Riopex reads the monitoring system your team already runs. This is the reasoning, published because every network team asks first.
There are two ways to know how much traffic a circuit carried. Collect it yourself, or read it from the system that already collects it. The first sounds more capable, and in a multi-country group it is the longest-lead part of the whole project.
Riopex reads. So the honest place to start is with exactly what that reading asks of you.
One read-level API key to a system your team already operates. The key is stored as an opaque reference into a secrets manager and resolved only when a job runs, so the secret itself stays in the vault.
The ask is small enough to be granted by the people who already own monitoring, which is the real unlock. One team can make that decision without convening four others, and the work then moves to the part that is genuinely hard.
Binding an interface to a service is that part, and it is a reconciliation problem rather than a collection one. Some platforms expose no stable interface index, so the practical key is the device host plus the sensor name, and matching on that automatically is how utilisation ends up attached to the wrong circuit.
So Riopex auto-binds only on an exact identifier match and puts everything else in a queue with ranked candidates. That is where the engineering went: into being right about which circuit a number belongs to. Compare that with what collecting the numbers ourselves would have cost.
Deploying polling infrastructure inside a group's networks means a security review per country, a penetration test, a data-processing agreement, a firewall change, a device credential, and an owner for each of those in each operating company. Six to twelve months per customer is a realistic estimate, and it produces numbers the customer already had.
It also creates a permanent liability. Every device credential we held would be a credential that can be lost, and every agent we shipped would be a component on someone else's network that we are responsible for patching.
Reading has its own price, and it is paid in what you inherit.
You inherit the other system's blind spots. An unmonitored interface stays invisible to us. A platform with ninety-day retention gives us ninety days. Where its coverage is thin, ours is thin, and it is reported as thin rather than smoothed.
One platform we surveyed derives its figures from session logs rather than interface counters, so its coverage is capped by policy and supports a diagnostic rather than a waste finding. That constraint is written into the connector rather than into a footnote.
Which platforms we read, what each connector takes, and the status of each one are on the monitoring systems page.
