Patch v1.5.4 has been released:
Improved speech-to-text reliability so selected app language is now applied more consistently during startup and language switching. Multilingual models are now loaded and applied successfully.
Fixed issues where speech-to-text could default incorrectly or stop responding in certain cases.
Enhanced vision and screen-awareness behavior to help prevent stale or repeated responses from previous screens, improving general overall accuracy when asking about what’s currently on your screen.
Fixed cases where automatic screen detection could fail or not update correctly on certain hardware setups.
Improved memory handling for the temporary visual context and system interactions.
Added a subtle animated “thinking” indicator in chat while the assistant is processing responses.
Fixed an issue where the startup greeting text could appear corrupted when using Russian on app reopen, while functioning normally afterward.
Improved overall assistant responsiveness and stability, including fixes for edge cases where responses could repeat or get stuck.
A higher quality speech-to-text option will also be included in a future update, allowing use of a more advanced transcription model. This will use more system resources but provide further improved accuracy over the current default models, as an optional feature.
