Every channel, rundown and account lives in one file. Snapshots are taken automatically, and each is opened and read before it is kept — a snapshot that cannot be restored is worse than none, because it satisfies the retention count and displaces a good one.

Every original is replicated to each configured tier. If the primary goes away, reads and ingest fail over automatically; when it returns, what landed elsewhere is copied back and the system returns to it. The quiet state is the whole point.

The server posts a signed JSON event to each webhook when a channel goes down or comes back, an output (SRT, UDP, SDI) stops or recovers, a file cannot be played, dead air airs, or a channel logs an error — retried until it is delivered.

Who started, stopped, deleted or suspended what — the first question when a channel goes off air. Kept ninety days. The address is the machine that connected, not one it claimed to be.

Scans the uploads/ directory and registers any MP4 files not already in the database. Use this after data loss to restore file records without re-uploading. Files will be probed for duration and resolution automatically.

The process writes its own vitals every 30 s to playout-flight.json — a bounded one-hour ring. After an unclean shutdown, the previous run's final state is read out loudly in the log and served here until the ring refills. A climbing RSS names a leak; a flat line before a crash points at power, kernel or driver.

Download-template accidents like NameName.mp4: preview lists every candidate with its before/after (long stems only — real names like “gogo” are never touched); repair applies and audits. Display already un-doubles; this cleans the stored data itself.

Share a code with someone to let them create an account. The code sets how many channels that user may create.