@dreamteamer/recordings

Audio recordings of your meetings, each matched to its meeting and transcribed on your own machine: Hebrew and English, speakers told apart.

npm install @dreamteamer/recordings @dreamteamer/meetings @dreamteamer/contacts @dreamteamer/companies @dreamteamer/users @dreamteamer/assets
npx dreamteamer compile

the flow

1. Drop audio into <an INBOX_FOLDERS root>/meeting-recordings/, named [date] [who] where you can (2026 08 30 1500 Northwind weekly.m4a, Dana Oren jul22.m4a; a Hebrew month name works too).
2. /fetch-new-recordings — dedupes by content hash, archives each new file under ${env:FILES_FOLDER}/meeting-recordings/, moves the original to the inbox root's trash/, and creates a meeting-recordings record.
3. /match-recording-to-meeting — links each recording to its meeting.
4. /transcribe-recording — writes the transcript into the recording record.

A file handed straight to a session, or a pasted transcript, skips the inbox: the transcribing-recordings skill covers both.

| alone | with others | |---|---| | recordings pulled from an inbox by content hash, matched to their meeting, transcribed; each meeting's recordings list generated from the recordings that point at it | meeting-docs reads the transcript: summarise, extract tasks, critique |

| needs | for | |---|---| | @dreamteamer/meetings (and what it needs) | the meeting a recording belongs to | | INBOX_FOLDERS and FILES_FOLDER, declared in your workspace's dreamteamer.vars | where recordings arrive (a newline-separated list of roots) and where they are archived (a local folder) | | ffmpeg, ffprobe, uv and about 1.6 GB of models, downloaded once | transcribing audio; a pasted transcript needs only uv | | TRANSCRIBING_MODELS_DIR, an absolute path exported in the environment that runs the transcriber | where the models live; without it they sit inside the skill folder, which compile and npm ci regenerate |