This guide explains the full Cassini workflow — from joining a room to opening a replay in the browser — and what happens at each stage.

The pipeline

Cassini’s record command is a three-stage pipeline. Each stage produces an artifact that the next stage consumes.

cassini record
  └── Stage 1: Capture   →  .run bundle
  └── Stage 2: Build     →  .meeting artifact
  └── Stage 3: Publish   →  .opus file

You can run each stage manually if you need more control. See Advanced usage below.

Stage 1 — Capture

Cassini joins the Nextcloud Talk room as a silent participant and records the audio stream. It writes everything into a .run bundle: raw audio, participant events, and timing metadata.

The capture stage ends when the call ends or when you interrupt Cassini with Ctrl+C.

Stage 2 — Build

The build stage processes the .run bundle. It runs the audio through:

• Transcription — a local speech-to-text model produces a time-aligned transcript
• Diarisation — the transcript is annotated with speaker turns where possible
• Packaging — the transcript and audio are combined into a .meeting artifact

This stage is the most time-consuming. On a modern machine, expect roughly 0.5× real-time — a one-hour meeting takes around 30 minutes to process.

Stage 3 — Publish

Publish takes a .meeting artifact and writes the final .opus file. This step is fast — it is mostly a repackaging operation with no additional processing.

Resumability

Cassini stores intermediate state in a .cassini-work/ directory next to your --out file:

My Meetings/
  2026-07-01 Team Sync.opus        ← final output
  .cassini-work/
    2026-07-01 Team Sync.run       ← stage 1 output (kept)
    2026-07-01 Team Sync.meeting   ← stage 2 output (kept)

If the command fails at stage 2, rerunning it skips stage 1 (capture already done) and retries from stage 2. This means you never lose a recording to a transcription failure.

The viewer

The Cassini viewer is a static web app bundled separately. Use cassini serve to start a local server pointed at a directory of .opus files:

./bin/cassini serve "./My Meetings"

The viewer gives you:

• Timeline scrubbing — click any point in the waveform to jump to it
• Transcript navigation — click any line in the transcript to seek to that moment
• Full-text search — search across the transcript to find when something was said
• Speaker filtering — filter the transcript by participant

Advanced usage

You can run each pipeline stage explicitly if you need to inspect intermediate artifacts or retry a specific step:

# Stage 1: capture only
./bin/cassini record --call "$CALL_URL" --out ./runs/meeting.run

# Stage 2: build from a .run bundle
./bin/cassini build ./runs/meeting.run --out ./meetings/meeting.meeting

# Stage 3: publish to .opus
./bin/cassini publish ./meetings --out ./site

# Inspect any artifact
./bin/cassini inspect ./runs/meeting.run
./bin/cassini inspect ./meetings/meeting.meeting

Nextcloud app install

Cassini can also run as a Nextcloud ExApp — registered via AppAPI, it records meetings server-side without requiring a separate machine. See the ExApp install guide for the full setup.