Help

When something sticks.

Nearly every issue is one of the three below. If yours isn't, the guide covers normal setup end to end, and obliviox@zohomail.in reaches the people who build RefineIQ.

Microphone permission denied

Voice Draft needs a one-time microphone grant for the extension itself — not for each website. If recording won't start or the pill says microphone access is off:

1.Click the Voice pill's "Allow microphone" action — it opens RefineIQ's own setup tab, the one place Chrome can reliably show the permission prompt. Click Allow there.
2.If you previously clicked Block, Chrome won't re-ask. Open chrome://settings/content/microphone, find the RefineIQ entry under "Not allowed to use your microphone", and switch it to Allow — then retry from the setup tab.
3.On a Mac, Chrome itself also needs the system permission: System Settings → Privacy & Security → Microphone → enable your browser, then restart it.
4.Check the right input device is selected at the top of chrome://settings/content/microphone — a disconnected headset that's still selected records silence.

Audio never leaves your machine: recording and transcription happen on-device, the permission is revocable any time in the same Chrome settings, and nothing can ever start listening on its own.

Model download stuck or failing

The one-time model download (~380 MB for text; the voice model adds ~73–195 MB depending on tier) streams from Hugging Face's CDN and is cached locally. If it stalls or errors:

1.Check your connection — and if you're behind a VPN, proxy, or strict firewall, try without it. The only hosts involved are Hugging Face's model CDN.
2.Make sure there's enough free disk space: roughly 1 GB free is a safe margin for the default tiers. Chrome silently evicts caches on a nearly-full disk.
3.Use the Retry action where the error appeared (popup or the in-page panel) — downloads resume cleanly and progress counts up honestly from where things stand.
4.Still heavy for your machine or connection? Pick the lighter model tier in Settings (~130 MB text, ~28 MB voice) — smaller, faster to fetch, still private.
5.If a download seems permanently wedged, remove and reinstall the extension. That clears the model cache entirely; settings take a minute to redo, and nothing else is stored.

After the download completes once, refining works fully offline — the network is only needed again if you switch to a model you haven't downloaded.

Slow refines & WebGPU requirements

On-device speed depends on whether your browser can use the GPU. RefineIQ uses WebGPU when available and falls back to WebAssembly (slower, but it works everywhere):

1.Use Chrome 113 or newer (or a current Edge, Brave, Opera, Vivaldi, or Arc) — RefineIQ requires 113+, and it's also the WebGPU baseline.
2.Check chrome://gpu and search for "WebGPU" — it should say Hardware accelerated. If it doesn't, enable "Use graphics acceleration when available" in chrome://settings/system and relaunch.
3.On machines without a usable GPU, refines still work via WebAssembly — expect a few extra seconds per refine. The lighter model tier in Settings narrows the gap considerably.
4.Very old or low-memory hardware (≤2 GB RAM) is happiest on the light tiers for both text and voice — that's exactly what they exist for.
5.If refines feel slower than they used to, check that the first refine after browser startup isn't the culprit: the model loads into memory once per session, then subsequent refines are fast.

Prefer a bigger model than your hardware can run? Bring-your-own-key cloud mode (Settings → Provider) sends prompts directly to your own OpenRouter or Hugging Face account — never through a RefineIQ server.

Still stuck? Chat with us — we'll ask for your browser and version, on-device or cloud mode, and what you expected versus what happened. RefineIQ keeps no logs, so that context is genuinely all we have to go on. Never include prompts or API keys.

or email obliviox@zohomail.in