What Czar Keep is

An organization with a server room has a different starting point from a team with a desk. It already owns a rack, a network segment, an identity system, and the people who run them. What it wants from a sovereign AI appliance is the same thing a team wants, models, retrieval, agents, and a record that stays inside, but it wants enough of it to serve the whole building, and it wants it in the form its room already takes. Czar Keep is the Czar for the data center: four or eight sovereign AI engines in one sealed chassis, running the Substrate, standing alone or joining a Czar Grid pool.

The Czar for the data center

Czar is the appliance. One Czar is one sealed node, serving one team or a small floor on its own subscription, air-gapped or connected only to systems the customer already secures. Czar Grid is several Czars joined into one inference pool for a facility. Czar Keep takes the engines a Grid is made of and puts four or eight of them in one chassis built for a rack. It is a Grid in one chassis, not a bigger Czar. The deskside form is traded for density, and that is the whole of the trade. The Keep runs the same Substrate, holds the same posture, keeps the same record, carries the same subscription model, and takes the same signed updates as every other Czar.

A Keep can stand alone, serving an organization's people and applications from one place. It can also join a Czar Grid pool, where its engines serve alongside the Czars already in the pool, under the same services and the same record. Which of these an organization chooses depends on what it is serving, and the choice can change later without anything being rewritten.

What carries over from Czar

The Substrate services on a Czar Keep are the same seven as on any Czar: Inference, Retrieval, Memory, Adaptation, Evaluation, Watchman, and Local Bridge. Inference serves open-weight models to the organization's applications and agents. Retrieval indexes the organization's own records and answers in place, with the passage each answer came from. Memory holds what the agents and applications need to carry forward. Adaptation tunes models on the organization's own data, on the appliance, and the adapter stays inside. Evaluation scores models, prompts, and agents against the organization's own test sets before anything is promoted. Watchman verifies what is running. Local Bridge permits the routes the organization has declared and denies everything else.

Agency Studio is where the organization authors, bounds, approves, and records its agents. An agent on the Keep is written against the organization's own systems, bounded to the actions the organization permits, approved by a named person, versioned, and recorded, and any run can be replayed from the record the organization holds.

The posture carries over unchanged. The Keep can run with no route out at all. Where the organization permits connections, they are declared one by one at commissioning: identity, the record systems, ticketing, logs, source control, the data stores. Updates arrive as signed bundles, by media or by the organization's own network. Support and remote sessions are always initiated by the customer, through the customer portal; nothing begins from the outside.

What a Czar Keep serves

A Czar Keep serves the organization's internal applications, which call the models over the organization's own network with no vendor in the path. It serves the organization's agents in Agency Studio. It serves retrieval over the organization's own records, so that a question about a policy, a contract, a procedure, or a dataset is answered from the source and cited to it. It serves adaptation and evaluation on the organization's own data. When it joins a Grid, it also serves the machines on a floor, alongside the Czars that were already there.

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.

Who it is for

Czar Keep is for platform and infrastructure teams that already own a rack and a server room and are asked to serve the whole organization from it: a global capability center serving the business units it supports, a bank's internal AI group serving analysts and operations, a laboratory serving its researchers over its own data, a government or defense organization serving a program whose material cannot leave its walls. The common shape is a hundred-plus people, one machine, and a boundary that does not move. A team that needs one node runs on Czar. A technology or AI team runs on Czar Forge. A facility whose floors are the main load runs a Grid of Czars. The Keep is for the room that already has racks and the people who already run them.

The family line

Czar Keep has its place in the family line. A Czar for a team, a Grid for a facility, a Keep for the data center, a Forge for the builders, a Chassis for the vendors. The Keep is the data center entry in that line, and it is the same appliance in every way that matters: the same services, the same posture, the same record, the same subscription for the configured appliance, hardware, platform license, and sustainment in one line, fixed for the term. What is different is where it sits and how much of the organization it can serve from there.