PureView capabilities.

Sweeps the ranges you give it by opening ordinary TCP connections, so it needs no special privileges on the host. A device that refuses a connection is recorded as present rather than missed, and the ports it tries include the ones network equipment actually answers on rather than only the obvious two.

Matches hardware addresses against a vendor database, reads identifying values over SNMP where a device offers them, and resolves a model from what it finds. It declines to guess when the answer is ambiguous.

Asks switches and routers what they can see next to them, and builds the adjacency from their own neighbour tables rather than from a drawing somebody maintained by hand.

Turns interface counters into rates, and knows the difference between a counter that reset because the device rebooted and traffic that genuinely went backwards. It refuses to average across a gap in collection rather than drawing a smooth line through a period it knows nothing about.

Watches upstream links and distinguishes the case everybody misses: the gateway answers, the internet does not. Where traffic is still moving it trusts the traffic over the ping.

Weights each measurement by the time until the next one rather than counting samples, and does not charge its own collection gaps to your provider.

Accepts log messages in both standard formats, including the awkward parts: a December timestamp arriving in January, and the byte order mark the specification permits. The listener stays off until you turn it on, so an upgrade never opens a port you did not ask for.

Produces configuration for each supported platform in that platform own syntax, and checks it before it goes anywhere: a policy set with no final deny is flagged, a rule referring to something that does not exist is refused.

Takes a structured snapshot of a device configuration, on a schedule and when something changes, and keeps what it does not recognise rather than discarding it.

A multi-step deployment that fails part way is rolled back to the state it started from, including restoring the settings it replaced.

Describes an intended network independently of vendor, checks it for the mistakes that matter, and refuses to produce a deployment plan while a critical problem remains unresolved.

Tracks which addresses are allocated, suggests the next free one inside the range you meant, and reconciles the plan against what is actually answering on the network.

Records sites, rooms and racks down to the unit a device occupies, so each device has a location as well as an address.

Tracks equipment through its life, including the relationships between items: what feeds what, what is hosted on what, and what depends on what.

Records how power reaches each item, so the question of what else goes dark has an answer recorded in advance.

Registers servers, reads their state, and runs authorised operations against them: services, processes, packages, scheduled jobs, networking and power.

Covers the machines on those hosts through their own interfaces: lifecycle, configuration, snapshots, storage and networking.

A host reports in using a token issued for it alone. The token is shown once and stored only as a hash, and it carries the organisation, so a host cannot report into somebody else account.

A threshold has to stay crossed for as long as you specified before an alert opens, and has to recover past it before the alert clears. Nothing opens during an approved maintenance window, and stale readings are treated as stale rather than as a real breach.

When a failure takes out everything behind it, the alerts behind it are marked as consequences and pointed at the thing that actually broke. The inbox shows you the cause.

Records incidents with a timeline that cannot be quietly rewritten, and holds the serious ones to their process: a major incident needs somebody named as running it, and needs evidence before it can be closed.

Administrator, operator and viewer. What you may do is decided when your request arrives, by one shared check, whatever route it took. A viewer is sent away from a page they may not have rather than shown a version of it that fails on use.

Beyond a role, a person can be given a site, a room, a rack, a group or a single device, and sees only that. A grant narrows; it never widens across an organisation boundary.

The credentials the software uses to reach your equipment are encrypted with a key belonging to that installation, and are never returned through the interface. One organisation credentials are not reachable from another.

Every change is recorded with who made it, what it affected, what it looked like before and after, and where the request came from. Recording a change never causes the change itself to fail, and no secret is copied into the record.

Installs on an ordinary Linux server inside the network it manages, from a bundle that carries everything above the operating system. A site needs no registry and no internet on installation day.

Some things a browser cannot do. The desktop application reaches a console cable and scans the network the operator is actually sitting on, and routes that work to the local machine rather than through the server.

A read-only console handed to people who should only watch. What it may not do is absent from the build rather than hidden in the interface, so an administrator running it still gets no console cable.

The bundle carries the application, its database and its front door as images, so an isolated site installs from media rather than from a network it does not have.

Applications for phones and tablets, for the people who need to see status away from a desk.

Each installation will carry a licence identifying its edition and entitlements, which it can check without a continuous connection.

Installations will ask whether an update applies to them, and how promptly updates arrive will be yours to choose, including holding them until somebody approves the change locally.

One place to see your deployments, their versions, your subscription and what you are entitled to.

Raise a request against a specific deployment, and where a problem needs us to look directly, authorise a session that expires by itself.

Guidance that states which versions it applies to, so a procedure never describes a screen your installation does not have.