segment

Record HLS (.ts segments + playlist) to S3, GCS, or Azure.

Example

{
  "name": "hls_rec",
  "type": "segment",
  "inputs": ["mix_video", "mix_audio"],
  "config": {
    "storageType": "s3",
    "storageConfig": {
      "bucket": "my-recordings",
      "region": "us-east-1",
      "accessKeyId": "<key>",
      "secretAccessKey": "<secret>",
      "prefix": "shows/{date}/",
      "filename": "composite"
    },
    "segmentDurationSec": 6,
    "maxPlaylistEntries": 10,
    "caption": {
      "showSpeaker": true,
      "textSplit": { "enabled": true, "length": 42, "lines": 2 }
    },
    "encoding": { "videoCodec": "h264" }
  }
}

Config

Field	Type	Default	Description
storageType	string	required	s3 | gcp | azure.
storageConfig	object	required	Provider credentials and path. See Storage config.
format	string	"hls"	Output format (hls).
segmentDurationSec	int	—	Segment length in seconds (0–60).
maxPlaylistEntries	int	—	Live sliding-window playlist size (0–1000). VOD / finalize includes all segments regardless.
encoding	object	—	See shared encoding. H.264 + AAC recommended.
caption	object	—	HLS WebVTT caption options. See caption.

prefix and filename support path templates. Filename has no extension; the engine appends by format (default basename index → index.m3u8).

caption

Omit caption to use defaults: speaker names are shown and one WebVTT cue is written per ASR final. caption.textSplit splits long captions only for this HLS subtitle rendition; other downstream consumers still receive the original avflow.asrText event.

Field	Type	Default	Description
showSpeaker	bool	true	Include speaker name in WebVTT cues when ASR is wired.
textSplit.enabled	bool	false	Turn on HLS caption display splitting.
textSplit.length	int	42	Max display columns per line (Latin=1, CJK/fullwidth/emoji=2). When enabled, 0 / omitted → 42. Cap 200.
textSplit.lines	int	2	Max lines per WebVTT cue when enabled; 0 / omitted → 2. Cap 10.

Wiring

Must connect to a single mixed stream — use video_mixer + audio_mixer for RTC rooms.

Wire asr into inputs for subtitle renditions. Each definite ASR result is written as a WebVTT cue as it arrives; caption.textSplit can split long cues for HLS display. The VTT (and subtitle playlist) are uploaded on finalize.

Storage config

s3

Works with AWS S3, Cloudflare R2, MinIO, and other S3-compatible APIs.

Field	Type	Description
bucket	string	Bucket name (required).
region	string	AWS region (optional for some S3-compatible endpoints).
accessKeyId	string	Access key id (required).
secretAccessKey	string	Secret access key (required).
sessionToken	string	Optional STS session token.
endpoint	string	Custom S3 API endpoint (R2, MinIO, …).
forcePathStyle	bool	Force path-style addressing (often needed for MinIO / custom endpoints).
prefix	string	Key prefix (may include templates).
filename	string	Playlist basename without extension (default index).

gcp

Field	Type	Description
bucket	string	GCS bucket (required).
credentialsJson	object	Service-account JSON object (required, inline).
prefix	string	Object prefix (templates allowed).
filename	string	Playlist basename (default index).

azure

Field	Type	Description
container	string	Blob container (required).
accountName	string	Storage account name (required).
accountKey	string	Account key (use this or sasToken).
sasToken	string	SAS token (use this or accountKey).
prefix	string	Blob prefix (templates allowed).
filename	string	Playlist basename (default index).

Path templates

prefix and filename may contain {var} placeholders:

Variable	Meaning
jobId	Job id
jobName	Job name
roomName	Room name (from upstream RTC source when available)
date	YYYY-MM-DD
time	HH-MM-SS
timestamp	Unix milliseconds
identity	Participant identity (when applicable)
trackSource	Track source
trackSid	Track SID
trackName	Track name
sourceName	Source component name (e.g. room_src)

Known variables with no value expand to empty; unknown {…} tokens are left unchanged.

Pricing

Encoding tier by encoded resolution and codec, plus successful upload egress. Image sinks keep their separate per-image pricing and do not use this A/V egress model. Sink pricing.