Whispermate – open-source alternative to Wispr Flow (github.com)

WhisperMate is an open-source, native macOS voice-to-text app that uses Groq’s Whisper Large V3 for fast, cost‑effective transcription. Released as a lightweight Swift/SwiftUI app (~1.35 MB vs 200+ MB for Electron alternatives), it delivers 400–800 ms inference times, minimal CPU/memory use, and features hold-to-record or continuous hotkey recording, auto-paste into the active app, overlay/full-window modes, and optional LLM-powered transformations (translation, tone/formality adjustment, custom glossaries). The project is MIT-licensed and the repo is available for auditing and local builds (Xcode 15+, macOS 13+). Significance for AI/ML practitioners lies in privacy, performance, and reproducibility: WhisperMate sends audio only to Groq’s API (no third-party servers), transcribes then immediately discards audio, and stores API keys securely in macOS Keychain—so data handling is fully auditable in source. The app’s architecture (AVFoundation audio capture, GroqAPIClient, Keychain helper) makes it a practical reference for integrating low-latency cloud inference into native macOS tooling. Limitations to note: it’s macOS‑only and requires a Groq API key (beta is free aside from API usage). WhisperMate positions itself as a lightweight, transparent alternative to heavier desktop speech tools like Wispr Flow.