on current versions.
Firefox 134+ on desktop. Click “Enable on-device AI” in the extension popup — a multilingual classifier model (~340 MB) downloads once. On Firefox 134–139 you may also need to set browser.ml.enable and extensions.ml.enabled to true in about:config; from Firefox 140 they're on by default.
It's not a lesser effort — it's a different engine. Chrome ships Gemini Nano, a small generative LLM that extensions can prompt, so it can write things: explanations, kinder rewrites, translations. Firefox's WebExtensions AI API (browser.trial.ml) instead exposes on-device classification models — they can judge tone (in ~100 languages) but can't write text. Same badge, different brains. When Firefox's generative backend matures for extensions, the rewrites follow.
Yes — free and open source under the MIT license. No paid tier, because there's no server to pay for.