Recording event JSON file

A time-stamped log of call events (tracks, active speaker, chat, transcript) written next to a cloud, cloud-audio-only, or raw-tracks recording. Live streams never write it.

Enable

Pass dataOutputs: ["event-json"] to startRecording() or to POST /rooms/:name/recordings/start (camelCase in REST too). Other values: transcript-webvtt, chat-webvtt. The file only captures events; it does not start transcription or chat.

Location and timing

Key: <s3key>.event.json, in the recording’s bucket (Daily’s, or your own S3 / OCI bucket). s3key is in GET /recordings/:id and the recording.ready-to-download webhook (s3_key).
Download link: data_outputs["event-json"] in GET /recordings/:id/access-link. GET /recordings/:id lists data_outputs.
Uploaded when the recording stops. If the call moves to another server mid-recording, a partial copy is uploaded at that point, so the file existing does not mean the log is complete. Not uploaded if it would exceed 50 MB.
The recording.ready-to-download webhook does not list this file and does not wait for it. If the file is missing when the webhook arrives, retry.

Format

{
  "format_id": "daily-event-json",
  "format_version": "2026-04-30",
  "meeting_session_id": "<meeting session id>",
  "recording": { "id": "<recording id>", "type": "cloud-recording | raw-tracks", "instance_id": "<instance id>" },
  "events": [ { "ts": 1760500001.204, "type": "<event type>", "participant_id": "<session id>", "data": {} } ]
}

recording.type is cloud-recording for cloud and cloud-audio-only. Each event has ts (Unix seconds, float, clock time), type, optional participant_id (session ID), optional data (per type). Skip unknown types.

Event types

`type`	Recording types	When	`data`
`recording-media-started`	cloud	once, media start; always index 0 of `events`; no `participant_id`	`meeting_session_id`, `recording_id`, `instance_id`, `recording_type`
`recording-media-started`	raw-tracks	once per track file, when Daily starts writing it	`trackSessionNum`, `uri` (`s3://<region>/<bucket>/<key>`), `contentType`, `mediaStartTime` (Unix seconds), `user_name`, `user_id`
`recording-media-finished`	raw-tracks	track file reached final upload state	`trackSessionNum`, `uploadStatus` (`complete` \| `failed` \| `no-media`)
`track-added`	all	track joined the recording (includes tracks present at start)	`trackSessionNum`, `trackId`, `kind` (`audio` \| `video`), `trackType` (string; common values `cam-audio`, `cam-video`, `screen-audio`, `screen-video`; others such as `rmpAudio` / `rmpVideo` for remote media players can appear), `displayName`, `paused`
`track-removed`	all	track left the recording	`trackSessionNum`, `trackId`, `kind`, `trackType`
`track-paused` / `track-resumed`	all	participant muted / unmuted the track	`trackSessionNum`
`active-speaker`	all	active speaker changed to a different participant; an entry at recording start is not guaranteed	none; speaker is `participant_id`
`transcript`	all, only while live transcription runs	transcript message arrived, including non-final results	the transcript message: `session_id`, `user_id`, `user_name`, `trackType`, `text`, `timestamp` (ISO 8601), `speech_timestamp`, `duration_seconds`, `transcription_delay`, `instanceId`, `is_final`; optional `rawResponse`, `speaker`, `speech_final`
`chat-msg`	all	app message (broadcast or targeted) with `event` of `chat-msg` or `message-reaction` (Prebuilt chat sends these)	the message as sent; check `data.event` to tell reactions from messages

trackSessionNum is unique per track within the recording and links all track-* and recording-media-* entries for that track. A republished track gets a new trackSessionNum.

Rules

events is not sorted by ts: the cloud recording-media-started entry is forced to index 0. Sort if you need time order.
Offset into media, cloud: event.ts - start.ts, where start is the recording-media-started entry. Raw-tracks: event.ts - start.data.mediaStartTime, where start is the recording-media-started entry with the same trackSessionNum (its ts is when the notification was logged, not the media start). Negative offsets (tracks present before media start) mean “at the start”.
Raw-tracks mediaStartTime is when writing started, not the first media sample. Good for ordering, not sample-exact sync.
active-speaker is written only on change. Silence is never written.
Stopping the recording does not write track-removed; a track without one ran to the end.
Raw-tracks tracks that are paused when they would be added get track-added on first resume.

Architecture & Monitoring

Troubleshooting

Privacy and Security

Scaling Calls

Features

Recording event JSON file

Enable

Location and timing

Format

Event types

Rules

Enable

Location and timing

Format

Event types

Rules

Related