What’s new

v1.0.35 · Build 46

Released September 26, 2026 · Apple silicon · macOS 14+

• Improve focused-field discovery and retry briefly when an editor exposes its Accessibility information late.
• Allow normal paste into the original ChatGPT or Codex desktop window when its input field is unavailable through Accessibility. The original app, window, and user-input activity must remain unchanged; secure input and known non-editable controls are excluded.
• Use the same compatibility handling for Paste Last Dictation and show clearer Recovery explanations.

Keep the chat input focused until dictation finishes. If automatic paste cannot be sent, copy the text from Dictation Recovery. Recovery items remain until dismissed or Saransh quits, so copy any needed text before updating.

v1.0.34 · Build 45

Released September 25, 2026 · Apple silicon · macOS 14+

• Meeting Recovery: retry transcription from locally saved audio after failure or interruption. Recovery uses samples already written to disk; it cannot restore deleted or unwritten audio.
• Audio retention: keep new recordings for 7 days, 30 days, or until you delete them. The default remains deletion after successful transcription. Failed or interrupted recordings do not expire automatically.
• Transcript playback: listen from matching passage timestamps, adjust speed, skip ten seconds, and optionally follow the current passage. Playback requires retained audio.
• Input checks: test microphone and computer audio locally for ten seconds. During capture, separate indicators show activity and reported errors.
• Chapters and sources: use your configured AI provider to create topic navigation and an editable AI note with decisions, actions, and supporting quotes. Source links open transcript passages inside Saransh; they are not public sharing links.
• Recording fixes: improve speaker-echo handling, ignore explicit non-speech annotations, and protect audio files during finalization.

Choose retention before your next recording to enable playback. Changing retention applies to new recordings; it cannot restore audio already deleted. Time-based expiry runs when Saransh starts.

Chapter generation shows where transcript text will be processed before you start. Cloud providers receive text, not audio, and charge your provider account separately. Available on-device Apple Intelligence or a server running on your Mac can keep this text processing local. Review every AI interpretation, even when a supporting quote matches.

Set up the new features · Read the FAQ · Review storage and privacy

v1.0.33 · Build 44

Released September 25, 2026

• Read transcripts by speaker, search passages, and copy individual turns.
• Edit passages while retaining matching timestamps and identify AI notes that need regeneration.
• Keep speaker correction controls visible and improve smaller-window layouts.

Download the latest version, or use Check for Updates in Saransh.