AI voice assistant
Features
Dictate anywhere
Hold a hotkey, speak, the words appear at your cursor. Mac and Windows, in any app you can type into — Slack, Gmail, VSCode, Notes, your Jira ticket, your kid's school form.
Local mode
Fully offline. Transcription on OpenAI Whisper or NVIDIA Parakeet, AI cleanup on Ollama with whichever open-source model you trust. No cloud, no servers, nothing leaves your machine — for the times your audio is not for sharing.
Cloud mode
Best-in-class accuracy and live web answers in one key. Transcription via OpenAI's gpt-4o-transcribe at $0.003/min — about 18¢ per hour of audio. The same key powers real-time facts via the Responses API. You bring the key, we take zero markup.
Search the web
Mid-sentence, ask Whisper to look it up — stock price, weather, last night's score, the spelling of your colleague's hometown. The answer arrives formatted and inline. No tab switch, no losing your train of thought.
Polish as you speak
Ramble in, structured out. "Hey Whisper, format as email" turns your mumbled draft into a three-paragraph email with a greeting and a sign-off. Write your own trigger for any context.
Translate as you speak
Speak Polish, paste English. Speak English, paste Mandarin. Speak Japanese, paste German. 90+ languages, both directions, in cloud mode and local mode — no extra app, no extra subscription, no Google Translate tab.
Ask about what's on your screen
Hover your cursor over an error, a chart, a UI bug — hit the dictate hotkey and ask. The screenshot under your cursor goes to OpenAI alongside your voice prompt, the answer pastes back inline. Image cost is well under a cent per question. Cloud mode, Pro feature.
Mute your music when you talk
Spotify mid-album, a YouTube tutorial open in a tab, a podcast playing on the side — Whisper mutes whatever your system is playing the moment you start recording, restores the volume the second you stop. Pair it with a soft start-and-stop cue in the same section if you want eyes-free feedback.
Rewrite what you've already written
Select text in any app — Slack, Gmail, your editor — hit the dictate hotkey, and speak the rewrite: "make it formal," "cut the corporate filler," "translate to German." The selection is replaced in place. Same hotkey, no menu. Works in Cloud mode and Local mode.
Instructions on a hotkey
Save instruction packs for any context — developer prompts, formal email, friendly Slack updates, school-newsletter bullets. Bind the first nine to Ctrl/Cmd + 1..9 and switch the active style mid-recording without leaving your app.
Your words, spelled right
Add brand names, acronyms, your colleague's surname, the German town you keep mistyping — once. Every transcription respects them, cloud or local, no AI tokens spent.
Um, uh, like — gone
One toggle, every "um" and "like" stripped on the way out. Deterministic, free, works with every engine. Now your dictation sounds like the third draft, not the first.