Free, on-device dictation for any Mac app.

Press Control-Space, talk, and clean text lands in Slack, Gmail, Cursor or Terminal. Transcription runs locally with Whisper, so your voice never leaves your Mac. Optional AI cleanup with a local model.

Download for Mac Why voice

Free during the beta · Apple Silicon (M1 or later)

Relay 0.2.2, released 2 June 2026. Release notes.

Get a license key and release notes by email (optional; the download does not require it)

Control-Space, speak, and the text lands in Slack. Screen recording, no audio.

Shown: the Relay bar as it appears at the bottom of your screen.

Your end result
Hey Sarah, quick update on the launch. Engineering is on track to land the API changes by Thursday, so we should be good to ship to beta on Friday morning. Let me know if you need anything from my side before then.
Cursor ˄
Ready Listening, click to stop Pasting, wait for the green check… Sent ✓
Raw
Clean
AI
Auto

Hit ⌃ Space, speak, and your words land in any app on your Mac.

Unlike Apple Dictation it formats and rewrites. Unlike Wispr Flow nothing is uploaded. Unlike Superwhisper there is nothing to configure.

Send to any app on your Mac
Pick the destination, dictate, done. The bar follows whatever you're focused on.
Cursor Code editor
Slack Chat
Gmail Email
Outlook Email
Terminal Command line
Claude AI assistant
Cursor Slack Gmail Outlook Terminal Claude ˄
Ready
Raw
Clean
AI
Auto

Plus Notion, Linear, GitHub, ChatGPT, iMessage, Discord, Figma, Google Docs, and every other text field on your Mac.

Speaking is roughly three times faster than typing.

In a 2016 Stanford and Baidu study, speech input ran at 153 words per minute against 52 typed, 2.93 times faster (Ruan et al., arXiv:1608.07323). Relay turns that gap into finished text in the app you are already using.

Speaking
153 wpm
Typing
52 wpm
Measured rates for English text entry in the study above. Your own numbers will differ; the calculator below uses yours.
Time savings
At your speed and volume, dictating instead of typing would save you:
30 min
Every working day
2.5 h per week 16 days per year
50 wpm
45 min per day
150 wpm

Defaults: 45 minutes a day of typing that could be dictated (messages, email, prompts, notes; not code or editing), 50 wpm typing, 150 wpm speaking, 5 working days a week, 250 a year, 8-hour days. Time spent correcting the transcript is not subtracted. Move the sliders to your own numbers.

Over a year, that is 16 days of working time.
About 125 hours, or 3.1 weeks, at the assumptions above.

Say the thought while it is still whole

Speech keeps pace with thinking in a way typing does not. Say the idea as it forms; Relay's Clean and AI modes handle the punctuation and structure.

Cleaner output than a fast draft

AI mode turns what you said into a coherent message, matched to the app it is going to: terse for Slack, fuller for email, literal for a terminal.

Easier on your hands

Long days at the keyboard add up. Dictation moves most of the bulk-text load off your hands; see voice typing for RSI for what it can and cannot do.

Download for Mac
Free during the beta · Apple Silicon (M1 or later)

Three levels of polish.

The same words, refined to the degree the moment calls for. Switch any time with ⌘1 / ⌘2 / ⌘3.

You said
“uh so basically we need to push the launch i think to thursday because the api stuff is still flaking. can you tell sarah?”
Raw
uh so basically we need to push the launch i think to thursday because the api stuff is still flaking. can you tell sarah?
⌘1 · Exactly what Whisper heard. Instant.
Clean
So basically we need to push the launch, I think, to Thursday because the API stuff is still flaking. Can you tell Sarah?
⌘2 · Same words, fixed punctuation and casing.
AI
Hey Sarah, we're pushing the launch to Thursday. The API integration is still unstable and we'd rather ship clean than on schedule. Will keep you posted.
⌘3 · Full rewrite, tone matched to the target app.
Reference

Every shortcut and voice command, in one place.

Open the reference

You choose where your text goes.

Transcription is always local. The AI polish step is opt-in and runs on whichever brain you point it at.

Always local Your voice is transcribed on-device with whisper.cpp and Apple's Metal GPU. No cloud service ever hears your audio.
Path A · Local model

Keep everything under your roof.

Point Relay at your own Ollama install, on your Mac or a Mac Mini on your network. Polished text never touches the internet. Slower than a hosted API, but nothing leaves the house.

Path B · Your API key

Plug in a frontier model when you want one.

Drop in an Anthropic, OpenAI, or other API key when you need top-tier quality. Requests are billed to your account and sent over your network, never proxied through us.

No telemetry in the app. Usage stats live in a local file on your Mac. Signing up for a license key stores your email address; see the privacy policy.

Try Relay on your Mac.

Free during the beta. Transcription stays on your machine.

Download for Mac

Apple Silicon (M1 or later) · Relay 0.2.2 · Release notes

Get a license key and release notes by email (optional)