Skip to content

Latest commit

 

History

History
192 lines (144 loc) · 6.13 KB

cli.md

File metadata and controls

192 lines (144 loc) · 6.13 KB

The cassini CLI from a checkout

This page is for people working in the repo. If you want to run Cassini on a Nextcloud, install the app instead — see Installing Cassini as a Nextcloud ExApp. If you want to read published recordings from outside Nextcloud, see Agent access to meeting recordings; that path needs no checkout.

From a source checkout, use the wrapper:

./bin/cassini

It builds the current cassini CLI from the Go module and runs it from your current working directory. Output paths are resolved from that directory.

First commands

See what the CLI exposes:

./bin/cassini --help

Validate the environment before expensive work starts:

./bin/cassini doctor

If doctor reports an unwritable Cassini cache or model directory, fix that path or point Cassini at a writable cache root before building:

export CASSINI_CACHE_ROOT="$PWD/.cache/cassini"
./bin/cassini doctor

The main flow

Set your Talk room URL:

export CALL_URL="https://cloud.example.com/call/<ROOM_TOKEN>"

Record a meeting and finish with one portable file:

./bin/cassini record --call "$CALL_URL" --out "./My Meetings/2026-03-11 Weekly Sync.opus"

Recording defaults to audio only. To retain camera video in the source capture, explicitly add --retain-video:

./bin/cassini record --call "$CALL_URL" --retain-video --out ./runs/weekly-sync.run

--retain-video=false explicitly selects the default. The portable .opus output contains audio only in either mode. CLI capture flags apply to that invocation; installed recordings use the administrator's saved Capture video policy. Rebuilding an existing source keeps its media and does not start another capture.

If Cassini fails after capture or during processing, fix the issue and rerun the same command with the same --out path. Cassini keeps resumable state in a hidden .cassini-work/ directory next to the target file and reuses the finished recording or finished meeting artifact where it can.

Inspect the resulting file:

./bin/cassini inspect "./My Meetings/2026-03-11 Weekly Sync.opus"

If you already have a recording, build the portable file from that:

./bin/cassini build /path/to/meeting.mkv --out "./My Meetings/Imported Meeting.opus"

Trying it without a real call

The portable-file path needs a real meeting. For local smoke tests and diagnostics, simulate mode writes a debug .run bundle:

./bin/cassini record --simulate --out ./runs/demo.run
./bin/cassini inspect ./runs/demo.run

Browse an already-generated sample site from this checkout:

./bin/cassini serve ./cassini-viewer/exports/static-meetings

If you have processed .opus recordings, build a browser view of them in one step:

cd cassini-viewer
npm install
npm run build
node ./scripts/export-static-meetings.mjs \
  --source-dir /path/to/your/processed-opus \
  --output-dir /tmp/cassini-opus-view
cd ..
./bin/cassini serve /tmp/cassini-opus-view

The explicit pipeline

To keep the internal working artifacts visible, run the stages yourself:

./bin/cassini record --call "$CALL_URL" --out ./runs/weekly-sync.run
./bin/cassini build ./runs/weekly-sync.run --out ./meetings/weekly-sync.meeting
./bin/cassini publish ./meetings --out ./site
./bin/cassini serve ./site

inspect reads any primary Cassini artifact:

./bin/cassini inspect "./My Meetings/2026-03-11 Weekly Sync.opus"
./bin/cassini inspect ./runs/weekly-sync.run
./bin/cassini inspect ./meetings/weekly-sync.meeting
./bin/cassini inspect ./site

See Artifacts and filesystem for what each artifact holds.

When several people shared one participant's device, separate their voices in a built bundle. diarize measures one participant's own track (it needs the Nemotron diarization model, see Separating voices on a shared device); apply rewrites the bundle from its original transcript and runs no model:

./bin/cassini speakers show ./meetings/weekly-sync.meeting
./bin/cassini speakers diarize ./runs/weekly-sync.run --speaker spk_... --out ./turns/spk_....json
./bin/cassini speakers apply ./meetings/weekly-sync.meeting --edits ./edits.json --turns-dir ./turns
./bin/cassini pack ./meetings/weekly-sync.meeting --out "./Weekly Sync.opus"

The harness

The local Talk lab and the showcase/demo flows live under cassini dev:

./bin/cassini dev stack up
./bin/cassini dev room create --name "Local room"
cp .envrc.example .envrc
direnv allow
./bin/cassini dev smoke
./bin/cassini dev fixture prepare-showcase
./bin/cassini dev player showcase --call-url "$CALL_URL"

The harness implementation is under harness/.

Viewer demo data

See Development in the viewer README for the meetings the viewer dev server reads and how to fill it.

The commands

cassini annotate   Read and write the tags and marks a packed .opus file carries
cassini build      Build a browser-ready meeting artifact from a recording
cassini dev        Access the local harness namespace
cassini doctor     Validate the local environment before expensive work starts
cassini insight    Ask one question of several meetings and keep the answer
cassini inspect    Inspect a run, meeting, site, or lower-level Cassini artifact
cassini meetings   Read the meeting recordings your Nextcloud account may access
cassini operator   Launch the separate cassini-operator binary
cassini pack       Pack a built .meeting bundle into a portable .opus file
cassini publish    Publish one or more meeting bundles as a static site
cassini record     Record a meeting into one portable artifact
cassini retag      Rewrite a packed .opus file's room, job and recording-start fields without re-encoding it
cassini serve      Serve a static Cassini site locally
cassini speakers   Separate the voices of people who shared one device

cassini meetings reads published recordings out of Nextcloud as a given Nextcloud user — see Agent access to meeting recordings.