Canonical's New AI Tool Wants You to Talk to Ubuntu Instead of Type

Canonical has announced Myna, a local speech-to-text dictation tool planned for Ubuntu 26.10. Built around GNOME on Wayland, it uses a push-to-talk mechanism and processes audio entirely on-device via a sandboxed Canonical Inference Snap. Three model sizes (lightweight, default, quality) will support NVIDIA GPUs, Intel NPUs, and CPUs. Audio is discarded after each session and no internet connection is required. The project is still early-stage with only a GitHub repo skeleton, and Canonical is seeking feedback from users who rely on dictation or assistive tools on Linux.