Platform/Ekur/Nusku
Ekur · one of four capabilities

Nusku

fleet control

Declare centrally, reconcile locally, report by exception. The control plane never touches the data.

What it is

Nusku is Ekur’s fleet-control capability. Client operators declare what runs where, from one node to thousands. The control model is simple: declare centrally, reconcile locally, report by exception. A node that is dark for a day converges back to the declared state when connectivity returns.

Control messages carry no mission data. Nusku distributes; Kittu admits, so a compromised control component can at worst deliver an artifact the node refuses. Losing Nusku stops future changes. Each node keeps running its declared state. Enkidu publishes to a client’s Nusku and has no presence inside it.

Publisher-to-fleet topology: Enkidu publishes signed bundles to the client’s Nusku Core; from the Core, hubs and nodes reconcile over the local mesh; the whole client implementation is client-operated with no publisher presence inside; a single node runs the same three components on one host.
Publisher-to-fleet topology. Source: Enkidu (S2).Scroll the figure horizontally →
Five invariants
A
Declare centrally, reconcile locally, report by exception.

Disconnection is a normal state, not a failure. A node dark for a day converges to the same state as one that never lost its link.

B
Control never carries data.

The management plane cannot become an exfiltration path — by construction, not by policy.

C
Nusku distributes; Kittu admits.

A compromised control-plane component can at worst deliver an artifact the node refuses. Movement is logistics, never trust.

D
Human authority; automation is bounded and reduce-only.

Capability widens only through a committed declaration. It narrows automatically.

E
Loss of Nusku costs the future, never the present.

Total loss of the control plane leaves every node running exactly as declared.

Three components
Agent — on every node: identity, enrolment, local reconciliation, verified install, local telemetry, bounded remediation.
Warden — at every hub: the unit of autonomy for a cluster; caches, aggregates, coordinates, recovers, and keeps doing so when cut off from the centre.
Core — at the centre: the registry, the signed desired state, the fleet twins, the console, the entitlements. One per implementation.

A single-node implementation runs all three on one host. Scale changes complexity, not function.

The eight-step operating cycle — declare, publish, reconcile, admit, apply, observe, report, remediate — with admission marked as Kittu’s step, and the loop closing only through a committed declaration.
The operating cycle. Source: Enkidu (S2).Scroll the figure horizontally →
What it administers
Fleet lifecycle Zone placement Policy application Access and entitlements Artifact distribution Patching Cryptographic posture Observation Remediation and automation
The publisher boundary

Enkidu publishes signed artifacts to a client’s Core and nothing more. Enkidu has no component, credential or channel inside the client’s implementation; administration inside it — and every model, agent, tool and policy the client authors for itself — is the client’s alone.

How it fits with the other three

Anshar defines zones; Nusku places resources into them.

Kittu admits; Nusku distributes only what the registry has activated.

Shedu holds every key; Nusku carries the declaration of which algorithms and protocols apply, and never key material.

Nusku is specified and is being built to its specification. No availability, customer or performance claims.

0

mission-data bytes a Nusku channel can carry. Control schemas structurally cannot express mission content. Source: Enkidu (S2).

Further reading
Nusku — Fleet Control S2 · ENK-TWP-2026-006Available on request
Related