FluidVoice v1.6.12

FluidMeet

• New Meeting, settings and history stay visible inside the meeting page. Starting another meeting waits for a running summary to finish or be cancelled. #1156
• Set your own start/stop meeting shortcut in Shortcuts settings; unassigned by default. Uses the saved meeting source. #1157
• Export transcripts as Markdown (.md), Text or JSON, with timestamps and speaker names. #1158
• Preserve microphone recordings from devices that provide integer audio formats. #1152
• Clearer microphone speech when other audio is playing. If echo filtering is unavailable, keep the speech and show a small possible-echo notice.
• Live captions finish phrases before recovering from a stall, with fewer interruptions. Floating captions can be dragged on macOS 27. Thanks to @argus-vault for the dragging fix. #1061
• Preserve recognized words near sentence endings and join unlabelled fragments to the preceding speaker.
• Edit text in place with a double-click. Rename and merge speakers from Edit speakers, confirm merges, and use Command-Z to undo corrections. #1187
• Meeting history loads details as needed instead of reading every recording up front.
• Keep processing versions consistent across saved meeting runs.
• Turn FluidMeet on or off from its settings without deleting saved meetings.
• Summaries can run on this Mac or with your own AI provider, with an optional custom prompt, and stay available after restarting the app.
• AI summaries show a brief accuracy reminder.
• Let Claude Code, Codex and other coding agents read meetings from FluidMeet settings → Integrations.
• Browser meetings are detected by default. End reminders cover manual recordings and Safari Meet even with automatic start detection off; confirmed call exits prompt sooner.

Other changes and fixes

• Updated the app icon with a silver mark and subtle edge shine. #1221

• Reduced startup delays when starting dictation.

• Dictionary + opens Add Manually without blocking normal dictation. Compact buttons keep the list tidy; Test word stays inside the Edit popup.

• Dictionary learning has clearer recording and save steps, pauses between examples, and editable spelling corrections.

• Clearer correction labels, accurate dictionary search counts, and feedback when nothing new is saved. #1151, #1165

• User dictionary spellings override defaults. Pending learned rules are cancelled when text is edited or observation ends.

• Saved dictionary pronunciations carry across Parakeet, Mini and Pico. Pronunciation learning is now on by default. Dictation keeps its normal processing until a voice word is trained.

• Selecting Fluid Speech Mini or Pico no longer shifts the Voice Engine page upward.

• Refresh idle AI connections to avoid stalls on the next request. Thanks to @andrewmitchell-automations. #1134

• Switch between Dictation, Edit and Command during a recording without a false “Dictation Unavailable” alert or restarting capture. Preserve the starting field during mode changes.

• Paste shortcut tests now reject unexpected modifier keys. #1193

• Disabled Command and Paste Last shortcuts no longer block reusing their keys. #1213

• Recovery Copy uses the current transcript and restores the retry window when copying fails. #1198

• Recovery and microphone-change popups stop redrawing after their countdown ends, reducing idle CPU use. #1218

• Login launches keep the main window hidden when requested, including after a restart. Opening FluidVoice from the menu bar reuses that window.

• Fluid Intelligence model downloads retry brief network interruptions and resume saved progress. Working models stay available during a failed update.

• Fixed AI requests to newer Claude models that reject a temperature setting.

Contributors

Thanks to @andrewmitchell-automations, @grohith327, @argus-vault, @yyjosef, @dxcore35 and @nickbecker for 1.6.12 fixes, improvements and reports.