Jarvis

Jarvis is an open-source macOS app that provides voice dictation and AI-driven text transformation for macOS. It allows users to record speech by holding a modifier key and inserts the resulting text directly into any active application. The software is deployed as a desktop application for Apple Silicon and Intel Macs, with a version available for iOS via TestFlight.

Users can choose between local processing for privacy or cloud-based APIs for speed. The application handles the transcription of spoken words and can automatically remove filler words, correct grammar, or reformat text into lists and emails based on user-defined prompts. It also supports basic system commands such as opening websites or setting timers. By utilizing a hold-to-speak mechanism, it integrates into existing workflows without requiring a dedicated interface for every interaction.

Key features

• Local transcription using Whisper or NVIDIA Parakeet via Sherpa-ONNX
• Integration with Ollama for local LLM processing
• Cloud transcription and AI options via Deepgram and Gemini
• Automatic removal of filler words like "um" and "like"
• Customizable prompt engineering for text formatting and behavior
• Hands-free mode toggle via keyboard shortcuts
• Support for Apple Silicon and Intel macOS architectures
• Zero telemetry and tracking by design

Jarvis is designed for users who require a private, subscription-free method of dictation. The architecture supports a fully offline workflow when paired with local models, though performance varies based on hardware specifications. Users with M1, M2, or M3 Max and Ultra chips experience faster processing, while standard hardware may see a few seconds of latency during AI post-processing. It is particularly suited for macOS users who want to automate text entry and basic app control without sending data to external servers. The software allows for deep customization of how the assistant behaves, enabling users to tweak how it formats mail or cleans up raw dictation.

Jarvis is a privacy-focused, MIT-licensed voice assistant.