Cue.
Push AI, not pull.
An ambient, voice-first Mac assistant. Hit a hotkey, speak, let go. Cue sees what you are doing, thinks about it, and actually takes action across your apps. What Siri should have been, finally built.
Press ⌥Space to talk to Cue
A colleague who was already watching, not a search box you had to remember to open.
Every other assistant makes you stop, switch context, phrase your thoughts for an empty chat window, then copy the answer back. Cue lives on a hotkey. It already knows the app you are in, the text you highlighted, the thing you just copied. You talk, it acts. The work does not stop.
What ships in the box
Real things, not buzzwords. Every capability here maps to code in the repo, with the gaps called out honestly further down.
Voice-first, hotkey-driven
Option+Space, speak, let go. No app to switch to, no window to find. The HUD floats over whatever you are already doing, Raycast-style, and disappears when you are done.
Sees what you are doing
Cue reads the active app, the frontmost window, your selection, your clipboard. It can OCR the screen on demand. Context goes into the prompt so the answer feels like a colleague who was watching, not a search box.
Actually takes action
Claude decides, the Accessibility API executes. Open apps, navigate, paste, press buttons, run Shortcuts, drive AppleScript. The "does stuff" layer is the whole point. Assistants that only answer are chat boxes.
Native Mac, not a web view
Swift and SwiftUI, top to bottom. No Electron. No React Native. The HUD is a non-activating NSPanel so it does not steal focus from whatever you are typing into. Menubar presence, global hotkey via Carbon.
Local speech recognition
Whisper.cpp runs on-device with Metal acceleration. Your voice does not leave your Mac on the way to being transcribed. Only the final text goes to Claude, and only when you invite it.
Direct distribution, no sandbox
Signed and notarised DMG, Sparkle for updates. Cue needs the Accessibility API, global hotkeys, and microphone access. The Mac App Store sandbox would gut the product, so it ships direct. You keep the receipt.
How Cue thinks
Layered so cheap things happen always, expensive things happen only when they earn it. Your battery, your API bill, and your attention all stay respected.
Passive sensors
Zero costActive app, selected text, clipboard. Cheap, local, always on. No screen capture, no mic, no Claude calls.
Screen sampler
Local computeScreenCaptureKit plus Vision OCR, sampled on demand or at a user-set cadence. Respects Screen Recording permission, off until you opt in.
Trigger
Your callOption+Space for v1. Wake word ("Hey Cue") for v2. Later, a tiny on-device classifier scores ambient moments as worth surfacing, quietly.
Transcription
On-deviceWhisper.cpp on-device turns your voice into text. Voice activity detection short-circuits empty recordings so Claude is not woken up for nothing.
Reasoning
API callContext blob plus transcript plus tool definitions go to Claude. Streaming response starts rendering in the HUD before the model finishes thinking.
Action
Real effectClaude picks a tool, Cue executes through the Accessibility API. Destructive actions ask first, everything is logged, undo is one keystroke away.
Shipping in public, no smoke
The scaffold is real and runs end-to-end. Some of the hot-path pieces are stubbed and being replaced, week by week. Founder's License buyers see every step.
Scaffold
XcodeGen project, menubar, HUD panel, Option+Space hotkey, audio capture, Claude streaming, Accessibility trust flow.
Real voice loop
Whisper.cpp integration, voice activity detection, Keychain-stored API key, first-run onboarding.
Context awareness
ScreenCaptureKit plus Vision OCR, clipboard and selection observers, context aggregator.
Action execution
Accessibility API expanded, Claude tool-use, AppleScript bridges, confirm-destructive safety model.
Motion and polish
Spring-based HUD transitions, listening pulse, sound design, custom Q icon.
Wake word
"Hey Cue" via Picovoice Porcupine, battery-aware modes.
Proactive triggers
Heuristic then learned classifier for moments worth surfacing, with a signal-to-noise dial.
Why Cue is not on the App Store
To be useful, Cue needs three things the Mac App Store sandbox does not allow: the Accessibility API so it can drive other apps, a global hotkey so you can reach it from anywhere, and microphone access without jumping through user-hostile prompts every session.
So Cue ships direct, as a signed and notarised DMG, with Sparkle handling updates. You buy it from Polyxmedia. You own the receipt. Apple does not take 30 percent of a product they would not let exist on their store anyway.
The flip side is that Cue respects the Accessibility boundary religiously. Every action is logged, destructive ones confirm, and you can turn the whole thing off in System Settings whenever you want.
Scope honesty
Other tools own other problems. Cue is sharper because it refuses to be everything.
✕ Not a ChatGPT clone
Chat is a fallback, the voice loop is the primary interface.
✕ Not a code editor assistant
Cursor owns that space and does it well. Cue lives in the OS, not the IDE.
✕ Not a meeting transcriber
Granola owns that. Cue is for when you are working, not when you are in a call.
✕ Not always-recording
Rewind owns that. Cue is invited in, not ambient in the surveillance sense.
✕ Not cross-platform
Mac first, Mac only, at least until the shape of the product is right. Windows would dilute the positioning.
✕ Not heading to the App Store
Sandbox incompatibility is fundamental. Accessibility access cannot live inside App Store rules.
Boring where it counts
Platform-native, mostly Apple frameworks, one LLM vendor, one wake-word vendor. Small surface area, long shelf life.
Founder's License
One-time. Lifetime updates on the v1 line. You get every build from scaffold to ship, your name in the credits, and a direct line to the person building it. Capped to the first 200 buyers.
Waitlist
Get notified at launch
- Email when the public beta opens
- Dev log updates
- No purchase needed
Founder's License
Lifetime v1, shipping in public
One-time, no subscription
- Every beta build from today
- Lifetime updates on the v1 line
- Name in the in-app credits
- Direct line to the builder
- Priority roadmap votes
- Bring your own Anthropic API key
Team
Deploy across a whole team
- Volume licensing
- Shared action registry
- Per-seat admin controls
- Invoiced billing
cue-continuum.app, macOS 14 Sonoma or later, Apple Silicon recommended
The first Mac assistant that actually does stuff.
Cue is being built in public by Polyxmedia. Founder's seats are limited to 200, priced to reward the people who back it before the product is finished. After that, pricing returns to £199 at launch.