Show HN: Velora – On-device macOS dictation (Whisper and a local LLM, no cloud) (github.com)

Velora, a groundbreaking new macOS dictation tool, has made its debut as an open-source solution that prioritizes user privacy by operating entirely on-device. Utilizing advanced speech-to-text technology from Whisper and a local large language model (LLM), Velora allows users to dictate text effortlessly within any application without needing internet connectivity. All audio and transcripts remain securely stored on the user's Mac, and even custom dictionary entries can sync privately through iCloud without compromising personal data. Unlike comparable offerings that rely on cloud services and subscription fees, Velora eliminates network calls during dictation, ensuring a seamless experience with no telemetry or analytics tracking. The significance of Velora for the AI/ML community lies in its innovative use of local machine learning techniques, allowing users to harness powerful dictation capabilities while maintaining data privacy. The system supports multilingual transcription, app-aware smart formatting, and features like speaker diarization for meetings, all while delivering rapid response times and a user-friendly interface. By enabling users to dictate and edit their text with high accuracy and efficiency, Velora sets a new standard for on-device speech recognition tools, challenging subscription-based models and highlighting the potential of open-source software in the AI landscape. The attention to customization and user control, along with its performance metrics, illustrates its potential impact and usability in various productivity environments.