Software Releases

Production-ready desktop applications, local AI engines, and native utilities built with an uncompromising emphasis on on-device speed, zero cloud telemetry, and clean user experience.

Dictate

1.0.0-previewmacOS Native • Local Engine

Speak. It types. Nothing leaves your Mac.

100% Air-Gapped<150ms Metal<35MB Idle RAM

A local-first, context-aware dictation companion for macOS. Hold Fn/Globe, speak naturally, and get cleanly formatted text in any app without your voice leaving your Mac.

Direct macOS Distribution

Install Standalone macOS Disk Image

Native Metal hardware-accelerated binary. No account, no cloud servers, no runtime dependencies.

Download for Mac (.dmg)1 MB

System: macOS 13.0+ (Ventura, Sonoma, Sequoia, or later)

Architecture: Metal

Engine: whisper.cpp GGUF Quantized

Interactive Hardware Simulation

Push-to-Talk Global Overlay & Active Cursor Injection

TARGET DOCUMENT — Notes.app

Active Frontmost Focus

# Sprint Planning & Engineering Sync

...KUBERNETES MIGRATION TUESDAY

Inserted cleanly • Zero cloud telemetry • Air-gapped memoryApple Neural Engine & Metal Pipeline

Context-Aware Reasoning

Dynamic App Profiles

The exact same raw speech transforms instantly based on the active macOS frontmost application.

Raw Spoken Input:"let's schedule the kubernetes sync for next tuesday and email anjali at acme dot io"

GENERIC OS DICTATIONPhonetic Mangling

"let's schedule the cuber net ease sync for next 2nd day and email on Julie at ack me.io"

✕Loses technical entity casing, corrupts colleague names, and breaks syntax.

DICTATE (MAIL & FASTMAIL)Contextual Formatting

Let's schedule the Kubernetes sync for next Tuesday and email Anjali (anjali@acme.io).

Preserves natural sentence casing, parses spoken email protocol links, and formats polite business typography.

Domain Adaptation

Custom Vocabulary & Phonetic Biasing

Provide phonetic biasing tokens for uncommon engineering terms, repo names, acronyms, and teammate names.

Active On-Device Phonetic Biasing Tokens:

KubernetesPostgreSQLAnjalivLLMSupabaseAlphaEarth

Tokens are passed directly into the whisper.cpp decoder loop as initial prompt biasing tokens, eliminating acoustic hallucinations without model fine-tuning.

Architectural Breakdown

Local On-Device Engine vs Cloud Dictation

Why local inference on Apple Silicon outperforms hosted server-based transcription tools.

Feature / Dimension	Dictate (This App)	Cloud Dictation (Whisperflow)	Cloud Speech APIs
Audio Processing	100% On-Device (whisper.cpp + Metal)	Remote Cloud Servers	Remote Cloud Endpoints
Privacy & Telemetry	Zero data leaves device; fully air-gapped	Audio stored & processed in cloud	Subject to API provider logging
Pricing	Free standalone app (No subscription)	$15 – $30 / month recurring	Usage-based token billing
Offline Support	Full offline functionality (Airplane mode)	Fails without internet connection	Fails without internet
Model Selection	Whisper Tiny, Base, Small, Medium, Large-v3-Turbo	Locked to provider's hosted model	Fixed remote endpoints
System Integration	Native macOS menu bar + cursor injection	Proprietary desktop client	Developer integration required

Ready to dictate locally? Grab the standalone installer for macOS.

Download DictationApp.dmg (1 MB)