Every cloud server in one place — measured properly, priced in full.

Hetzner Cloud, DigitalOcean, TransIP and OVHcloud. What you can do differs by provider and is declared rather than assumed: rescue mode, a web console, ISO mounting, placement groups and native delete protection are not offered everywhere, and where a cloud cannot back a control it is greyed out with the reason rather than offered and then failed. OVHcloud is the sharpest case: a VPS there is ordered rather than created from an API call, and removing one cancels a subscription with a notice period — so we send you to OVHcloud’s own panel for both, instead of offering a button that would quietly mean something different from the one beside it. The integrations page lists what works where.

No. Connect a provider and you get its own metrics — CPU, network and disk throughput, every five minutes — plus the full lifecycle, cost, backups, snapshots, volumes and networks, with nothing on the machine. The agent is optional and adds what no cloud API reports: memory, disk space, load average, uptime, the heaviest processes, what is filling the disk, the state of your services, and an alert when the server stops reporting. It is one command and takes about a minute.

Because those numbers only exist inside the machine. A cloud API sits outside the virtual server and can measure what passes through the hypervisor — CPU time, network packets, disk I/O — but it cannot see how much of your RAM is in use or how full your filesystem is, any more than your electricity meter can see what is in your fridge. That is why the agent exists, and why a rule on memory or disk stays greyed out until it is installed rather than being saved as a rule that could never fire.

The server’s own price plus everything billed alongside it: backups, each attached volume, each attached IP address and snapshot, and any manual line you add yourself, like a licence or a support contract. The IP line matters — with most providers a public IPv4 is billed separately, so a figure without it makes every server look slightly cheaper than it is. Prices are stored and shown net of VAT in the currency your provider quotes, nothing is double-counted against the organisation Spend report, and the total is labelled an estimate: your invoice is the final word.

No. The duration on a rule is an averaging window, not a stopwatch: every reading across those minutes is averaged, and the average is what the threshold is compared against. A backup that pins a core for sixty seconds does not move a fifteen-minute average past 85%. Recovery has a margin too — the average must come back about 10% under the threshold before anything is called recovered — so a value sitting on the line cannot alert and recover all night. You get two messages at most per alert, and if a spike was too short to fire you get nothing at all, not even an all-clear.

Delete destroys the machine at your provider, and the billing stops. Stop tracking leaves the machine running and billing, and only removes it from InfraNest — it disappears from the list, dashboards, cost roll-up, metric poller, alert rules and exporter, and it comes back with its history, tags and settings if you track it again. The confirmation dialog says which of the two you are about to do, including whether a delete takes effect now or at the end of a period you have already paid for.

It depends on the cloud, and InfraNest tells you which one you are on rather than guessing. Every server on a private network can reach every port every other server exposes on its private address, with no allow step in between. On Hetzner the cloud firewall cannot filter that traffic at all — its rules apply to the public interface — so who is attached is the only control. DigitalOcean’s firewalls do filter VPC traffic. Before you confirm an attach, the dialog says which of those applies, read from a capability the provider declared.