/modes/dictationdictation mode

Talk like yourself. Type like you meant it.

⌘⇧␣default hotkey — yours in Settings

Hold the key, speak at normal speed, let go. Clean, punctuated text lands wherever your cursor was — in any app on your Mac. The difference from every dictation tool you've tried before: it knows where it's typing, and it edits like someone who writes for a living.

Anatomy of a press

  1. 1
    Hold, speak, release

    Push-and-hold by default (or toggle mode: press to start, press again to stop). A live level meter shows you’re being heard, and Esc cancels mid-recording without spending an API call.

  2. 2
    Whisper transcribes

    Audio goes to Groq’s whisper-large-v3-turbo — fast, and spooky accurate on accents. Prefer the full model or English-only distil? Both are a dropdown away, as is pinning a language.

  3. 3
    Haiku edits like a copyeditor

    The transcript passes through Claude Haiku with one job: remove filler, add punctuation, resolve self-corrections ("let’s meet Tuesday — no wait, Wednesday"), fix homophones from context, and preserve your meaning exactly.

  4. 4
    It checks which app you’re in

    The frontmost app shapes the tone: polished sentences for email, contractions and lowercase energy for chat, technical terms with backticks for editors, clean prose for notes.

  5. 5
    Text lands at your cursor

    Pasted in via the clipboard, which is restored immediately after — images and files on your clipboard survive intact. Your words appear; nothing else changes.

You say it like this. It lands like this.

In Gmail — polished
hi sarah, um, just following up on the proposal, no rush but like, if you could look at it this week that would be great, thanks
Hi Sarah — just following up on the proposal. If you could take a look this week, that would be great. Thanks!
In Slack — still you, just tidier
omg yes, that deploy was, uh, that was rough, lol, anyway it’s fixed now, don’t touch anything for a bit
omg yes, that deploy was rough lol. Anyway it’s fixed now — don’t touch anything for a bit

Casual stays casual. The editor removes noise, not personality.

In Cursor — technical terms intact
the bug is in the parse config function, it’s returning undefined when the yaml is empty
The bug is in parseConfig() — it returns undefined when the YAML is empty.

Code editors get backticked identifiers automatically. Say the function name; get the function name.

It learns how you talk

Fix a word it got wrong a few times — your cofounder’s name, your product’s spelling — and Wisspa starts auto-applying that correction. Add custom vocabulary for jargon Whisper shouldn’t have to guess at. And per-app profiles let you set tone and vocabulary per tool: formal in your email client, unhinged in the group chat, precise in the editor. Calibration is a one-minute onboarding step, and the silence guard uses Whisper’s own confidence scores — so quiet speakers don’t get thrown away, and dead air never reaches the transcriber.

Boring reliability, on purpose

Empty recording? Silently dismissed. Network hiccup? One automatic retry, then a graceful error — never a lost utterance. Three-minute rant? Processed, with a heads-up. Every dictation lands in your local history — searchable, filterable, re-injectable with one click, exportable to CSV, pruned automatically, and never anywhere but your Mac. Prefer to approve before insert? There’s an edit-before-insert review window one toggle away.

Tune it in Settings

Recording mode (hold vs toggle), input device, mic sensitivity, STT model and language, max recording length (10–120 s), sounds and volume, the completion chime, review-before-insert, per-app profiles, vocabulary, and word corrections. Everything’s a dropdown or a slider; nothing requires a config file.