The Neural Engine changes the math

Every M-series Mac ships with a chip built for exactly this workload, and most apps ignore it. Cue's on-device transcription now runs on the Apple Neural Engine.

• Around 8x faster than our previous offline engine. Long recordings finalize in a blink instead of a wait.
• Live words while you speak. Streaming preview stays on, so you see text as you talk even with no internet.
• Almost no battery cost. The Neural Engine sips power. Your CPU stays free and your fans stay quiet.

Until now, choosing on-device meant trading speed for privacy. That trade is gone. On your Mac, private is the fast option.

Built to not fall over

Fast is worthless if it is flaky, so the new engine got the boring work too:

• Warm-up on launch, so your first dictation of the day is as fast as your fiftieth.
• The engine self-heals if it ever gets into a bad state, instead of silently degrading.
• Cancelling a recording fully clears the buffer. Nothing bleeds into your next dictation.

Milestones, redesigned

Your profile now tells the story of everything you've dictated. Every word counts toward ten voice ranks — starting at Whisper and climbing through Thunder all the way to Vox Machina at one million words.

• A new profile card with your rank, total words, and a progress bar to the next level.
• Round, honest thresholds. Ranks land on clean numbers, so you always know how far the next one is.
• Your stats at a glance — words per minute, top app, daily streak, and typing time saved.

Local is free. Full stop.

Every on-device model in Cue is now free, for everyone, with no account. Download the best model your Mac can run and dictate as much as you want. No meter, no time limits, no trial that runs out.

The reasoning is simple: on-device transcription runs on your hardware, so it costs us nothing. Charging for it would be charging you rent on your own Mac.

In Private Mode, nothing ever leaves your machine. Not audio, not text, not analytics about what you said. That is not a policy promise, it is how the app is built.

Cloud transcription is still there when you want maximum accuracy, and it is priced the same honest way: cents per use from a wallet you top up. No subscription. Ever.

Paste without touching your clipboard

Until now, every dictation app on macOS pasted the same way: put text on your clipboard, fake a Cmd+V. Your clipboard manager filled up with fragments of everything you ever said.

Cue now types text directly into the target app. Your clipboard stays exactly as you left it.

Cleanup by Apple Intelligence

On Macs with Apple Intelligence, Cue can now use it for the automatic cleanup pass. On-device, instant, and free, because it runs on your hardware too.

A better first five minutes

Onboarding got a full redesign. It is shorter, it explains the private/cloud choice honestly, and if you quit halfway it resumes where you left off instead of starting over.

Hush

Coffee shops, open offices, a TV in the next room. Real life is loud, and dictation has always paid the price.

Hush is Cue's new suppression layer. It runs on your Mac, in real time, before audio reaches transcription.

• Background noise goes away. Keyboards, traffic, espresso machines.
• Other voices too. The conversation behind you stops leaking into your text.
• You control the strength. A live slider, from gentle to aggressive, so you can tune it to your room.

Cloud dictation that does not flinch

When you use cloud transcription, Cue no longer depends on a single provider being up.

• Automatic failover. If the primary provider will not connect, Cue switches to a backup at connection time. You dictate, it works, you never see the switch.
• Bring your own backup. You can add your own provider keys as backups, stored locally like all your keys.

Sound and feel

• A new sound theme for start, stop, and completion.
• Feedback volume now tracks perceived loudness, so quiet and loud sounds actually feel even.

An overlay that behaves like it belongs

The floating orb is how you meet Cue a hundred times a day, so we rebuilt how it feels.

• Right-click works everywhere. The orb now has a proper context menu even while you are in another app. Switch modes, open settings, or quit without touching the main window.
• Put it where you want. New positions on the left, right, and bottom center, and the orb is dock-aware so it never hides behind your Dock.
• A quick tray on hover. Rest your cursor on the idle orb and a small tray of actions slides out. Move away and it tucks itself back in.

Long recordings, no drama

Under the hood, streaming transcription got a stability pass:

• Long dictation sessions no longer degrade or freeze partway through.
• The live preview now shows your full transcript as you speak, not a truncated tail.

macOS niceties

• A proper app menu, so standard shortcuts behave.
• Fixed window chrome after fullscreen cycles.
• A show-in-Dock setting for people who want Cue to be invisible.

One place for intelligence

Cleanup, Edit, and Translation all run on AI. Until now, configuring that AI was scattered. This release brings it into one Intelligence tab with a simple choice:

• Cue Cloud. Zero setup. We run the model, you pay per use in cents.
• Bring your own key. Plug in your Anthropic, OpenAI, Google, or other provider key and Cue talks to them directly.

Your keys stay home

Bring-your-own keys are stored locally on your Mac, in the system keychain. They are never synced to our servers. Not encrypted-and-synced. Just never sent.

Cleanup that never blocks you

The automatic cleanup pass got stricter rules:

• A hard 10-second budget. If the model is slow, you get your raw transcript instead of a wait. Your words are never held hostage by a loading state.
• Cleanup respects your mode. In Private Mode, automatic cleanup stays on your Mac. Cloud cleanup only happens when you have explicitly turned it on, and it is labeled when it does.
• Per-feature models. Want a fast model for cleanup and a smart one for Edit? You can set that now.

Edit Mode

Dictation gets words onto the screen. Edit Mode fixes them once they are there.

Select any text, in any app. Press the Edit hotkey. A small panel appears with your selection. Say what you want, or pick a preset, and the rewrite streams back in place of what you selected.

• Works everywhere. Mail, Slack, your browser, your editor. If you can select it, Cue can rewrite it.
• Presets for the common stuff. Proofread, shorten, change tone, turn notes into prose.
• Or just say it. "Make this sound less angry." "Turn this into three bullet points." Your voice is the instruction.
• Streaming. You watch the rewrite happen instead of staring at a spinner.

The details

• The Edit panel is searchable. Start typing to filter presets, or keep typing to write a custom instruction.
• Keyboard-first: arrow keys, Enter, and Escape all do what you expect.
• Edits are saved in history alongside dictations, so you can always get the original back.

Translation Mode

Speak in one language, paste in another. Pick a target language, hit the hotkey, and talk. Cue transcribes what you said, translates it, and pastes the result right where your cursor is.

• Any direction. Dictate in Hindi and paste English, or the other way around.
• Same flow as dictation. One hotkey, no app switching, no copy-paste dance.
• Full history. Translations live in your history next to your dictations.

Corrections that know where you are

Cue's cleanup pass now sees context: the app you're in, the window title, and the text around your cursor.

• Mid-sentence dictation stays mid-sentence. No more surprise capital letters when you dictate into the middle of a line.
• Jargon survives. Terms from the page you're on are less likely to get "corrected" into something else.

Reliability

• Re-granting microphone or accessibility permission now restores everything instantly. No restart needed.
• Start and stop dictation straight from the menu bar tray.
• Fixed a class of stale-transcription bugs when cancelling with Escape.

Cue 1.0

This is the one we've been building toward. Cue 1.0 brings streaming offline models to your Mac. Fully on-device, real-time transcription that just works.

Streaming Offline Models

• Fully offline, fully streaming — transcription happens as you speak, with no cloud round-trip
• Optimized for real-time speech — low memory and fast inference
• Works without internet — airplane mode, coffee shop with bad wifi, wherever you are
• Privacy by default — your voice never leaves your machine

What this means

Previous offline models processed audio in chunks. You'd speak, wait, then see text. Streaming changes that. Words appear as fast as you can say them, even without a connection.

Optimized for Apple Silicon

Runs natively on the Neural Engine for M-series Macs. Expect:

• Near-zero latency transcription
• Minimal CPU and memory overhead
• No thermal throttling, even in long sessions

Why 1.0

We started Cue as the context layer between your brain and your computer. Streaming offline transcription was the foundation that had to be right before everything else. Notes, meetings, agents. All of that is next. But this is the base.

What's next

Notes Mode and Edit Mode are in progress. Stay tuned.

Focus sessions

Writing with your voice works best when you're in the zone. Cue 0.7 adds Focus Sessions — a dedicated mode that removes distractions and helps you stay in flow.

How it works

• Press ⌥ + F to start a focus session
• Cue enters a minimal overlay mode — just your words on screen
• Set a timer (15, 30, 60 minutes) or go open-ended
• When you're done, Cue copies everything to your clipboard or pastes it where your cursor was

Focus Sessions are perfect for drafting emails, writing documents, or brain-dumping ideas.

Dictation analytics

Curious how much time your voice saves? The new Analytics tab in Settings shows:

• Words dictated this week, month, and all-time
• Time saved — estimated based on your average typing speed vs. dictation speed
• Accuracy rate — how often Cue gets it right on the first pass
• Most active apps — where you dictate the most
• Streak tracker — consecutive days using Cue

All data stays local. Nothing is sent to our servers.

Other changes

• Added support for macOS 15.3 Sequoia features
• New onboarding flow for first-time users
• Fixed an edge case where Rewrite mode would lose formatting in rich text editors
• Improved wake-word detection reliability when using AirPods

Under the hood

Cue 0.6 is the biggest performance update since launch. We rewrote the audio pipeline and processing engine to be dramatically faster and lighter.

What changed

• CPU usage down 60% — Cue now idles at near-zero CPU. Fans stay quiet.
• Instant activation — press ⌥ + Space and Cue is listening in under 50ms
• Streaming is smoother — text appears character-by-character with no jank
• Battery impact reduced — on M-series Macs, Cue barely registers in Activity Monitor

New audio engine

We moved from a buffered processing model to a streaming architecture. Instead of waiting for chunks of audio, Cue now processes speech as it arrives — reducing latency from ~300ms to under 100ms.

Apple Silicon optimized

The offline model now uses the Neural Engine on M1+ chips for inference. This means:

• 2x faster offline transcription
• 40% less memory usage
• No thermal throttling during long sessions

Fixes

• Resolved memory leak during extended dictation sessions (2+ hours)
• Fixed text insertion delay in VS Code with many extensions
• Improved handling of Bluetooth microphone reconnections
• Cue now gracefully handles sleep/wake without losing state

Auto-formatting

Dictation shouldn't mean a wall of text. Cue 0.5 listens to how you speak and structures your text automatically.

What it detects

• Lists: Say "first... second... third..." and Cue creates a numbered list
• Paragraphs: Natural pauses longer than 2 seconds start a new paragraph
• Headings: Say "heading" or "title" followed by your text to create a heading
• Line breaks: Say "new line" for an explicit break

Smart context

Cue adapts formatting to the app you're in:

• In Notion and Google Docs, it uses native formatting (Markdown rendered)
• In Slack and Discord, it uses their markdown syntax
• In plain text editors, it keeps things clean with simple line breaks

Bug fixes

• Fixed cursor jumping in Safari text fields
• Resolved an issue with duplicate words when network latency spikes
• Improved handling of dictation in password fields (Cue now skips them)

Custom vocabulary

Every field has its own language. Cue now lets you add words, names, and phrases it should always get right.

How to use it

• Open Settings → Vocabulary
• Add words one by one, or paste a list
• Each entry can include a spoken form and a written form (e.g., "reactjs" → "ReactJS")

Where it helps

• Developer tools: framework names, CLI commands, variable naming conventions
• Medical & legal: terminology that generic models stumble on
• Names: colleagues, clients, product names — spelled exactly how you want

Custom vocabulary works in both online and offline mode.

Other improvements

• Reduced memory usage by 30% during long dictation sessions
• Fixed a bug where Cue wouldn't activate in full-screen apps on macOS Sequoia
• Improved punctuation accuracy for rapid speech
• Menu bar tooltip now shows your current language and mode

Offline mode

The most requested feature since launch. Cue 0.3 includes a local Whisper model that runs entirely on your Mac.

How it works

• Download a 150MB model once (Settings → Offline Model)
• Cue automatically switches to offline mode when there's no internet
• Or force offline mode from the menu bar for maximum privacy

Performance

On M1 and later, offline transcription runs at near-real-time speed. On Intel Macs, there's a slight delay but accuracy stays high.

What stays the same

• Same keyboard shortcut (⌥ + Space)
• Same auto-corrections
• Same app compatibility
• Audio is still never stored — online or offline

Other fixes

• Fixed rare crash when switching audio inputs mid-dictation
• Improved Slack integration — handles thread replies correctly
• Added a "What's New" badge in the menu bar for updates

What's new

Context-aware corrections

Cue now uses surrounding text to fix transcription errors smarter. If you're writing code, it won't capitalize variable names. If you're in an email, it'll add proper punctuation.

8 new languages

We've added support for:

• Spanish, French, German, Portuguese
• Japanese, Korean, Mandarin, Hindi

Switch languages from the menu bar or say "switch to Spanish" mid-dictation.

Other improvements

• 50% faster cold-start time on Apple Silicon
• Fixed cursor placement bug in VS Code terminals
• Improved handling of technical jargon in coding contexts
• Menu bar icon now shows active language

What is Cue?

Cue is a macOS app that turns your voice into text — anywhere you can type. Press ⌥ + Space, start talking, and watch your words appear in real time.

Highlights

• Works in every app — Slack, Notion, VS Code, Mail, Chrome, and more.
• Real-time preview — See your words as you speak, not after.
• Powered by Deepgram — Fast, accurate transcription that handles accents and technical terms.
• Privacy first — Audio is never stored. Ever.

Getting started

1. Download Cue from cue.app.
2. Grant microphone and accessibility permissions.
3. Press ⌥ + Space and start talking.

That's it. No account required for the free tier.

What's next

We're working on offline mode, more languages, and Windows support. Stay tuned.