A Keep in the server room

The server room is already there. The organization owns a rack, a segment, an identity system, a ticketing system, a log store, and a team whose job is to keep them running. Czar Keep is built to go into that room as one more thing the team is responsible for, and to ask less of the team than most of what is already in it.

The rack and the segment

Czar Keep arrives as one sealed chassis holding four or eight sovereign AI engines and running the Substrate. The infrastructure team chooses the rack and the network segment. The Keep can sit on a segment with no route out at all, or on one the organization already secures. Nothing about it requires a new room, a new zone, or a new way of working. It is placed where the organization's other sovereign systems are placed, under the same physical controls, and it is one thing to watch rather than a stack of things to keep consistent.

Routes declared at commissioning

Commissioning is where the organization decides what the Czar Keep may talk to. Each connection is declared one by one: the identity system, so that people and applications are who they say they are; the record systems and the data stores that Retrieval will index; ticketing, so agents can read and propose; the log store; source control, where the organization keeps its own code. Local Bridge permits those routes and denies everything else.

There is no default allow list and no route the organization did not declare, and the declaration itself is on the record. When the organization later wants another route, that is a change the team makes deliberately, not a setting an application discovers on its own. The posture is the same as on any Czar; the Keep brings it into a room where it is already the expectation.

Standing alone or joining a pool

A Czar Keep in the server room can stand alone, serving the organization's applications, retrieval, and agents by itself. It can also join a Czar Grid pool. Where the organization already has Czars, on a laboratory floor, in a program office, or beside a line of machines, the Keep joins them as one pool with the same services, the same agents, and the same record, and the density in the room serves the floor's machines as well as the building's people.

The Keep is a Grid in one chassis, so joining a Grid is not a migration. The infrastructure team declares the route, and the pool grows. Nothing is rewritten, and nothing that was on the record leaves it.

Signed updates, by media or by the network

Updates to a Czar Keep arrive as signed bundles. On an air-gapped segment, the team carries the bundle in on media, the Keep verifies the signature before applying it, and Watchman verifies what is running afterward. On a connected segment, the bundle travels over the organization's own network to the Keep, on a route the team declared. In neither case does the Keep reach out to fetch anything, and in neither case does an update begin from outside. The team decides when a bundle is applied, and the record shows what was applied and by whom.

Support the customer initiates

Nothing begins from the outside. When the infrastructure team needs help with its Czar Keep, it opens a ticket in the customer portal. If a remote session is needed, the customer initiates it, for a window that expires. On a sovereign segment the organization attests first, confirming that the data disks are detached, the network is isolated, the console is supervised, and the person requesting has the authority to do so, before a session is approved. No one outside the organization has a standing path into the Keep. The organization's own record shows every session that was opened and every one that was closed.

What the team is responsible for, and what it is not

The infrastructure team is responsible for a Czar Keep in the way it is responsible for everything else in the room: power, cooling, the rack, the segment, the identity system, and physical access. It is responsible for declaring routes, for applying signed bundles when it chooses, and for holding the record. It owns the data, the index, the adapters, and the agents the organization authors in Agency Studio.

It is not responsible for building the platform. The seven Substrate services, Inference, Retrieval, Memory, Adaptation, Evaluation, Watchman, and Local Bridge, are what the Keep runs, and they are sustained under the subscription. The team does not assemble a serving stack, patch model runtimes, or integrate several tools and keep them in step. It is not responsible for a vendor's cloud, because there is none in the path. And it is not responsible for approving what the agents do; the people who own the work are. Agents on a Czar Keep prepare and propose; a named person approves, and every step is on the record the organization holds. Where a Keep serves a floor, safety stays on the robot.

A server room is a place an organization already trusts. The Keep is built to earn the same trust in the same way: by being one sealed thing in a rack, doing what the team declared it may do, and nothing else.