Navi · broadcast agent
The first agent built for live broadcast operations
Navi reads every Sense on every tap, folds related faults into one incident with a cause, and puts it in front of master control in thirty seconds — with the clip, the timecode, and the fix.
- SCTE-35
- Loudness targets
- Caption checks
- EBU R128
- Fault to master control alert
- 30s
- Senses read on every tap
- 9
- Local models, your building
- On-prem
Measured on a customer feed.
Audio, video, and metadata on one timeline.
No per-token costs. No media leaves the facility.
Watch it work
One bad insert. Three alarms. One incident.
Break 4 of 6 on a live sports feed goes wrong three ways at once. On the left, what your wall shows tonight. On the right, what Navi sends to master control.
Your wall tonight
Separate alarms. No cause, no owner.
What Navi sends
One incident. One cause. One action.
- 23:41:07.02MetadataSCTE-35 splice-out fires. Break 4 of 6 starts, ninety seconds scheduled — and the wrong insert goes to air.
- 23:41:11.40AudioShort-term loudness on the insert reaches −17.8 LKFS, 6.2 LU over the configured target.
- 23:41:12.90CaptionsCaption stream drops on the commercial feed. The configured caption presence check fails.
- 23:41:15.00NaviNavi holds. A few frames over target is a transient, not a violation — raising it now would be a false alarm.
- 23:41:29.60NaviThe condition has held for eighteen seconds. Navi ties the loudness jump and the caption drop back to the same splice.
- 23:41:37.02Master controlMaster control is alerted once, thirty seconds after the insert went to air, with the cause and the swap to run.
Replay of a representative incident; the loudness and caption values shown are typical, not a customer log. The thirty-second fault-to-alert figure is measured on a customer feed.
Why thirty seconds
Fast enough to fix, slow enough to be right
Loudness is measured over a window — a few frames over target is a transient, not a violation. Navi waits for the condition to hold, correlates it with everything else on the tap, then alerts once. One incident beats thirty alarms nobody trusts.
It does not alarm on a transient
A brief excursion over the configured loudness target is not a compliance failure. Navi confirms the condition is sustained before it says anything, so the alerts your operators get are the ones worth standing up for.
One alert, not three
The loudness jump, the caption drop, and the splice are one fault with three symptoms. Navi sends the fault. Nobody spends the first ten minutes agreeing on what happened.
The alert lands inside the break
Thirty seconds into a ninety-second break leaves a full minute to swap the insert. That is the difference between a fix and a filing.
What it buys you
The break still goes out in spec
A ninety-second break that fails at second four is recoverable at second thirty-seven and unrecoverable at second ninety-one. Navi buys your operator the minute that decides which one it is.
Every tap, every hour
Coverage does not thin out overnight, at the end of a long shift, or on the channels nobody has a monitor pointed at.
The same standard every time
SCTE-35 timing, configured loudness and R128 checks, caption presence and timing, and your house policy per channel, genre, and daypart — applied identically on feed 1 and feed 40.
A trail for the morning after
Every detection and every action, human or agent, lands on one timeline with program timecode. Post-air review starts from the record instead of from memory.
What Navi does
Correlate. Decide. Route.
One incident, not thirty alarms
Navi reads every Sense on every tap and merges related detections into a single story with a cause. Alarm fatigue is a design failure — correlation is the fix.
Fluent in your standards
SCTE-35 timing, configured loudness and R128 checks, caption presence and timing — plus your house policies per channel, genre, and daypart. Good means whatever your ops manual says it means.
Acts on your terms, defers to your operators
Alerts ship with the clip, the timecode, and the reasoning. Actions are configurable per workflow — alert, annotate, ticket, webhook — and your operators stay authoritative on the air chain.
Fluent in the language of broadcast
| Standard | Covers | What Sense monitors |
|---|---|---|
| SCTE-35 | Ad insertion cue messaging | Validates splice points, cue durations, and timing accuracy against SCTE-35 markers in real time. |
| Configured loudness | Program and commercial loudness | Monitors audio against the loudness targets you configure per channel, genre, and daypart. |
| Caption checks | Caption presence and timing | Checks caption presence, synchronization, and completeness against the caption rules you configure. |
| EBU R128 | Loudness normalization (EU / streaming) | European loudness standard. Monitors integrated loudness, true peak, and loudness range for broadcast and streaming playout. |
Division of labor
The Senses detect. Navi decides.
Each of the nine Senses is a specialist in one domain — loudness, captions, video integrity, language, safety. Navi is the generalist above them: it reads every detection in context, on program time, and turns them into calls your team can act on.
Navi, in plain terms
See Navi on your feeds
Thirty minutes, your test stream, Navi and the Senses live. We connect to a feed you provide and run the session on that signal.