Claude Code lives in a terminal, and a lot of people assume dictation stops at the terminal door. It doesn’t. A terminal is a text field like any other: put the cursor there, hold the key, talk.
What’s different
Two things. First, terminals don’t wrap or format, so long instructions arrive as one line. That’s fine for Claude Code, which reads it the same way. Second, you’re usually giving a task, not chatting, so the structure matters more than in a message.
Use the Smart key
Speaky Box’s Smart key treats a terminal running an AI agent the same way it treats Claude’s desktop app: it turns what you said into a structured brief. Task, context, constraints, done-when. Claude Code does noticeably better with that than with a stream of thought, because it plans from the brief.
For quick replies inside a session (“yes, do that”, “the second option”), use the Dictate key. Plain words, nothing rearranged.
Settings that help
- Add your repo’s service names, file names and libraries to the Words list. Terminals show every typo.
- Pick a key that isn’t used by your terminal. Right option or caps lock are safe; fn can conflict with some terminal emulators.
- If your terminal uses vim keybindings, make sure you’re in insert mode before you hold the key.
Works in
Terminal, iTerm2, Warp, Ghostty, Kitty, and the integrated terminals in Cursor and VS Code. Anywhere the cursor blinks.