config.env.example

# Copy to config.env (in ECHO_HOME, next to the scripts) and fill in.
# All keys are optional - without them the note simply stays in the session folder.

# Filesystem path where process_meeting.py --obsidian copies the finished note:
#OBSIDIAN_VAULT_PATH=~/Library/Mobile Documents/iCloud~md~obsidian/Documents/MyVault/Meetings

# For the obsidian:// deep link when you click a meeting in the app:
# vault name and the notes folder inside it (must match OBSIDIAN_VAULT_PATH).
#OBSIDIAN_VAULT=MyVault
#OBSIDIAN_DIR=Meetings

# Stop the recording by itself after this many minutes of silence in the system track,
# so a forgotten session doesn't run for hours. Default 15; 0 turns it off.
# Keep it generous: waiting for a call to start is real silence - 8-minute gaps happen.
# Only fires when system audio is being captured - a mic-only recording never auto-stops.
#AUTOSTOP_SILENCE_MIN=15

# How long the 16 kHz WAVs survive after the transcript is written. The raw CAFs
# (~1 GB per hour per track) are deleted immediately either way; the WAVs are 8x
# smaller and let you re-run the pipeline on real audio - which is the only way to
# tune diarization and the hallucination filters. Default 2 days.
# Sessions whose transcript came out empty keep their audio regardless.
#KEEP_WAV_DAYS=2

# Language whisper transcribes in. Fixed instead of auto-detected because on Russian
# speech peppered with English loanwords (deploy, meeting, feature) the model keeps flipping to English and even
# translating - over half a real meeting was lost that way. Set to your language's
# ISO code (ru, en, de, ...); "auto" restores per-chunk detection for mixed calls.
#WHISPER_LANG=ru