
HeyPace
A voice assistant that sees your Mac.
About HeyPace
HeyPace is a private voice agent for Apple Silicon Macs that can see what is on your screen. You hold ctrl+option, ask a question about the window in front of you, and Pace (the app's short name) answers out loud or carries out an action you approve, such as clicking a button, snapping a window, or adding something to your calendar. Voice, screen understanding and planning run locally by default, and the app needs no account to use. It is built by an independent developer and the source code is published on GitHub under the MIT License.
The problem it targets is the trade-off most desktop assistants force on you: to get help with what is on your screen you usually have to send screenshots, audio and context to a cloud service. Pace takes the opposite position. In local mode it states that it ships nothing off the machine, and the in-app Privacy dashboard measures bytes from a local audit log, showing 0 bytes for local inference and how much went where if you turn on a cloud planner. Conversation memory, journals, flows and taught skills are ordinary local files under ~/Library/Application Support/Pace/.
The pricing shape is unusual for this category. The full app is free to try against Apple Foundation Models on macOS 26, and a one-time $29 purchase adds a bundled local model stack (a Qwen3-4B planner, Qwen3-VL-4B vision model, Qwen3 TTS and a Kokoro-82M sidecar) so nothing extra has to be installed. Checkout is currently handled by a manual email path rather than a hosted store, and the public build is a preview that is ad-hoc signed and not notarized.
You download the preview zip, move Pace to Applications and approve it once in System Settings, because the current build is not notarized. On first launch macOS asks for Microphone, Screen Recording and Accessibility permissions, which cover voice input, screen context and approved actions. After that Pace lives in the menu bar with no dock icon and updates itself through Sparkle.
To use it, you hold ctrl+option, speak, and let go. Pace merges a local vision model, Apple Vision OCR and the macOS Accessibility tree into one element map of the current screen and hands that to the planner. The planner decides whether to answer or to act. Answers are spoken back with on-device text-to-speech, using an Apple Premium voice, the bundled Qwen3 TTS or the Kokoro sidecar. Actions such as clicking, typing or opening an app go through Accessibility, and Pace asks for approval before it executes them. Reversible changes show a 5-second undo banner, and if a click misses, Pace says so.
By default the planner runs on your Mac, using Apple Foundation Models in the free tier or the bundled MLX models after purchase. In Settings, under Planner, you can opt in to a consented Codex or Claude CLI, or a bring-your-own-key API. When a turn goes off-device, the capsule turns amber and the turn is written to the local audit log. Always-listening mode is an explicit opt-in rather than the default.
- •
Screen-aware voice questions - Ask about whatever is on screen and Pace builds an element map from a local vision model, Apple Vision OCR and the Accessibility tree before planning a reply. Push-to-talk on ctrl+option is the default, so the microphone is not open all the time unless you switch always-listening on in Settings.
- •
Approved actions on your Mac - Commands like "click submit", "snap this window left" or "add this to my calendar" are carried out through macOS Accessibility. The action executor covers clicks, typing, app and URL opens, calendar and reminders, and it asks for approval first. Reversible changes leave a 5-second undo banner.
- •
Learning mode that teaches the next click - Pace can draw on your screen, point at the next thing to click, and advance only when you actually click the right element, which turns it into an on-screen guide for unfamiliar apps rather than only an assistant that acts for you. The site gives Figma, Cursor, Logic Pro and internal admin tools as examples, and the verification loop watches your click events so it only moves forward when you did the right thing.
- •
Flows, recipes and taught skills - You can record a flow to replay clicks and keystrokes, install a pre-built recipe by voice, or teach a skill in plain language. Taught skills re-read the screen and re-plan on each run instead of blindly repeating clicks, and they are stored as ordinary markdown files you own. Pace also registers pace:// deeplinks and App Intents for Siri, Shortcuts and Spotlight, so a Raycast script or a Shortcut can start listening or send a chat, and a deeplink goes through the same approval pipeline as your voice.
- •
On-device meeting notes - Pace records your microphone and the other side's audio as separate tracks, transcribes them on the Mac, and uses a privacy-pinned local model to write the summary, action items and decisions, even if you use a cloud planner elsewhere. Recordings and notes follow a retention window that defaults to 30 days.
- •
Local activity journals - Watch mode logs when your screen meaningfully changes and a separate app-usage journal records which apps had focus and for how long. You can ask "what did I do today?" or "what did I research about X?" and Pace answers from those journals, which stay on disk with 7-day retention and per-source enable and clear controls in Settings.
HeyPace fits Mac users on Apple Silicon who want a voice assistant that understands their screen but do not want their screen and audio streamed to a cloud service. Privacy-minded professionals, such as people handling client documents or internal tools, get help with on-screen tasks while keeping inference on the machine, and they can verify that through the audit log rather than taking it on trust.
It also suits people who sit in a lot of calls and want meeting notes without uploading the meeting. Because both sides are captured as separate tracks and summarised by a local model, the notes workflow stays on the Mac even for users who allow a cloud planner for other tasks.
Power users and developers will find the automation side useful: recorded flows, voice-installed recipes, plain-language skills stored as markdown, pace:// deeplinks and App Intents for Raycast, Shortcuts and Siri, and an MCP bridge for self-hosted tools. Since the code is MIT licensed, developers can also build Pace from source in Xcode.
The current preview has real requirements to keep in mind: macOS 26, Apple Silicon and at least 16 GB of RAM (32 GB for the bigger bf16 model), plus an approximately 8 GB model download for the paid local stack. Users who are not comfortable approving an unnotarized app in Privacy and Security may prefer to wait for a signed release.
Pricing
One time purchase, not a subscription
- Try$0/mo
- Pace$29 once
From the vendor pricing page, 2026-10-21

















