Oyma requires macOS 13.0 (Ventura) or newer, with macOS 14.2+ (Sonoma) recommended for optimal ScreenCaptureKit audio. While compatible with Intel Macs, Apple Silicon (M1/M2/M3/M4) is strongly recommended for hardware-accelerated Whisper transcription.
Native macOS Engineering
Oyma is written natively for macOS, utilizing Apple’s low-level system frameworks including CoreText, AppKit, ScreenCaptureKit, CoreAudio, and CoreML. Because it is not packaged inside a heavy Electron wrapper, Oyma delivers exceptional battery efficiency, sub-millisecond typing response, and minimal RAM overhead.
Supported Operating Systems
- macOS 14 Sonoma (14.2 or later): Recommended. Includes Apple’s latest ScreenCaptureKit system audio tap enhancements and CoreML Neural Engine optimizations.
- macOS 15 Sequoia: Fully supported with full compatibility for Apple Silicon architectures.
- macOS 13 Ventura (13.0 or later): Minimum supported operating system. Basic ScreenCaptureKit features are supported.
- macOS 12 Monterey and older: Not supported due to the absence of Apple ScreenCaptureKit audio stream APIs.
Processor & Hardware Recommendations
| Hardware Architecture | Note Editing & Search | Meeting Audio Capture | On-Device Whisper Transcription | Local Ollama LLM Inference |
|---|---|---|---|---|
| Apple Silicon (M1–M4) | Instantaneous (<1ms) | Flawless | Hardware accelerated (ANE) | Supported (Unified Memory) |
| Intel Core (x86_64) | Instantaneous (<2ms) | Flawless | CPU-only (slower) | Cloud BYOK recommended |
Memory (RAM) Guidelines
- 8 GB Unified Memory: Excellent for note editing, search, audio recording, and Whisper Base/Small models.
- 16 GB Unified Memory: Recommended sweet spot. Allows concurrent recording, Whisper Medium/Large models, and running 7B-parameter local LLMs (e.g., Llama 3 8B) in the background.
- 24 GB to 128 GB Unified Memory: Optimal for running larger local models (e.g., Command-R, Mistral Nemo) alongside high-capacity video conferencing sessions.
Disk Space
The application bundle itself occupies under 120 MB. Downloaded CoreML Whisper speech models range from 75 MB (tiny) to 1.5 GB (large-v3), stored locally in your application support directory.
Apple Neural Engine Architecture Across M-Series
The Apple Neural Engine (ANE) included on M-series chips provides specialized hardware acceleration for automatic speech recognition:
- M1 Series (M1, M1 Pro, M1 Max, M1 Ultra): 16-core ANE capable of 11 trillion operations per second. Whisper Base transcribes audio at roughly 6x to 8x real-time speed.
- M2 Series (M2, M2 Pro, M2 Max): 16-core ANE capable of 15.8 trillion operations per second, delivering approximately 8x to 10x real-time Whisper speed.
- M3 and M4 Series: Advanced neural cores with up to 38 trillion operations per second, executing Whisper Large-v3 models with remarkable thermal efficiency.
Thermal Efficiency and Fan Acoustics
Unlike unoptimized background Python scripts that peg CPU cores and cause laptop fans to spin loudly during calls, Oyma‘s CoreML speech pipeline executes quietly on the dedicated neural block. Your MacBook remains silent and cool, preventing unwanted fan hum from bleeding into your microphone.
