RosettaOps™ Idle Resources
Waste detection and cleanup
Nobody is using it.
You are still paying for it.
Idle machines, disks nobody reattached, load balancers with nothing behind them. We find twenty eight kinds of it across your accounts, show you the list, and switch off whatever you approve.
What gets found
Far more than forgotten machines
Most waste tools stop at compute. That is where everyone has already looked, so it is where the easy savings have already been taken. The surprises are usually elsewhere, and the same scan finds them on every cloud you have connected.
Compute
Machines stopped and forgotten, idle containers, Kubernetes clusters with nothing scheduled, quiet big data clusters.
Storage
Disks nobody reattached, snapshots older than anyone remembers, and disks on the wrong storage class or paying for performance they never use.
Network
Load balancers with nothing behind them, idle gateways, reserved addresses attached to nothing, unused private endpoints.
Data
Quiet databases and caches, warehouses nobody queries, idle search and streaming, document and graph stores left running.
AI and ML
Model endpoints kept warm for traffic that stopped months ago. Idle GPU is the most expensive idle thing on the bill.
Leftovers
Alarms watching resources that no longer exist, security rules protecting nothing, gateways nobody calls. Small on their own, and they never leave.
It reports before it touches anything
This is the part that decides whether a cleanup tool gets switched on or quietly disabled after one bad afternoon.
Reporting is the default
A scan tells you what it found and what it would cost to keep. Nothing is removed until you turn that on, per kind of resource.
Deletion is not just off, it is not allowed
In reporting mode the credentials the scan uses are read-only. A mistake in our code cannot delete anything, because nothing it holds has permission to.
Your thresholds, your snapshots
You decide how many days idle counts as idle, separately for each kind of resource, and whether a snapshot is taken before anything goes.
Production accounts can be marked untouchable, whatever any rule says.
Then stop it coming back
Cleaning up once is a good afternoon. The waste returns the following week unless something is watching.
Autostop on idle
A machine nobody is using stops on its own. Data stays where it is, so work carries on when it starts again. The developer who went to a meeting does not pay for the afternoon.
Scheduled shutdown
Development and test environments do not need to run at three in the morning. Put them on a schedule and stop paying for the nights and weekends nobody works.
Spot instances
Spot prices,
without losing the work
Spot instances are spare cloud capacity at a fraction of the on-demand price. The catch is that the provider can take them back at about two minutes notice, which is why most teams only trust them with work they are willing to lose. Two minutes is enough, if something is ready for it.
1. The warning arrives
A rule in your own account is listening for the interruption notice, so nothing depends on us noticing in time.
2. The work is saved
Machines that can hibernate write memory to disk and keep it. Machines that cannot are captured as an image before the provider takes them, so nothing has to be rebuilt.
3. It picks up again
Hibernated machines resume where they stopped. The rest relaunch from the image they were captured into, once capacity returns.
That changes what spot is good for. A long training run or an overnight batch job can take the discount, because an interruption costs minutes rather than the whole run. You choose which workloads are eligible, and everything else keeps on-demand capacity.
Common questions
How do I find idle resources across my cloud accounts?
RosettaOps scans every connected account for twenty eight kinds of idle and orphaned resource, from stopped machines and unattached disks to idle databases, load balancers, Kubernetes clusters and AI endpoints. It reports what it finds before anything is touched.
Will it delete something I still need?
Not unless you ask it to. Scanning is report-only by default, and in that mode the credentials it uses are read-only, so deletion is not just switched off, it is not permitted. You choose which kinds of resource it may act on, and it can take a snapshot first.
Can I shut machines down automatically when nobody is using them?
Yes. Machines can stop on low usage or on a schedule, so development and test environments are not paying for nights and weekends. Data stays in place, so work resumes when the machine starts again.
Can I use spot instances without losing work when they are reclaimed?
Yes. A rule in your own cloud account listens for the interruption warning. Machines that can hibernate write memory to disk and keep it, so work resumes where it stopped. Machines that cannot hibernate are captured as a machine image before the provider terminates them, so they can be relaunched rather than rebuilt.
How many days before a resource counts as idle?
You set the threshold, and you set it separately for each kind of resource. A stopped machine and an old snapshot rarely deserve the same patience.