Speak a two-minute prompt into Claude Code's terminal, Cursor's chat, or a ChatGPT tab — and get clean, punctuated text at your cursor with the filler words already stripped. No copy-paste, no cutoffs. Typally exists because we hit this exact problem ourselves.
⬇ Download for Windows — 7-day free trialTypally started as an internal tool. We were living in Claude Code all day, and the best prompts — the ones with real context about the bug, the constraints, what we'd already tried — were long. Speaking them was obviously faster than typing them. But in our experience, built-in voice input kept cutting off after about a minute, mid-thought, and browser dictation meant transcribing in one tab and pasting into another. So we built a hotkey that records as long as you hold it, transcribes with Whisper large-v3, cleans up the "ums", and types the result into whatever window has focus. That tool became Typally.
AI coding tools reward context. Most people type 40 words per minute and unconsciously compress their prompts to save effort — dropping the reproduction steps, the constraint, the "and don't touch the auth module." Speaking at 150 words per minute removes that tax: you give Claude the full picture because saying it costs nothing. The quality difference compounds over a day of prompting.
Use it if it covers you — it's built in and free. The reasons we ended up building Typally: recording length (hold-to-talk with no practical limit), cleanup (filler words and punctuation fixed before the text lands), custom vocabulary for jargon-heavy prompts, and one tool that works identically in every app rather than only inside Claude Code. If your prompts are short, built-in is fine; if you think in paragraphs, you'll feel the difference in a day.
7-day free trial, $7.99/month after. Add your stack to the dictionary, hold the key, and give Claude the context it deserves.
Yes. Typally types into whatever window has keyboard focus — Windows Terminal, PowerShell, WSL, VS Code's integrated terminal, or the Claude Code desktop app all work the same way.
No practical one — it records for as long as you hold the hotkey (or until you toggle it off in toggle mode). Long, detailed prompts are exactly the use case it was built for.
Out of the box, Whisper large-v3 handles technical English well. For your specific stack — package names, internal repo names, acronyms — add them to the custom vocabulary and they'll transcribe correctly.
Yes — anywhere you can place a cursor on Windows. It's not integrated with any one tool; it types at the OS level, which is why it works with all of them.
No. Audio is transcribed over an encrypted connection and immediately discarded. Your history is stored only on your device.