Everything, fully offline.

Medical notes, legal drafts, NDAs, internal code — there are categories of work where 'send to the cloud' is a non-starter.

Offline mode keeps every byte on your machine. No audio uploaded, no text uploaded, no telemetry on transcription content. Air-gapped networks, regulated industries, and paranoid founders all welcome.

Pick per session from Settings → Models. Whisper for accuracy, languages, and AI features. Parakeet for raw speed on CPU.

Same hotkey, same overlay, same dictionary. You just change which model transcribes.

Whisper mode exposes three performance profiles. Each one tunes model size, chunking, and GPU settings — so you trade speed for accuracy explicitly.

Old laptop on battery? Pick Fast. Workstation with a GPU? Accurate gives sharper results. Whisper auto-detects CUDA, Vulkan, and Metal.

Point Whisper at an Ollama instance on localhost and you get the same trigger word, the same presets, the same Always-On mode — but the LLM runs on your machine.

Pick any model you've pulled: Llama 3.1, Mistral, Qwen, whatever suits your hardware. Configure it once, use it anywhere the hotkey works.

Whisper's multilingual models have a built-in translate task. Speak Japanese, Ukrainian, Polish, or any of 90+ languages — paste English. One pass, no extra model, no API call.

Nobody else in this category ships a built-in offline translator. Flip 'Translate to English' in Settings and every dictation switches mode.

Every preset from Cloud mode works offline. Save instructions, stack them, switch them from the overlay.

They run through Ollama instead of OpenAI — same shape, same syntax, same results.

The dictionary works the same offline as online. Add once, applied to every transcription, no tokens spent.

Language-agnostic and model-agnostic. Names, acronyms, product terms — all respected.

The same deterministic filler-word cleaner runs in offline mode. Works with either engine. Works with any language.

No AI needed — just fast pattern matching after the transcription is done.

Select text in any app — Slack, Gmail, Notion, your editor. Hit your dictate hotkey, say what you want done — "make it formal," "cut the corporate filler," "translate to German" — and the selection is replaced in place with the rewritten version.

It's the same trick as polish, but starting from text you've already written. In Local mode the rewrite runs on whichever Ollama model you've configured, never leaving your machine. No second hotkey, no menu, no copy-paste shuffle.

No audio uploaded. No text uploaded. No telemetry on transcription content.

The only network traffic in offline mode is license verification and update checks — and both can be audited with any network monitor.

Monthly is $9. Lifetime is a one-time payment — buy once, own forever, no recurring bill.

Models are free. Inference is free. No per-minute cost, no per-token cost. Only the app is paid.