Dicticus

A fully local dictation app for macOS and iOS. Hold a key or trigger a shortcut, speak, release — accurate text appears instantly. Fully private, no cloud dependency.

Features

• Multi-Platform — Native apps for macOS (Menu Bar) and iOS (Universal App).
• System-wide push-to-talk — Works in any text field on macOS.
• Siri Shortcuts & Action Button — Trigger iOS dictation from anywhere, even when locked.
• On-device ASR — Whisper large-v3-turbo via WhisperKit on macOS; Parakeet TDT v3 via FluidAudio on iOS — both run on the Apple Neural Engine.
• AI cleanup mode — Grammar, punctuation, word-order, and filler-word correction via Qwen3.5-4B (llama.cpp), on both macOS and iOS. Optional and opt-in — triggered by its own shortcut, with a deterministic fidelity guard that blocks the model from rewriting what you actually dictated.
• Spoken punctuation — Say "comma", "new line", etc. and it's inserted deterministically before cleanup.
• Auto language detection — German and English, no manual switching.
• Live transcript panel (macOS) — while you hold the dictation key, a small panel near your text cursor shows the words as you speak; the pasted text still comes from the full final decode. Turn it off in Settings → General.
• Media auto-pause (macOS) — playing media (Music, Spotify, browser video) is paused, or muted when it can't be paused, while you dictate, and resumes on release.
• Local History & Search — Browse and search your past transcriptions with FTS5.
• Custom Dictionary — Define your own replacements for technical terms or names; import/export as CSV/JSON and add bundled starter packs.

Requirements

macOS

• macOS 26+ (Tahoe)
• Apple Silicon Mac (Intel Macs launch the universal binary, but transcription without the Neural Engine is too slow to recommend)
• ~1 GB disk for the ASR model, plus ~2.7 GB for the optional AI-cleanup model (downloaded only if you enable AI cleanup)

iOS / iPadOS

• iOS 17+
• iPhone 15 or later (for Action Button optimization), or any modern iPhone/iPad. AI cleanup requires a device with ≥5 GB RAM.
• ~1.2 GB disk for the ASR model (Parakeet TDT v3), plus ~2.7 GB for the optional AI-cleanup model.

Installation

macOS

1. Download Dicticus.dmg from Releases
2. Open the DMG and drag Dicticus to Applications
3. Launch Dicticus — it's Developer ID signed and notarized, so it opens normally (no Gatekeeper override needed).
4. Grant Microphone and Accessibility permissions when prompted.

iOS

1. Build from source using Xcode 26.
2. Grant Microphone permission during onboarding.
3. Download the ASR model (one-time setup).
4. Follow the in-app guides to setup the Start Dictation Siri Shortcut.

Manual uninstall

Prefer the script: ./scripts/uninstall.sh quits Dicticus, removes the app bundle, preference files, and the LaunchAtLogin entry, then prompts before deleting the ~3 GB model cache.

To uninstall by hand, remove these four locations:

1. /Applications/Dicticus.app
2. ~/Library/Preferences/com.dicticus.*.plist
3. ~/Library/Application Support/Dicticus/ (contains the Qwen3.5 AI-cleanup model, the WhisperKit ASR cache, and history.sqlite)
4. ~/Library/LaunchAgents/com.dicticus.app.plist (LaunchAtLogin entry; only present if you enabled "Start at login")

TCC permission entries: macOS does not let any app clean its own Privacy & Security entries. After uninstalling, open System Settings > Privacy & Security and remove Dicticus from Microphone, Accessibility, and Input Monitoring before reinstalling — otherwise stale TCC entries from previous bundle copies can block hotkeys.

Updating the app icon

The app icon is generated from a single source: assets/icon-master.png (1024×1024 PNG, committed to the repo).

To change the icon:

1. Replace assets/icon-master.png with a new 1024×1024 PNG.
2. Run ./scripts/generate-icons.sh to fan the master out to:
   • all ten macOS appiconset PNGs (16/32/128/256/512 × 1x and 2x)
   • the single iOS AppIcon.png (iOS 18 unified icon model)
3. Commit the regenerated PNGs along with your master change — generated outputs are tracked in git so PR reviewers can spot unintended icon changes visually.
4. Clean-build (scripts/build-dmg.sh or xcodebuild -scheme Dicticus -configuration Release clean build) before verifying in Finder; LaunchServices caches stale icons aggressively. If the new icon does not appear after a clean build, run killall Dock Finder and sudo mdutil -E / to nuke icon caches.

Usage

• macOS: Hold a hotkey, speak, release. Two configurable hotkeys — plain dictation (default ⌃⇧S) and AI cleanup (default ⌃⇧D).
• iOS: Trigger the Start Dictation shortcut via voice, Action Button, or Back Tap.
• Dictionary: Manage replacements in Settings to improve accuracy for specific terms.
• History: Use the History tab on iOS or the Menu Bar on macOS to find past transcriptions.

Privacy

All processing happens on-device. No audio or text is ever sent to any server. The app makes zero network calls during operation (model downloads happen once on first launch).

Tech Stack

Component	Technology
App shell	Swift 6 + SwiftUI, MenuBarExtra (macOS)
ASR	macOS: WhisperKit + Whisper large-v3-turbo · iOS: FluidAudio + Parakeet TDT v3 (both CoreML, Apple Neural Engine)
LLM	llama.cpp + Qwen3.5-4B Instruct Q4_K_M (Metal) — macOS + iOS
Database	GRDB + SQLite (FTS5)
Live Activity	ActivityKit (iOS)
Text injection	NSPasteboard (macOS) + UIPasteboard (iOS)

Building from Source

# Prerequisites: Xcode 26+, xcodegen
brew install xcodegen

# macOS
cd macOS
xcodegen generate
xcodebuild -scheme Dicticus -configuration Release build

# iOS
cd iOS
xcodegen generate
# Open Dicticus.xcodeproj in Xcode 26 and Run on Device/Simulator

License

Private project. All rights reserved.