Chrome extension — free & private
Refine prompts. Privately.
RefineIQ rewrites your rough AI prompt into a clear, well-structured one — on your device by default. No account, no tracking, no backend.
// works with ChatGPT, Claude, Gemini — and any text box
01 — In action
Invisible until you need it.
01 / 06 — Select text — the pill appears where you work.
02 — Voice Draft
Or just say it.
Focus any prompt field and a small Voice pill appears — press it (or ⌥⇧V / Alt+Shift+V) and speak your rough draft. A live waveform and a growing transcript preview show it's hearing you.
Two taps from spoken to sent
The transcript lands at your cursor pre-selected, so the Refine pill is already waiting. Speak, refine, send.
Hands-free by default
Hold to talk or tap to toggle — and once you've spoken, ~3 seconds of silence finishes the recording on its own. Hard-capped at 60 seconds, never discarding your speech.
Nothing to play back — by design
Audio exists only in memory while it's transcribed, then it's gone. No recordings, no files, no cloud speech service. The multilingual model runs entirely on your device.
// on-device speech-to-text — multilingual by default, audio never leaves your machine
03 — Capabilities
Everything a prompt needs. Nothing it doesn't.
Refine in place
Select text in any field and a fluid Refine pill glides in — no modal, no context switch. Review the rewrite, hit Replace; your original stays one Esc away.
⌥⇧R / Alt+Shift+ROn-device AI
A small language model runs entirely in your browser via WebGPU, with a WebAssembly fallback. One consented download, then it works fully offline.
SmolLM2 · ~380 MB, onceVoice Draft
Speak your rough draft into any prompt field. A live waveform shows it's hearing you, the transcript lands at your cursor pre-selected — and transcription runs entirely on your device.
⌥⇧V / Alt+Shift+VA scratchpad that always works
A reliable refine surface in the toolbar popup, immune to hostile page editors. Type or paste a draft, refine, copy — every time.
in the popupBring your own key
Prefer a larger model? Connect OpenRouter or Hugging Face with your own API key. Prompts go directly to your provider — never through a RefineIQ server. There isn't one.
optional · off by defaultWorks everywhere
Purpose-built write-back for ChatGPT, Claude, Gemini, AI Studio, and Perplexity — plus any site with a text box, and per-site control when you want it off.
+ ProseMirror · Lexical · Slate · Draft.js04 — Three steps
Installed to refined in two minutes.
Add to Chrome
One click, no account, nothing to configure. Free — requires Chrome 113+ (the WebGPU baseline).
Set up the model
Open the popup and start the one-time, consented download (~380 MB). It's cached locally — RefineIQ then works fully offline.
Select & refine
Highlight your rough prompt, click Refine, replace. Or press ⌥⇧R and keep typing — the rewrite lands where you work.
05 — Privacy
No account. No tracking. No backend.
There is no RefineIQ server — nothing for your prompts to be sent to. By default everything runs on your device and works offline. If you opt in to a cloud provider, your prompt goes directly to that provider with your own key. We never see it; there is no code path that could.
0
servers behind the product
100%
on-device by default
1
optional key — your own
06 — FAQ
Asked, answered.
What does RefineIQ do?
It rewrites a rough prompt into a single, clearer one — built to preserve your intent and add nothing you didn't say; always plain text, under 500 characters. Select text in any field and click Refine, press Alt+Shift+R, or use the scratchpad in the popup.
Is it really private?
Yes. By default everything runs on your device and nothing you type leaves your machine. There is no RefineIQ server, no account, and no tracking of any kind.
Do I need an API key or an account?
No. On-device mode needs no key and no account — just a one-time model download. A key is only for the optional bring-your-own-key cloud mode, and it stays on your device.
How big is the model download?
Roughly 380 MB for the default model (a lighter ~130 MB tier is available in Settings), downloaded once and cached. On a machine with WebGPU it loads in a few seconds thereafter.
Does it work offline?
In on-device mode, yes — fully, after the one-time download. Cloud mode needs a connection, since it calls your own provider.
Which sites does it work on?
Any page with a normal text field, textarea, or editable box — with purpose-built support for ChatGPT, Claude, Gemini, AI Studio, and Perplexity. You can turn it off for specific sites under Settings → Sites.
Is my voice recorded?
Voice Draft is push-to-talk and always local: speech is transcribed by a small model running entirely on your device. Audio is never stored and never sent anywhere — RefineIQ deliberately avoids the browser's built-in speech recognition because Chrome implements it server-side.
Is it free?
Free, with no paid plans, subscriptions, or ads. If it saves you time, there's an optional support link in the popup — that's the only monetization.
Does it save tokens or money?
Yes, in two ways. On-device refining is free and uses no AI provider tokens or credits — you're not paying an API to clean up a prompt. And a clearer, tighter prompt gets a usable answer on the first try, so you burn fewer tokens on the back-and-forth of fixing a vague one.
Can my team use it with sensitive data?
That's the point of on-device mode: your drafts are processed by a model running in your own browser, so sensitive text stays inside your environment and is never sent to RefineIQ — there is no RefineIQ server to receive it, log it, or leak it. Cloud mode is strictly opt-in and, when enabled, sends prompts directly to the provider you chose with your own key.
Which browsers are supported?
Chrome and Chromium browsers on desktop — Edge, Brave, Opera, Vivaldi, Arc. Chrome 113 or newer is required (it's also the WebGPU baseline; devices without WebGPU fall back to a slower on-device path). A Firefox build is maintained for portability.
07 — Install
Ready when you are.
Install RefineIQ and send a clearer prompt on your very next draft — free, and private by default.
Free. Private. On-device.