The instinct of your best NOC engineer, on every site.
Multi-agent diagnosis of every fault against your own docs — cited — then a field tech in the loop before dispatch.
live scene · follows your cursor
A fault fires at 3am. The clock is already running.
Today
A national carrier runs tens of thousands of cell sites. Any one can fail at 3am — a rectifier, a transport link, a power plant.
The clock
From minute zero an SLA clock is running. The alarm storm lights up dozens of nodes and hides the one site that matters.
The shift
The senior engineers who could read a fault by instinct are retiring. That instinct doesn't scale — and it doesn't answer the phone at 3am.
The question
What if every fault got the instinct of your best engineer — grounded in your own docs, and never dispatching a crew without a human's say-so?
That is Arc: specialist agents under one Orchestrator, reasoning on Vultr and answering to a human in the field.
The site
What Arc watches: one cell-site equipment shelter.
Racks, transport links, the −48 V DC power plant, the patch panel — down to the operator's phone. Every node feeds the live signal stream, and when a single rectifier drifts the whole site is on the clock.
One Orchestrator. Six agents. One human veto.
Correlationfuses the alarm storm with the network topology to isolate the one site that matters — and its equipment class.
One fault, one responder — scored, not broadcast.
Diagnosed fault signature
score = 0.5·competence + 0.5·difficulty-fit, in-zone preferred.
Scoring the candidates
scanningThe one responder
A. Rossi
Senior energy specialist
Only the best-fit responder is paged — not a broadcast.
A. RossiCompetence 1.00 (power, rectifier), senior level fits a complex fault, and in the site's zone → the one notified.
From a live fault to a cited action report.
The live event log, the network topology, and the agent's reasoning — from the first anomaly at PAR-021-NORD to the confirmed dispatch, on one screen.
See it in action
Watch Arc run — from alarm to cited dispatch.
One incident, a few minutes: detection, grounded diagnosis, the field veto, then remediation and a cited action report.
The agent proposes. The field decides.
The field-validation loop
The moment Root-Cause lands a hypothesis, a push hits the on-call technician's phone. They test it on-site: confirm, and the pipeline advances; refuse — with a counter-measurement — and the agent pivots and re-diagnoses. The human always holds the veto.
Every cause cites your own docs — down to the page.
Agent claims
Rectifier module PN-RECT-48-2000 is the fault origin
Site draws from the −48 V DC power plant on feed B
Swap procedure requires an isolation step first
Carrier's own docs
Rectifier Maintenance Manual
cited · p. 42
PAR-021-NORD site dossier
cited · p. 7
Field Repair Procedure R-118
cited · § 3.2
Grounded, validated, fast.
0
specialist agents
under one Orchestrator
0
human veto
field-validated
0%
cited claims
grounded on your docs
<0 min
to first action
detect → dispatch
Arc · network operations
Not retrieve-then-answer. A real agent — grounded, telecom-native, on Vultr.
Arc · live event stream · site PAR-021-NORD