Per-app dictation tone on a Mac: casual in Slack, formal in Mail, verbatim in your terminal

The same words want different polish depending on where they land. A Slack message can stay loose and conversational; an email usually wants to read a notch more formal; and a command you dictate into a terminal should be typed exactly as spoken — the last thing you want is an AI “improving” a shell command. Most dictation tools apply one cleanup style everywhere. OpenWhisp now sets it per app.

If you’ve used Wispr Flow, this is the feature people rave about — and here it runs entirely on your Mac.

Pick a tone per app

In Settings › Profiles, each app gets a Refine preset:

• Verbatim — no AI pass at all; your exact words.
• Minimal cleanup — just capitalization, punctuation, and spelling.
• Casual — a relaxed, conversational tone.
• Formal — tightened up and more professional.
• Custom — write your own one-line tone instruction.

Turn on “Apply per-app profiles,” add an app, and choose its preset. For a custom tone, a small prompt editor appears below the table so the settings screen doesn’t re-sprawl.

Code editors and terminals stay verbatim — automatically

You don’t have to configure the most important case. OpenWhisp ships knowing the common terminals and code editors — Terminal, iTerm2, Warp, WezTerm, kitty, Ghostty, Alacritty, Hyper, plus VS Code, Xcode, JetBrains IDEs, Sublime, Zed, Cursor, and more — and defaults them to verbatim. A spoken command is never reshaped by an AI. Unknown apps fail open (they just inherit your global setting), and an explicit preset always wins if you want to override the default.

It shapes cleanup — it never translates or invents

The tone presets are deliberately conservative. Casual and formal reuse OpenWhisp’s existing cleanup guardrails with a single added sentence about tone — same language, transform-only, no preamble. They keep the safety check that stops a small model from quietly translating your words: dictate in Russian and the formal preset still hands you Russian. And a preset only ever shapes a cleanup pass that’s already running — if your global AI cleanup is off, a per-app tone quietly does nothing rather than switching the model on behind your back.

How it fits with Modes

If you use Modes (saved dictation styles you invoke by name), the precedence is simple: an active Mode wins, then the per-app preset, then your global cleanup dial. So you can set a sensible per-app baseline and still override it on demand with a Mode when a particular message needs something special.

Local, like the rest of it

Per-app tone runs on the same on-device cleanup model as everything else in OpenWhisp — no cloud call, no account. It’s the kind of small, per-context control that makes dictation feel like it’s actually paying attention to where you’re typing, without giving up the local-first guarantee.