Home/Platform/Devices & Discovery
The estate
Know the environment. Choose where to act.
An agent on a machine is not a nice-to-have here — it is the thing that makes a tunnel, a created credential and a target-side confirmation possible at all. Machines without one can still be found, still be worked on, and still be reported. They just cannot do those three things, and we would rather draw you that boundary than sell around it.
The boundary
No agent means no tunnel and no credential creation.
This is the part most platforms blur. An unmanaged machine is not a managed machine with fewer features switched on — there is nothing on it to dial out, and nothing on it to create an account with. Everything else follows from those two facts.
How a machine gets here
Found, then enrolled, then targetable.
Discovery tells you what exists. Enrollment turns one of those things into a machine you can prove work on. They are two steps, and a good onboarding walks the gap between them deliberately.
-
Discovery finds what is out there
A scan run from a machine that already has an agent, plus reads of the directory you already run, build a picture of the estate: hosts, operating systems, who is privileged, what looks unmanaged. It is the first honest inventory most clients have had in years.
-
Enrollment makes a machine answerable
The agent generates its own key pair, the private half never leaves the machine, and the device is bound to exactly one tenant. From that moment it can dial the tunnel, run a signed job, and answer an independent check about its own state.
-
Targeting picks it for an operation
An operation names a target and the platform resolves what that means: is there an agent, is there a tunnel, which credential reference applies, what privilege has to be leased. The model names the operation; the code derives the least privilege that operation needs.
- We do not manage network appliances — firewalls, switches, security boxes. Reaching those through a local gateway object is something we intend to build, not something you can buy today.
- We do not claim macOS or Linux agents. Windows is what ships.
What a row actually carries
Every field on a device row can name its source.
An inventory that cannot tell you where a value came from is a list of assertions. Colour-code the same row by provenance and it becomes something you can reason about — especially the fields that come from a scan rather than from the machine.
Installed software is collected from every enrolled machine and kept. What is still being built is the surface on top of it — a searchable software view, and a planner that reasons over what is installed before it proposes a fix. Today it is a collected, retained inventory, and we would rather say that than call it software asset management.
Targeting
An operation names a machine. The platform works out what that costs.
You do not hand a script a hostname here. The target is resolved into a set of facts, and those facts decide what privilege gets leased and whether the work can be proven at all.
- The model names the operation. Code derives the least privilege it needs — the model never gets to ask for more.
- A target with no agent is not silently downgraded. The plan says it is brokered, and the record says the same.
- One machine, one remote session. If someone already holds it, you find out before you connect.
- Bulk work fans out per target. Each machine gets its own isolated record, so one failure in fifty does not smear across the batch.
What people ask before rolling an agent out.
Can you work on a machine with no agent at all?
Yes, within a real limit. The work is brokered from a managed machine nearby and the identity has to come from your vault or from the client at the moment of use. What you cannot get is a created credential, a direct tunnel, or a confirmation from the target's own agent — because there is no agent on it to give you one. The record reflects that, rather than dressing it up as the same thing.
What does the agent actually do when it is idle?
It holds an outbound connection, reports its own state, and waits. It does not accept inbound connections and there is no port to open for it. Work only reaches it as a signed job that arrived down a tunnel the machine itself dialled. See the trust center.
What happens if a machine moves to another client?
It enrolls into the new tenant and nothing comes with it. Records stay with the tenant that owned the work when it happened, because evidence is a point-in-time statement about a device, not a property the device carries. See Roles & Multi-Tenancy.
Do you replace our RMM?
Not as a blanket claim, and we will not make one. Name the specific capability and we will tell you straight whether it ships, whether it is narrower than you think, or whether it is not there. What we do claim is the operation loop: leased privilege, model-blind execution, a target-side confirmation and a record. How we compare →
How do you find machines nobody told you about?
A scan brokered from a machine that already has an agent, plus reads of the directory you already run. The gap between "in the directory" and "actually enrolled" is usually the most uncomfortable and most useful number in an onboarding.
Point at your hardest machine
Ask what we can and cannot prove on it.
Bring the server nobody wants to touch. We will tell you which half of this page applies to it before you spend a minute on a rollout.