Use case
Talk to the agent instead of typing at it
Prompting a coding agent is writing, not commanding — long sentences, full of proper nouns from your own codebase. That is the worst thing to type and the best thing to say out loud.
The proper nouns are the whole problem
A prompt to a coding agent is dense with names no general recogniser has seen: your repository, your service, the library you adopted last month, the colleague whose branch you are on. Get one wrong and the agent goes to the wrong file.
The personal dictionary is where you put them, once. It applies after recognition, so a term you add takes effect on the next sentence.
A terminal is not an email
File your terminal under the terminal tone and spoken punctuation and path fragments survive intact instead of being tidied into prose. File your editor separately. Voicecape follows whichever window has focus.
Long prompts are where speaking wins
A two-line prompt is faster to type. A paragraph explaining what you want changed, why the last attempt was wrong and what to leave alone is not — and that is the prompt that actually gets a good result.
Questions about this
Does it work in any terminal?
Yes — Voicecape inserts at the cursor in whatever app has focus, including Terminal, iTerm and the terminal inside an editor.
Will it mangle file paths?
File the app under the terminal tone and path fragments and spoken punctuation are kept as said.
Can I dictate in Korean and English at once?
Yes. Mixed-script sentences are the case the cloud model is built for.