Features
Dictation that becomes an agent.
Dictation is free and unlimited, transcribed on your computer. Pro turns the same key into an agent that writes, answers and acts, takes meeting notes and offers the next step. The agent features are on macOS for now. Dictation runs on macOS, Windows and Linux.
Pro
The voice agent
One shortcut. Ambient decides from what you say whether to write, answer, act or recall. The Ask shortcut (⌃ ⌥ Space) always goes to the agent. Free includes 25 AI uses a week.
Demo of the voice agent. Write: you say “hey can we push the sync to thursday”. Talking to a person? It polishes your words and pastes them where your cursor is, like dictation. Answer: you say “what’s the difference between UTC and GMT”. A question gets a short answer in a glass card at the top right, with sources. It searches the web only when it needs to. Act: you say “add lunch with Ana on Friday at 1”. Reminders and events go into your Mac’s own Calendar and Reminders. It opens apps and links, and runs your Shortcuts. Recall: you say “what did I say about pricing yesterday”. It finds your own things: what you said, what was on screen, your files and calendar. Nothing to connect.
One key, four outcomes
Write
Talking to someone? Your words are polished and pasted where your cursor is, exactly like dictation.
Answer
Ask a question and a glass answer card appears near the top right, with sources. It searches the web only when the answer needs it.
Answer about this
“Summarise this” or “what does this error mean” uses the text of the window in front of you.
Act
“Remind me at 5 to call Sam” or “add lunch with Ana Friday at 1” lands in your Mac’s own Reminders and Calendar. It also opens apps and links, and runs your Shortcuts by name.
Recall
“What did I say about pricing yesterday?” is answered from your dictations, meetings and, if you turn it on, screen memory.
Writing stays the default
When it isn’t sure, it writes. Dictating an email stays dictating.
Pro
Meetings
Notes for any call, with no bot joining and nothing to connect. Free includes 2 meetings a month.
A live nudge during a call
Any call
Zoom, Google Meet, Microsoft Teams, FaceTime and Slack huddles. If it plays on your Mac, it works.
You and them
Your microphone and your Mac’s audio are transcribed separately on your Mac, so it knows who said what.
Notes as you talk
Rolling notes build during the call, so you can look back mid-meeting.
Live nudges
They asked you for something: Find it. You promised something: Remind me. Plus follow-up questions worth asking.
After the call
A summary, decisions, action items, questions for you, and a follow-up email draft ready to send.
Audio never saved
The audio is transcribed on your Mac and thrown away. Only the text is kept, on your Mac.
Pro
Suggestions
Switch it on and it offers one next step at the right moment. It is decided on your Mac, and nothing is sent unless you accept. During a meeting, nudges use only the recent transcript text.
A suggestion from an email thread
Long threads: summarise them. Replies: dictate the gist, get the email. A date in an email or Slack message: add it to your calendar.
Writing
A long draft: tighten it.
Browsing
A long page: summarise it. A product page: check prices elsewhere. Rephrasing the same search: just answer it.
Work
A spreadsheet formula error, a stack trace or a failing test: explain it and suggest a fix.
Meeting prep
A few minutes before an event in your calendar, it offers a short brief to get you ready.
Quiet by design
One line, Enter or Esc, gone in seconds. Dismiss a kind of suggestion and it backs off.
Pro
Finds your own things
Nothing to connect. It looks where your things already are.
Open windows
The text of the windows open in every app, read when you ask.
The last two minutes
A short rolling buffer, in memory only, for “what was that number I just closed?”.
Screen memory
What was on screen, saved as text, encrypted on your Mac. Kept 7 days by default.
Spotlight
Your files, Mail messages and notes, through the Mac’s own index.
Calendar
Your events, read through macOS.
Dictation history
Everything you’ve said to it.
Drive and Gmail
Searched in the browser you’re already signed in to. No login handed over.
Privacy
It reads your screen only when you let it.
Everything that looks at your screen is off until you turn it on, one capability at a time. When it is on, here is how it behaves.
Free
Dictation, free forever
Unlimited, on every platform, transcribed on your computer.
On-device transcription
Speech-to-text runs locally with open models (Whisper, Parakeet, Moonshine, SenseVoice and more) with GPU acceleration where available. Your audio never leaves your computer.
Works offline
Once a model is downloaded, dictation needs no connection at all.
99 languages
Pick the languages you speak. Whisper models cover 99 of them, and can translate what you say into English.
Choose your model
Trade speed for accuracy: small, fast models for quick notes, larger ones for long-form writing. Switch any time.
Any app
Text is pasted wherever your cursor is: email, chat, docs, code editors, the terminal.
Push-to-talk or toggle
Hold the shortcut while you speak, or tap once to start and again to stop. The shortcut is yours to choose.
Custom words
Add names, jargon and product terms so they come out spelled your way.
History and files
Every transcription is kept on your device: search it, play it back, re-transcribe it. Drop in an audio file to get the text.
AI polish (Pro)
Punctuation, no filler words, lists formatted, “scratch that” applied. Hosted with nothing to set up, or on your Mac with Apple Intelligence. Only text is sent, never audio.
macOS, Windows, Linux
Native builds for all three, plus command-line flags for scripting and Wayland shortcuts.
Coming soon · computer use
Next, it uses your apps for you.
Say “make a Linear ticket for the login bug” and Ambient opens the app and does the steps on screen, the way you would. No app to connect, and it asks before anything that sends, deletes or buys.