Getting Started with Speakeasy
This guide walks you through installing Speakeasy on macOS, granting the required Apple security permissions, and completing your first voice dictation in under two minutes.
System Requirements
| Hardware / OS | Recommended Specification | Minimum Supported |
|---|---|---|
| Operating System | macOS 14.5+ (Sonoma) or macOS 15.0+ (Sequoia) | macOS 13.0 (Ventura) |
| Processor | Apple Silicon (M1, M2, M3, M4 — all variants) | Intel Core i7 / i9 (AVX2 CPU fallback) |
| Unified Memory (RAM) | 16 GB Unified Memory | 8 GB Unified Memory |
| Disk Storage | 3 GB free disk space (for Core ML weights) | 1.5 GB free disk space |
Step 1: Download & Install the macOS App
Mount the Disk Image (.dmg)
Download the latest universal DMG from the Download page. Double-click the DictationApp-1.0.0.dmg file, then drag the Speakeasy icon into your Applications folder.
Speakeasy is cryptographically signed and notarized by Apple. If prompted with "Speakeasy is an app downloaded from the Internet. Are you sure you want to open it?", click Open.
Step 2: Grant Required macOS Permissions
Because Speakeasy runs entirely on your device and injects keystrokes directly into active applications, macOS requires specific permission grants in System Settings → Privacy & Security.
Microphone Permission
Speakeasy requires access to your microphone to capture your voice. The audio is streamed directly to the on-device CoreAudio ring buffer and is processed locally by the Apple Neural Engine. It is never transmitted over the internet or saved to disk without your permission.
Path: System Settings > Privacy & Security > Microphone > Toggle On
Accessibility Permission
Speakeasy uses the macOS Accessibility API (AXUIElement) to inject converted text at your active cursor position in any app (Cursor, VS Code, Claude Code, Slack, Safari, Mail). Without Accessibility access, Speakeasy cannot paste text automatically.
Path: System Settings > Privacy & Security > Accessibility > Toggle On
Screen & System Audio Recording (For Bot-Free Meetings Only)
If you plan to use Speakeasy to transcribe Zoom, Google Meet, Microsoft Teams, or Slack calls without inviting a bot participant, grant System Audio Recording access via Apple's ScreenCaptureKit API. This captures inbound meeting audio directly from your audio bus.
Path: System Settings > Privacy & Security > Screen & System Audio Recording
Step 3: Core ML Model Initialization
On first launch, Speakeasy loads the optimized whisper-large-v3-turbo model converted to 16-bit floating-point Core ML format.
- Pre-compiled Engine: Speakeasy compiles the model directly to Apple Neural Engine (ANE) hardware instructions during the first 15 seconds.
- Memory Footprint: Occupies approximately 950 MB of unified RAM when active.
- Zero Background Idle: When dictation is inactive, the model drops into low-power sleep mode, consuming 0.0% CPU and zero battery overhead.
Step 4: Your First Dictation
- Click into any text field or editor (e.g. Cursor, Slack, Apple Notes, Terminal).
- Press and hold Option (⌥) + Space.
- Speak naturally: "Review the authentication middleware and refactor the JWT token validation logic."
- Release the keys. Your spoken words are transcribed, auto-punctuated, and typed directly into the active editor in under 450 milliseconds.
Prefer not holding keys? Open Preferences → Shortcuts and switch from Hold-to-Talk to Push-to-Toggle. Tap once to start dictating, tap again to finish.