v0.14.1 Highlights
A small patch release. It fixes model downloads, makes the mobile app nicer to use, and gives the IO tracer a full rebuild.
Bug fixes
- Voiceprint enrollment on desktop no longer needs a download. Before, enrollment could fail with only
fetch failedwhen the voiceprint model could not be fetched from Hugging Face. The desktop app now ships with the model, so enrollment works offline from the first try. - Failed model downloads now tell you what went wrong. Before, a failed download showed a raw error such as
fetch failed. It now shows a plain sentence, for example that the connection failed, the server returned an error, or the download stopped early. Each failed model also has a Retry button, so you don't need to restart anything. The voiceprint settings page on Web and mobile shows the same explanation. - The Android app stops asking for the microphone on every visit. Before, Settings > Hearing showed "Allow microphone access" each time you opened it, even after you had allowed it. The picker now lists your microphones right away. If you have not allowed access yet, it opens the system dialog for you.
- The mobile stage no longer shows a stuck hearing error. With Listen on and no speech recognition provider, a status pill above the composer showed an error that never went away, even when nothing was listening. The composer microphone and voice presence already show your input state, so we removed the pill.
Product updates
Mobile
- The composer is a little narrower while idle, so more of your character shows. It widens again when you start typing, attach an image, reply, or join a call.
Voice
- Voice draft problems no longer squeeze the draft. Before, an error appeared as a red line that made the draft area smaller and made a working draft look broken. You now see a small warning icon with a count, and you can hover over it to see the list. Microphone and wake word problems now show there too, not only voice input errors.
Look and feel
- We removed many decorative borders across the app. Selected cards, voices, characters, Flux packages, and AIRI Cards now use a soft fill and their check mark or badge to show the selection. The chat panel, provider tags, and module pages look cleaner. Select menus and dropdowns now match the text inputs.
- The character switcher on mobile now crops previews from the top, so tall models keep their heads in view.
Languages
- Korean got a big update. One term now names one concept (for example, Speech, Hearing, and a character's memory), and 467 strings that showed in English are now translated. Thanks, @Daeil-Jung!
No upgrade action is needed. One note for desktop users who upgrade: an older voiceprint model download can stay in your app data folder. To clear it, use "Delete all data" in Settings > Data.
Full Changelog: v0.14.0...v0.14.1