Stop typing your AI prompts.
TongueType is a Mac voice dictation app that works in VS Code, Cursor, the terminal, and your AI chat panes. Hold a key, ramble a prompt, release — the text lands at your cursor. Everything runs on-device, so your code and prompts stay on your Mac.
macOS 14+ · Apple Silicon · Free forever, Pro if you want
What macOS dictation gets wrong
Apple’s built-in dictation has improved a lot, but if you spend your day inside developer tools and AI chat apps, you’ve probably noticed it gets flaky:
- Inside Electron-based editors (VS Code, Cursor, Windsurf, the Claude and ChatGPT desktop apps, Slack, Discord), built-in dictation often inserts only the first word and drops the rest, or doesn’t insert anything at all.
- In the Terminal and iTerm, dictation results are unpredictable. Sometimes it works. Sometimes it doesn’t.
- Technical vocabulary — library names, identifiers, API terms, command-line flags — gets mangled into the nearest common word.
- You can’t train a custom vocabulary, so you correct the same things over and over.
TongueType sidesteps these because it doesn’t hand text to the OS via the same code path. With Accessibility permission granted, it types the transcript at the cursor directly. If you skip that permission (which the app treats as Optional), the transcript lands on the clipboard and you press ⌘V — which is exactly how Electron apps want you to insert text anyway.
Where it works
Code editors
VS Code, Cursor, Windsurf, Zed, JetBrains, Xcode, Sublime, Nova. Hold the hotkey, dictate a code comment or commit message, release. The text appears where your cursor is.
AI chat apps
ChatGPT, Claude, Gemini, Perplexity, and the desktop apps for each. Talk out a multi-paragraph prompt instead of typing it. Your prompt goes straight to the chat box.
Terminals
Terminal, iTerm2, Warp, Alacritty, Ghostty, kitty. Useful for long commit messages with git commit -m, multi-line shell heredocs, and verbose error descriptions.
GitHub, GitLab, Jira
PR descriptions, issue comments, code reviews. Web forms and Electron desktop apps both. Anywhere your cursor lands, TongueType can deliver text.
Team chat
Slack, Discord, Microsoft Teams, Linear. The same Electron caveat applies — built-in dictation is unreliable here, TongueType is not.
Notes & docs
Notion, Obsidian, Bear, Apple Notes, Drafts, iA Writer. Capture an idea before it disappears, then go back and edit. Faster than typing, by far.
Where it’s actually faster than typing
Long AI prompts
The single biggest win. You think faster than you type, and AI assistants reward context, so the more you tell them, the better the answer. Holding the hotkey for thirty seconds and talking out the full problem — what you tried, what failed, what you actually want — produces a better prompt than the half-baked sentence you would have typed instead.
This is the moment to turn on Double-tap to latch (under Settings → General). Two quick taps start recording, one more tap stops it. No holding required while you think.
Commit messages and PR descriptions
The good ones include why, not just what, which means more sentences than anyone wants to type. Dictating a commit message takes about ten seconds and ends with a paragraph instead of a sentence. PRs benefit even more.
Code comments and documentation
The friction of typing kills documentation discipline. Talking it out kills the friction. Dictate the comment, fix the punctuation, move on. Same for JSDoc blocks, docstrings, and README paragraphs.
Code review feedback
GitHub and GitLab inline-review comments are exactly the kind of short paragraphs that take longer to type than to think. Hold the hotkey, explain why this approach worries you, release. The reviewer who actually leaves substantive notes is faster than the one who types.
Debug logs and bug reports
“What happened, what I expected, what I tried” is a paragraph nobody likes typing. It’s also the format that gets bugs fixed.
Technical vocabulary, your way
OpenAI’s Whisper model is pretty good on technical terms out of the box — far better than Apple’s. But it doesn’t know your codebase, your library choices, or your team’s acronyms.
TongueType’s postprocessing rules close that gap. Map whatever Whisper writes to whatever you actually want. A few examples:
- “Kubernetes” stays Kubernetes, not “cucumbers.”
- “to-do” becomes
TODO. - “open bracket” becomes
{, “close bracket” becomes}. - Your library names, your team’s product code names, internal acronyms — map them all once and forget about it.
Rules live on your Mac and sync across your devices via iCloud if you opt in.
Your code stays on your Mac
TongueType runs the Whisper model locally on Apple Silicon’s Neural Engine. There’s no transcription API behind the curtain. Your prompts, your code, your unreleased side project, the internal API endpoint you accidentally said out loud — none of it gets uploaded anywhere.
This matters more for developers than most other dictation users. Voice transcription services routinely log audio for “quality improvement,” and an AI prompt full of proprietary code is exactly the kind of thing you don’t want sitting in someone else’s training-data bucket.
If you’re curious about the technical side, the file-transcription page covers the same on-device guarantee for audio files, and the privacy policy spells out the rest.
Stop typing what you could say.
Free to download, free to try, and your prompts never leave your Mac.
macOS 14+ · Apple Silicon · Direct download, no App Store