Features
- macos-native banner styling and hover close
- granola-style banner layout
- drop the UN notification layer
- route banner window actions through the main window start and stop paths
- floating banner webview with auto-dismiss progress
- floating banner window with nspanel conversion
- pure banner window state controller
- branded icon set and a properly named dev binary
- call banners, detection settings and a saving state
- event-driven call detection with optional auto-stop
- pre-titled sessions, early file save and an activity clock
- beam-search quality profile for the offline whisper pass
- add audio_from frontmatter field and split_transcript_at
- auto re-transcribe and audio cleanup settings
- optional auto re-transcribe and audio cleanup
- curate the lineup to turbo and large-v3
- model manager with selection and deletion
- roles, re-transcribe model selection and deletion
- default to aac, keep flac as an option (stage-0.7 format change)
- keep-audio settings toggle and whisper Use/Active gating
- audio playback with click to seek and a re-transcribe action
- re-transcribe a call with whisper
- whisper large-v3-turbo pass via whisper-rs
- retain call audio as stereo flac next to the transcript
- record into the open note and continue its transcript
- new note button, sidebar hover and selection states
- session view with notes editor and live transcript
- sidebar shell with folder-backed session list
- light and dark themes following system with override
- interim transcript decoding while speaking
- folder-backed session ipc with single-writer routing
- file format v2 with title, notes and parser
- settings dialog with output folder and model manager
- main screen on shadcn
- in-app model registry and downloader with progress
- json config with output dir and active model
- start/stop ui wired to capture+asr pipeline
- capture->vad->asr->md pipeline
- ordered transcript store with atomic md writes
- system audio via core audio process tap (cidre)
- mic capture via cpal, 16k mono chunks
- silero vad + parakeet tdt v3 via sherpa-onnx
- model download script (parakeet v3 int8 + silero vad)
Fixes
- hide main window on close and reopen it from the Dock
- set minimumSystemVersion 14.2 so whisper.cpp release build links
- entitle the microphone under the hardened runtime
- review findings: window capability, hover reset, recording-live tracking
- ad-hoc sign bundles so notification registration works
- saving a stopped session no longer blocks starting the next one
- speech-only offline pass, segments split at region seams
- always pad flac's trailing block to a full block size
- de-flake queue-drain tests via a non-spawning seam
- surface suppressed background errors, refetch stale transcript
- delete-after-retranscribe no longer requires keep_audio
- keep pre-audio_from transcript text; atomic enqueue
- record audio_from when continuing without prior audio
- clear manual re-transcribe tracking on early rejections too
- track manual re-transcribe clicks to fix auto-error reporting
- re-validate queued re-transcribe jobs before they run
- kill the whole process group on watchdog timeout
- distinguish retired-but-downloaded from missing in Settings
- keep the settings dialog on screen with a scrolling body
- reclaim stale download temp files and correct model manager copy
- disarm delete confirmation on dialog close and stale refresh
- bound afconvert/afinfo subprocess calls with a timeout
- recover crashed continuation audio instead of truncating it away
- remove stale flac on encode failure, resolve crashed continuation audio
- chain the done-phase refresh's state updates to avoid seeding a remount with stale data
- fix wav-fallback collision and reconcile prefill length with transcript offset
- refresh a remounted SessionView after a background re-transcribe
- reject re-transcribe/audio-path reads during the stop-to-encode window
- prefill a continued session's audio instead of overwriting it
- scope the asset protocol to the output dir at runtime
- guard against an empty re-transcribe wiping the transcript
- flush the in-progress wav periodically so a crash is not fatal
- stop losing duration and pending edits when continuing a note
- visible sidebar hover, filled selection without the bracket, button press states
- keep selected sidebar row hover from dimming, use destructive token
- final review findings for stage 0.6
- flush active-session edits before stop, normalize empty titles
- clear stale session data on switch, dedupe refresh listener
- handle set_theme/set_output_dir IPC failures in settings dialog
- correct max-speech forced-cut timestamp offsets
- preserve foreign UTC-offset dates verbatim on rewrite
- guard settings dialog download button against double-clicks
- green start button, reset elapsed timer on record start
- pin dark: variant to html.dark instead of prefers-color-scheme
- require decoder/joiner files and id-scope download tmp path
- restore missing yo as ye, drop stray unk tokens
- don't block speaker tap start on tap_auto_start
- feed vad's trailing sub-window tail on flush
Other
- style(capture): cargo fmt
First release.
Install
Open the disk image, drag simstim into Applications, launch it. Signed
with a Developer ID and notarized by Apple, so there is nothing to
bypass. ASR models download inside the app on first run
(Settings > Models).
Requires macOS 14.2+ on Apple Silicon.
shasum -a 256
e631c05dc9f32e3868ac023f3c0d250c9ee787c6307e6ba627ca8f5cf975ea06 simstim_0.1.0_aarch64.dmg