"My prompts are lazy because typing is slow."
Speaking 3x more context costs you the same 30 seconds. Better context in, better answers out.
Voice prompting on Mac
The best prompts are long and specific, and nobody wants to type them. Talk it through like you would to a colleague; Lyre pastes a clean, complete prompt into ChatGPT, Claude, Cursor or your terminal.
7-day money-back guarantee · macOS 13+ on Apple silicon
Sound familiar?
"My prompts are lazy because typing is slow."
Speaking 3x more context costs you the same 30 seconds. Better context in, better answers out.
"Voice mode answers out loud, I want text."
Lyre only writes. You keep the chat, the code assistant and the history you already use; you just stop typing the prompt.
"I don't want my prompts going through another cloud."
Transcription and cleanup run on your Mac. The only place your prompt goes is the AI you paste it into.
Real examples
Claude Code
Okay so refactor the the billing service so webhooks are um idempotent, use the webhook id as the key, and like add tests for retries.
↓ Lyre
Refactor the billing service so webhooks are idempotent. Use the webhook id as the key, and add tests for retries.
ChatGPT
Can you, can you give me a uh two-week plan to learn SQL, I have I have about an hour a day and I already know Excel.
↓ Lyre
Give me a two-week plan to learn SQL. I have about an hour a day and I already know Excel.
Cursor
So this component re-renders on every keystroke, I think it's it's the context provider, find out why and fix it.
↓ Lyre
This component re-renders on every keystroke. I think it's the context provider. Find out why and fix it.
How it works
No app to open and no window to manage. Lyre waits in the notch until you need it.
Move the pointer to the top of the screen. The Lyre panel drops down from the notch (on any display, from the top center). Click the mic.
Talk naturally: pauses, "um", restarts, all fine. Click the check when you're done.
Lyre transcribes and polishes on your Mac and copies the result. Press Cmd+V in any app: email, Slack, Notes, your editor, ChatGPT or Claude.
Features
Drops "um", "uh", "like", repeated words and false starts, and keeps your meaning and your tone. It doesn't rewrite you into corporate speak.
By default the panel only says "Copied to clipboard". Want to read or edit first? Turn "Just copy" off and the text shows up before it's copied.
No window to find. Hover the top of the screen and the panel is there; move away and it's gone. On displays without a notch it's the top center.
Whisper Large v3 Turbo and Qwen3 run on the GPU with Metal. A short dictation is ready to paste in a few seconds on an M-series Mac.
Transcription and polish both run locally. No account in the app, no cloud processing, no recordings kept.
Speak Portuguese, Spanish, French, German, Hindi or any of the ~99 languages Whisper understands, and paste English. Or keep the same language.
On a plane, on bad Wi-Fi, behind a strict firewall. After the one-time model download, Lyre needs no internet to dictate.
Pick speed or accuracy: three Whisper sizes and four language models, from 1.7B to 8B. Or turn polish off for verbatim transcripts.
Private by design
Every step runs on your Mac's own chip. There's no Lyre server that hears you, because there's no server doing the work.
Your voice is recorded into memory on your Mac.
whisper.cpp transcribes it with Apple silicon's GPU.
A local language model removes fillers and, if you want, translates.
The clean text is copied. Audio is never saved to disk.
Pricing
Everything included on both plans. Pay yearly and save 35%.
Flexible. Cancel any time.
$5/month
Coming soonThat's $3.25 a month.
$39/year
Coming soon7-day money-back guarantee. Not for you? Refund yourself from your account, no questions asked. Prices in USD; sales tax or VAT added at checkout where it applies.
Needs a Mac with Apple silicon (M1 or later) on macOS 13+. Payments by Dodo Payments.
FAQ
Yes. Anything that accepts paste works: terminals, Cursor, VS Code, Zed, ChatGPT, Claude, Gemini and Perplexity.
Whisper Large v3 Turbo handles most technical vocabulary (API, JSON, Kubernetes, React). Unusual names can come out wrong; the "Just copy" off setting lets you glance at the text before pasting.
No, on purpose. If you say "write a poem", Lyre writes down "write a poem" for you to paste. It cleans the words, it doesn't act on them.
A Mac with Apple silicon (M1 or later) on macOS 13 Ventura or newer. Plan for about 4 to 7 GB of disk for the models (Whisper Large v3 Turbo is 1.6 GB; the recommended Qwen3 8B is 5 GB, and there are smaller ones). 16 GB of memory is comfortable for the recommended models; 8 GB Macs should pick the smaller ones.
No. Audio is recorded into memory, transcribed by whisper.cpp and polished by a local model, all on your Mac. Audio is never written to disk. The only network calls are the one-time model downloads from Hugging Face, the licence check and update checks.
Apple Dictation types exactly what you said, fillers included. Lyre cleans the text up first (no "um", no repeated words, no false starts), can translate into another language as you speak, and works the same in every app because it goes through the clipboard.
Not yet. Today Lyre copies the clean text and you paste it with Cmd+V. Typing into the focused app and a global keyboard shortcut are next on the roadmap.
Whisper understands around 99 languages, including English, Spanish, Portuguese, French, German, Italian, Dutch, Hindi, Japanese, Korean and Chinese. You choose an input language and an output language; set them to different languages to translate as you dictate.
No trial, but there's no risk either: if Lyre isn't for you, get a full refund from your account within 7 days of paying, no questions asked.
$5/month or $39/year · 7-day refund · Apple silicon Macs