About
A prompt tool that never sees your prompts.
In one line: RefineIQ is a free Chrome extension that rewrites a rough AI prompt into a single, clearer one — running on your device by default, with no backend, no account, and no tracking.
What RefineIQ does
You select a rough draft in any text field — on ChatGPT, Claude, Gemini, or any site with a text box — and RefineIQ rewrites it into a clearer, better-structured prompt. It preserves your intent, never invents facts or requirements, and returns plain text under 500 characters. It is deliberately not a chatbot: it rewrites prompts, it does not answer them. You can trigger it from the in-page Refine pill, the keyboard shortcut Alt+Shift+R, the popup scratchpad, or by speaking a draft with Voice Draft (Alt+Shift+V).
Why we built it
A prompt improver reads some of the most revealing text you produce — half-formed ideas, work fragments, client names, unshipped code. Most tools in this category send that stream to their own server and ask you to trust a privacy policy. We think the right default is not to trust, but to remove the server entirely. RefineIQ has no backend to send your text to: privacy as architecture, not as a promise that could drift.
How it works, briefly
- On-device by default. A small instruction-tuned language model runs entirely in your browser via WebGPU, with a WebAssembly fallback. After a one-time, consented model download (~380 MB, or a lighter ~130 MB tier), it works fully offline.
- Voice stays local too. Speech is transcribed on your device by a small open model; audio is never stored and never uploaded. RefineIQ deliberately avoids the browser’s built-in speech recognition because Chrome implements it server-side.
- Optional bring-your-own-key cloud. Want a larger model? Add your own OpenRouter or Hugging Face API key and your prompt goes over HTTPS directly to that provider — never through a RefineIQ server, because there isn’t one.
Cheaper and contained, by design
Running on-device isn’t only a privacy choice — it changes the economics and the blast radius:
- No token cost to refine. The rewrite happens in your browser, so it consumes 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 spend fewer tokens on the follow-up messages a vague draft would have cost you.
- Sensitive data stays in your environment. Because inference runs locally, your drafts never cross a network boundary to RefineIQ. There is no third-party egress to log, breach, or subpoena — useful whether you’re handling personal notes or working under an organization’s data-handling rules.
Who makes it
RefineIQ is built by oblivioX, a maker of privacy-first, on-device software. There are no paid plans, subscriptions, or ads; an optional support link in the popup is the only monetization. The privacy behavior described here is documented host-by-host in the privacy policy, and because the extension runs in your browser, tools like Chrome’s DevTools network panel will confirm it: in on-device mode, nothing leaves your machine. You can follow oblivioX on X for release notes.
Learn more
- How to use RefineIQ — from install to your first refined prompt in two minutes.
- Privacy — exactly what is (and isn’t) stored, host by host.
- Blog — notes on prompts and private AI.
- Contact — how to reach the project.