OpsTrellis v0.2.0

The first public release of OpsTrellis: voice dictation for macOS and Linux. Hold a key, speak, let go, and clean text appears in any app. Offline by default.

Install or update

curl -fsSL https://raw.githubusercontent.com/mtr1920/opstrellis/main/install.sh | bash

Or download a file below: .dmg for macOS (Apple silicon), .deb or AppImage for Linux (x86_64). The installer checks every download against SHA256SUMS.

Highlights

• Dictate anywhere: push-to-talk or hands-free, with a live recording pill.
• Offline by default: local speech recognition (Moonshine, Whisper, Parakeet, SenseVoice), plus optional offline AI polish with S1-mini by Superwhisper.
• Single-key shortcut on macOS: hold 🌐 fn to talk, double-tap for hands-free. Ctrl+Shift+Space always works as a fallback.
• Developer-aware cleanup: code names, paths, flags and tech terms come out right; prompts for ChatGPT, Claude, Cursor and coding agents are cleaned up.
• Command mode, snippets, dictionary with auto-learn, per-app style, meeting notetaker, history and insights.
• Bring your own provider: OpenAI-compatible endpoints, Deepgram, AssemblyAI, Google, Azure, ElevenLabs and Anthropic, with automatic fallback.
• AI polish keeps your dictionary words: cloud models are told your terms, and if any AI polish (cloud or offline) rewords or drops a dictionary word you said, your plain text is inserted instead.
• Safer pasting: if you switch windows while it's processing, the text goes to your clipboard instead of the wrong app.
• Logs (Settings → About) show where time goes, never what you said.

Notes

• macOS: the app isn't signed with an Apple Developer ID yet. After installing, allow Microphone and Accessibility (and Input Monitoring for 🌐 fn) in System Settings → Privacy & Security.
• Free for personal and other noncommercial use; commercial use needs a license, see COMMERCIAL-LICENSE.md.