A Grid in one chassis

It is tempting to read Czar Keep as the large model in a range: the same appliance, scaled up. That reading gets the engineering backwards. The Keep is not a Czar made bigger. It is a Czar Grid made denser. The engines inside it are the engines a Grid is built from, and the chassis is the thing that changed, because the room it goes into is a different kind of room.

The engines a Grid is made of

A Czar Grid is several Czars joined into one inference pool for a facility. Each Czar brings its engine to the pool, and the pool serves the floor as one thing: one set of services, one set of agents, one record. A Czar Keep starts from the same engines. It places four or eight of them inside one sealed chassis, running the same Substrate, and presents them the way a Grid presents its nodes. What an organization gets from a Keep is what it would get from a Grid of that many Czars, in one enclosure instead of that many.

That is why the right description is a Grid in one chassis. The unit of capacity is still the engine. The services still run across the engines the way they run across the nodes of a pool. Nothing about the software was rewritten to make the Keep; the engines were brought together in a form that fits a rack.

Density, and what it is traded for

Czar is built to sit beside a team. It stands on a floor or in a corner, it does not need a rack, and it serves the people around it. That is the right form for a team or a small floor, and it is the wrong form for a server room, where the floor space is in racks and the people who run the room expect equipment that mounts, powers, and cools the way everything else in the room does.

A Czar Keep trades the deskside form for density. In place of several Czars standing in a corridor, there is one chassis in a rack the organization already owns, on a segment the organization already runs. The trade is a trade of form only. The engines are not different engines. The models are the same open-weight models. The record is the same record. An infrastructure team that has run racks for years places the Keep the way it places anything else, and a platform team that has used a Czar finds the same services waiting.

Standing alone or joining a pool

A Czar Keep can stand alone. An organization with a hundred-plus people, a server room, and no robot floor runs the Keep as its sovereign AI appliance: the internal applications call its models, the organization's agents run in its Agency Studio, retrieval indexes the organization's own records, adaptation and evaluation run on the organization's own data. Nothing about standing alone is a reduced mode. It is the Keep doing what a Grid does, for people rather than for a floor.

A Keep can also join a Czar Grid pool. Where a facility already runs Czars for its floors and the organization wants to serve its people and its machines from one pool, the Keep's engines join the Czars' engines. The services are the same services, the agents the same agents, the record the same record. The pool simply has more engines in it, some of them in a chassis in the server room and some of them in nodes on the floor. The organization can begin with the Keep alone and join a pool later, or begin with a Grid and add a Keep later. In each case the change is to the configured appliance described in the agreement, not to how any of it works.

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.

What does not change

Everything that makes a Czar a Czar carries over to a Czar Keep unchanged, and it is worth being exact about the list.

The Substrate does not change. Inference, Retrieval, Memory, Adaptation, Evaluation, Watchman, and Local Bridge run on the Keep as they run on a single Czar and as they run across a Grid. Agency Studio is the same place to author, bound, approve, and record agents.

The posture does not change. The Keep can run with no route out at all. Where the organization permits connections, they are declared one by one at commissioning, such as identity, the record systems, ticketing, logs, source control, and the data stores, and Local Bridge permits those routes and denies everything else. Updates arrive as signed bundles, by media or by the organization's own network. Support and remote sessions are initiated by the customer through the customer portal; nothing begins from the outside.

The record does not change. Every request, every agent run, every approval, and every update is on the record the organization holds, on its own hardware, readable by the people it designates and by no one else.

The subscription does not change. One line covers the configured appliance, hardware, the platform license, and sustainment, fixed for the term. Usage does not move it. More people calling the models, more agents, more retrieval, and more adaptation leave it where it is. More engines, a second Keep, or a Keep joining Czars in a Grid is a change to the configured appliance described in the agreement.

A Grid in one chassis is a precise description, not a slogan. The organization that understands a Grid already understands the Keep. The only new thing to learn is where it goes.