Sayso is a free, open-source dictation app for macOS, Windows, and Linux that runs speech recognition locally by default. Press a configurable hotkey (Option+Space on macOS, Ctrl+Space on Windows, Ctrl+Alt+Space on Linux) to speak, see a live overlay with waveform and words, and have the transcribed text pasted into the focused application while your clipboard is restored. It’s written in Rust with GPUI; macOS uses a Swift engine with WhisperKit and FluidAudio, Windows uses transcribe-rs (ONNX Runtime and whisper.cpp), and Linux uses sherpa-onnx. Bundled local models include Parakeet, Nemotron, several Whisper variants and others (Parakeet Unified is the default at ~1.3 GB), and cloud providers (OpenAI, Groq, ElevenLabs, Deepgram, AssemblyAI, Mistral, or any OpenAI-compatible server) are optional. Built-in "AI styles" are editable TOML prompts that post-process transcripts (Clean, Email, Message, etc.) and can be applied per app or site; a dictionary and history (text stored until deleted; audio expires after 30 days) support accuracy and retrieval. Sayso preserves on-screen text if insertion or styling fails and can import data from Pindrop.
Installation is provided for macOS (signed DMG or Homebrew cask), Windows (user installer), and many Linux formats (.deb, .rpm, Flatpak, AppImage, AUR, Nix), with docs listing Wayland/X11 and desktop-specific permissions. Privacy defaults keep audio and text local; enabling a cloud model or an AI style sends data to the chosen provider and API keys are stored in platform keyrings. Official builds check releases once daily (IP visible to GitHub) but this can be disabled; source builds avoid that check. The code is licensed GPL‑3.0 and can be inspected, modified, and built without cost.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.