Broadcast

Roadmap surface for encoder paths, HLS shards, and operator desks. Compare the multi-app layout in fornevercollective/grok-cli (notes, terminal, ML, sync).

Built from Erika dataset metadata for browsing, filtering, and static profile lookup.

Live broadcast capabilities

Train stays a static site. This tab documents where a future broadcast stack lives: same patterns as mature operator repos that bundle notebook UI, terminals, collaboration, and ML hooks (for example the structure described in fornevercollective/grok-cli — React/TS notes front-end, Node backends, deploy scripts).

Set PUBLIC_BROADCAST_PIPELINE_URL at build time for a button to your yt-dlp / FFmpeg / HLS viewer (or any live desk origin).

Ingest & transcode

RTMP/SRT listen, direct pulls, YouTube via yt-dlp — own a small always-on service; Train only links to it.

Telemetry & shards

WebSocket segment ledger, FFmpeg progress, off-ramp webhooks — mirror the “instrument everything” path from your local pipeline viewer.

Operator notebook

Jupyter-like cells, markdown runbooks, and AI assist panels — reference implementation families in grok-cli (grok-notes-ts, grok-notes-backend).

Terminal & sync

Embedded shell for health checks, ffmpeg probes, and git deploy hooks; collaboration / offline sync when you promote this from placeholder to product.

Next integration step

  1. Deploy pipeline or Game on a real origin with HTTPS.
  2. Set repository variables for PUBLIC_GAME_CONSOLE_URL and optionally PUBLIC_BROADCAST_PIPELINE_URL (plus PUBLIC_BROADCAST_EMBED_HOSTS to allowlist iframe hosts in production).
  3. Use Sports for the public shell and Broadcast for operator-facing links, embed, and docs.