Every meeting-notes and voice-memo tool I tried wanted the same thing first: an account, a subscription, and my audio uploaded to someone else’s server. muesly is the opposite of that. It captures, transcribes, and summarizes everything you say, entirely on your own machine. No cloud, no server, no account, just a single self-contained desktop app.

The whole point is that nothing leaves your device unless you explicitly ask it to. Transcription runs locally using Whisper or Parakeet models, GPU-accelerated and in-process, with the transcript appearing in real time as you speak. Summaries are generated by a local model by default, though you can point them at your own Ollama server or a cloud provider if you want to. Recordings, transcripts, and summaries all live in a local SQLite database you can export or delete at any time.

The audio path is where most of the careful work sits. muesly does dual capture of your microphone and system audio at once, with proper mixing (ducking, clipping prevention) and voice-activity filtering so only speech reaches the transcription engine. GPU acceleration is auto-detected at build time: Metal and CoreML on macOS, CUDA or Vulkan on Windows and Linux.

Built in Rust on Tauri, it runs on macOS, Windows, and Linux. Available at muesly.ai.