F*ck Wispr Flow – Open-sourcing Jarvis: private, local, free (github.com)

Akshay Aggarwal has open-sourced Jarvis, a privacy-first, local voice-dictation and AI assistant for macOS that transcribes speech and pastes text wherever you’re typing. Designed as a lightweight alternative to commercial products like Wispr Flow, Jarvis runs natively (Electron) with a minimal waveform UI, is distributed as signed/notarized DMGs for Apple Silicon and Intel Macs, and was built in three months before being released under the MIT license. You trigger recording with the Fn key, speak, and Jarvis transcribes and formats your text instantly — all without cloud accounts or telemetry. Technically, Jarvis uses Deepgram Nova-3 for low-latency real-time transcription or OpenAI Whisper for transcription, and can use Google Gemini for formatting; API keys are entered locally and never leave your machine. It stores data under ~/.jarvis/, requires macOS 10.13+, mic and accessibility permissions, and exposes a repo for build (npm install; npm run build/dev) and contribution. For the AI/ML community this matters because it provides a reproducible, privacy-preserving voice+LLM workflow you can audit, extend, or integrate with your own models and keys — ideal for researchers and developers who need local control, reproducibility, and fast prototyping without vendor lock-in.