v2026.9.25-2
The Linux dock and app menu show the Dictate icon instead of a generic microphone.
AI agents and tools can read these release notes as plain Markdown.
This release covers 17 pull requests.
Highlights
- The Linux dock and app menu show the Dictate icon instead of a generic microphone.
- The packaged engine includes the X11 hotkey library, so a normal desktop session can dictate without being in the
inputgroup. - Pasting into a terminal uses plain text.
- The dictation list and Copy last dictation follow when the words were spoken.
- A
.debupdate downloads while Dictate stays open. - Stable and beta are separate builds.
Changes
- The About dialog no longer shows a model name. Version, package, and the update channel stay.
- Starting a meeting, and the All / Meetings / Quick filter, are on the beta channel only. The normal channel is dictation, and every saved note stays in the one list.
- Dictation into a terminal, including the Grok CLI, is inserted with Shift+Insert as plain text. Ctrl+V in that CLI treats a WebKit clipboard as a file chip.
- A
.debupdate downloads while Dictate stays open. The app installs that package and restarts only after a second click. - Stable and beta are separate builds. The installed version decides which one you have. About no longer switches between them.
Fixes
- The Linux dock and app menu show the Dictate icon instead of a generic microphone. The packaged app's own launcher entry used a fallback icon that hid the one the
.debinstalls. Existing entries are repaired on next start. - The packaged engine includes the X11 hotkey library, so a normal desktop session can dictate without being in the
inputgroup. - Pasting into a terminal uses plain text. The focused window is recognized from its process and window class, including Ghostty, not from a single app name.
- A disconnected engine no longer writes stock demo sentences into the dictation list.
- Dictation stays on CPU when every NVIDIA GPU is bound to vfio for a VM. The RTX 4090 passed through to Windows is not taken.
- The dictation list and Copy last dictation follow when the words were spoken. A crash recovery that stamped old notes with a later finish time no longer floats those notes above a new dictation.
- A saved backend this build does not ship, including the removed xAI, OpenAI, and Gemini backends, is rewritten to Parakeet when the desktop UI loads. The window was reporting
faster-whisperfor that stale setting even though the engine was already running Parakeet.
Contribution record
CHANGELOG/records/2026.9.25-2.md lists every pull request in this release once, with its related issues and thanks.