Editor recovery, task checkboxes, heading styles, transcription reliability, speech-to-text settings, session UI polish, and audio cleanup were improved.
Editor
Editor crashes are handled more safely so a bad node view or failed transaction is less likely to bring down the app
Heading levels in notes now render at distinct sizes, making H1, H2, and H3 easier to tell apart
Simple task checkboxes are back, with direct todo/done toggling while preserving existing task status data
Floating editor menus now have clearer borders and more consistent shadows
Top-level headings in AI-rendered output now scale down so they sit closer to body copy
Transcription
Interrupted batch transcriptions now save progress more reliably so recovered sessions keep completed transcript text
Local batch transcription is more resilient for very short Qwen recordings and long Parakeet, Omnilingual, and Qwen recordings
Speaker count hints are passed through live and batch transcription providers to improve diarization where supported
Transcript-only batch responses are now persisted with synthesized word timings, and empty batch output surfaces as a no-speech error instead of disappearing
OpenAI Whisper batches now request word timestamps, and synthesized timings no longer trigger audio seek when clicking transcript words
Soniqo batch recordings preserve each channel so multi-mic captures stay distinguishable in the transcript
Inactive sessions now open the transcript at the top instead of jumping to the latest line
The collapsed live transcript footer is shorter and tighter, giving notes more room during recording
Batch transcription no longer doubles up progress between the header and the panel — skeleton rows fill the panel with cleaner spacing while batching
Regenerating a batch transcript now keeps the transcript panel visible while progress runs
Browser-based meetings are now detected during active listening so recording can stop automatically when the meeting ends
Speech-to-Text Settings
Speech-to-text provider badges have been cleaned up so models are not shown as beta when they are no longer beta
Legacy local models are now labeled as deprecated, the legacy Anarlog provider card has been removed, and local model folder/delete controls live directly in the model picker
The STT settings layout has been simplified — language warnings appear as a top banner, legacy local choices stay hidden unless already selected, selected and downloaded model rows are more compact, and local model actions reveal on hover
The Soniqo model list now only advertises Parakeet Streaming and Parakeet Batch, while existing sessions on legacy Soniqo model IDs still resolve
Clearing every spoken language now sticks across restarts instead of being replaced by OS defaults
Data & Cleanup
Deleted notes now stay deleted after restart once undo is no longer available
Retained recorder artifacts are cleaned up in the background, and cached audio state refreshes after deletion
When audio retention is set to delete recordings, audio is preserved until transcript words exist so unprocessed sessions can still be transcribed
UI & Changelog
Sidebar and tab chrome now share a more stable layout, with the left sidebar locked to a consistent 200px width
Main language settings now use localized display text
The chat shortcut now appears on inactive transcript-backed sessions, tucking into a peek state while scrolling and revealing on hover near the bottom
Auto-enhance skip reasons now display in the chat shortcut slot, fading smoothly between the message and the button
Locked calendar provider rows now reveal their pro upgrade action only on hover or focus, keeping the sidebar quieter
The transparent transcript resize divider no longer renders a gray gap under the session border
Changelog pages are now available on the Anarlog website, with cleaner release summaries
