akshayaggarwal99/jarvis-ai-assistant

What it solves

Jarvis AI Assistant is a free, open-source voice dictation tool that allows users to convert speech to text anywhere on their computer. It eliminates the need for expensive subscriptions by providing a local-first approach to transcription and AI-driven text refinement.

How it works

Users hold a designated key (like the Fn key) to record their voice, and upon release, the recorded audio is processed into text. The system supports two primary modes of operation:

• Local Mode: Uses local models like Whisper or NVIDIA Parakeet (via Sherpa-ONNX) for transcription and Ollama for local LLM post-processing to clean and format text.
• Cloud Mode: Uses Deepgram and Gemini for faster, cloud-based transcription and refinement.

Who it’s for

People who want a private, fast, and free alternative to paid voice dictation apps, particularly those who prefer running AI models locally to avoid telemetry and subscriptions.

Highlights

• Local-First Privacy: Full support for local transcription (Whisper/Parakeet) and local LLMs via Ollama.
• AI Post-Processing: Automatically removes filler words ("um", "like"), fixes grammar, and can rephrase or bullet-point text.
• Customizable Behavior: Full prompt engineering capabilities to tweak how the assistant formats emails or cleans dictation.
• System-Wide Integration: Works across any application by inserting text directly into the active window.
• Crosspoint Support: Available for macOS (Apple Silicon and Intel) and iOS (via TestFlight).