Recording
Capture session audio from LiveKit, Pipecat, or BYO ingest — playback by session.
Settings → Recording chooses which agents record session audio. Recording is on by default for new agents. Capture is adapter-specific; Parlot normalizes every source into the same playback and redaction pipeline.
Sources
| Source | How audio lands |
|---|---|
| LiveKit | Room Composite egress → Opus/OGG object (stream-copied when already conformant) |
| Pipecat | Agent uploads stereo WAV segments (~30s) via upload-grant → stitch + normalize |
| BYO ingest | Agent uploads via the same grant → same stitch/normalize/redact path |
| BYO link-only | External URI stamped for UI; rejected when tenant PII redaction is on |
Playback always uses a signed URL for the session’s current playable asset (redacted when policy requires).
Why enable recording
Recorded audio powers:
- Playback on the Timeline tab
- Evidence review next to goal and quality scores
- Alignment between what was said and what the agent did
Turns, spans, and logs still appear when capture is on. Synced call audio needs recording enabled.
Configuring agents
Settings → Recording uses the shared New agents allowlist and Known agents overrides
(Allowlist, Forced on, Forced off). Recording defaults to on for new agents. See
Agent policy allowlists for how those controls work, common
setups, and how parlotize(record=…) can override Settings.
If the Known agents list is empty, run an instrumented agent once so Parlot discovers the deployment.
Privacy
Recordings follow your PII redaction policy. Redaction beeps are close-gated (need turn PII indexes × media segment bounds against the stitched file). Retention of recording objects is also limited by your plan.
Related
- Agent policy allowlists — New agents vs Known agents
- Timeline — listen and scrub
- Organization settings — broader workspace controls
- Pipecat guide —
ParlotRecordingProcessorplacement