Canonical has introduced Myna, an early-stage AI tool for voice dictation on Ubuntu that runs entirely on local hardware. The feature is planned for Ubuntu 26.10, scheduled for release in October. It uses a push-to-talk system and focuses on privacy by keeping all processing on the device.
The tool was announced by Jean-Baptiste Lallement, Canonical's Director of Engineering for Ubuntu Desktop. It builds on plans outlined in April by Jon Seager for implicit AI features in Ubuntu, including speech-to-text capabilities.
Myna is designed for GNOME on Wayland. Users hold a hotkey to dictate, after which transcribed text appears at the cursor position. Recognition occurs in a sandboxed Canonical Inference Snap that supports models of varying sizes on CPUs, NVIDIA GPUs, or Intel NPUs.
No internet connection is required once the model is installed. Audio data stays in a temporary buffer and is discarded after each session. Features such as wake words and dictation in password fields are not included at this stage.
The project remains in early development, with only basic files on GitHub so far. Canonical is seeking feedback from users who rely on dictation tools, and daily builds of Ubuntu 26.10 may include Myna in the coming weeks.
