Corti watches macOS's microphone-in-use signal. When Zoom, Slack, Google Meet, Discord, or another app grabs the mic, capture begins automatically. The owning process becomes the source tag on the note.
[AUTO] identifies whichever app owns the microphone
[DUAL] captures your mic and far-end audio separately
[LOCAL] runs Parakeet on CPU or Metal
[LIVE] shows timestamped Me / Them transcript rows
[DURABLE] streams bounded writes into the Vagus inbox
NO CALENDAR HOOK. NO “JOIN MEETING” BOT. NO ACCOUNT.
CORTI // LIVE● REC
REAL REACT FRONTEND // DETERMINISTIC TAURI FIXTURE // RETINA CAPTURE
LIVE WRITE PATH
Corti live transcript write path
01MIC OWNERZoom
→
02COREAUDIOMe + Them
→
03PARAKEETlive ASR
→
04SYNC_ALLbounded chunk
→
0500-INBOXplain Markdown
LIVE WRITE AUTHORITY
THE NOTE EXISTS BEFORE THE CALL ENDS.
With the local backend, Corti appends a bounded window every 1–10 minutes. Each complete chunk is OS-synced before memory is reused. A crash can lose the active interval—not the whole conversation.
State: transcribing → State: transcribed
VISIBLE REFLEX
READ THE ROOM. WHILE YOU'RE IN IT.
The menu-bar reader displays call-relative timestamps and Me / Them labels as speech regions close. Its idle microphone test uses the same local ASR without saving audio or creating a note.
Audio expires on schedule. Notes live on in the brain.
PIPELINE
The lit stages are active right now.
LOCAL MODE
YOUR AUDIO. YOUR MODELS.
Select private local Parakeet transcription, Metal acceleration, far-end diarization, echo cancellation, and one-minute durable inbox windows from the app.
Every quality-relevant choice stays visible and local: backend, runtime, models, diarization, echo cancellation, retention, and live filing cadence.
SETTINGS // LOCAL
Private local inference remains one runtime choice; AWS Transcribe is also available.
The ear writes Markdown. Two retrievers make it recallable.