Loquela
Patch Notes

Android v0.15.2-hotfix.3: Enter works where apps disagree

Hotfix 3 fixes Search and Go in Google and smaller apps without bringing back accidental Discord sends, and adds focused Enter diagnostics for future editor quirks.

3 min read
Version
0.15.2-hotfix.3 (33)
Channel
Google Play open testing
Status
Available in Google Play open testing

Loquela v0.15.2-hotfix.3, version code 33, is now available through Google Play open testing in Australia, the United Kingdom, and the United States.

This build keeps the first v0.15.2 release's voice-model and diagnostic changes, plus the simpler board defaults from the first hotfix. It finishes the Enter correction after real-device testing found that some Google and smaller apps publish conflicting field flags: their fields request Search or Go while also looking multiline to the keyboard.

Loquela now treats explicit Search and Go actions as navigation requests, while still treating multiline Send fields as ordinary newline composers. That preserves the Discord behavior testers asked for without leaving Google Search, browser-style address fields, and similar app surfaces unable to submit.

Added

  • Debug now has an opt-in visual console for recent Loquela warnings and errors.
  • The console shows timestamps, severity, fixed categories, and privacy-safe descriptions rather than raw system logs.
  • Debug now has a separate, default-off Enter key diagnostics view. It records the editor's field type, declared action, Android Enter flags, and Loquela's dispatch result, never the app name or typed text.
  • Debug now shows recent transcription failure reasons directly in the app, with a separate choice for how many entries appear on screen.
  • Keyboard customization now has an Enable extra boards switch. Emoji remains available when it is off.

Changed

  • Voice-model cards now say Select Model and assign the model to its Batch or Sequential slot without silently changing the active mode.
  • Both uses the same selected Sequential and Batch models, so either model can be changed independently and reused when switching modes.
  • Home and setup status now reflect the models required by the selected Batch, Sequential, or Both mode.
  • Onboarding recommendations respect the device language and no longer suggest an incompatible starter model merely because it fits the device tier.
  • The feedback shortcut now sits at the bottom of General Settings with a clear external-link treatment.
  • Diagnostic list sizes control only what is shown on screen. The separately bounded local report keeps the complete events that still fit.
  • Extra boards are off by default. Enabling them restores Favorites, emoticons, kaomoji, ASCII art, saved shortcuts, custom boards, and the board rail.
  • The symbols key once again changes between pages 1/2 and 2/2 by default. Holding it opens Emoji and any enabled extra boards.

Fixed

  • Both no longer behaves like Sequential-only when its older dedicated model slots were empty or stale.
  • Installed model choices remain usable across modes without creating duplicate Batch and Both selections.
  • Warning and failure history no longer discards older diagnostic events merely because the on-screen display count is smaller.
  • Explicit Search and Go fields now perform their declared action even when the app also supplies misleading multiline or no-Enter-action flags.
  • Multiline Send fields still insert a newline, preventing accidental sends in Discord-style composers that own a separate Send button.
  • Done, Next, Previous, and ordinary single-line Send fields continue to perform their declared app action, and the Enter key changes its icon and accessibility label to match.

Privacy

  • The warning/error console is off by default and stores only fixed Loquela event categories in app-private storage.
  • Enter key diagnostics are separately opt-in and exclude app identity, typed text, surrounding text, and editor contents.
  • Diagnostic history does not contain typed text, transcripts, audio, clipboard contents, raw touch coordinates, account details, or secrets.
  • Nothing in the diagnostic console is uploaded automatically.

Known issues

  • Nemotron remains a beta model whose speed, memory, battery, temperature, and editor behavior need broader physical-device testing.
  • Both requires Nemotron plus a compatible installed Batch model and uses more storage, memory, and processing than either pass alone.

For the design behind Batch, Sequential, and Both, read How Loquela is building live and two-pass local voice typing.

Get Loquela, browse Voice Models, or send feedback.