Where the pool sits in the building
An inference pool is a sealed node in the building's own rack, in a room the site already secures. Here is how operators decide which room, which network, and how many nodes, for a floor or a whole facility.
· 4 min read
IN SHORT
The pool sits in the building's own rack, in a conditioned space the site already secures, on the facility's own network. One Czar is a pool for a small floor; several joined as a Czar Grid are one pool for a whole facility. The robots it serves are on the floor or outdoors; the node is where the site keeps its network equipment, and nothing it holds leaves that room.
The question comes up early in every conversation about an inference pool, usually from the facilities engineer rather than the automation lead. Where does it go. The answer is short, and the reasons behind it say a good deal about what the pool is.
It goes in the building's own rack. Czar is a sealed node, installed in a conditioned space the site already secures, on the facility's own network. It is not on the floor among the robots, it is not in a vendor's cloud, and it is not in a trailer outside. It lives where the site's network equipment already lives, and the site's existing rules about who may enter that room apply to it.
The room
Most facilities already have the room. A warehouse has a space near the dock office where the switches and the warehouse system's servers sit. A plant has a controls room or an IT closet inside the plant segment. A terminal has an operations building. The pool goes there, in a rack the site owns, under the site's physical access control.
The reasons are practical. The space is conditioned, which the node wants. It is secured, which the node's contents warrant, since every camera feed, capture, task record, and adapted model on the site is on that hardware. And it is where the facility network terminates, which is the point.
On a cold-storage site the room is on the warm side of the wall. On a terminal or an intermodal yard it is indoors while every machine the pool serves is outdoors. The node is never where the weather is. The network is what reaches the machines.
The network
The pool serves inference over the facility's own network to every robot in the building, from any vendor, within each robot's latency budget. That makes the network part of the installation. Element 31 engineers install the pool with the operator's automation team, the integrator, and the robot vendors, and part of that work is confirming the network reaches where the robots actually are, at the far end of a yard, inside a freezer chamber, along a line.
On a connected site, the node sits on the segment the site chooses, and Local Bridge permits only the routes the site declares: the robot vendors' fleet software, the warehouse or execution system, and whatever else the site named. Egress is denied by default. On an air-gapped site there is no route out at all, and updates arrive as signed bundles carried into the room on media. The room is the same. The posture is a decision.
One node or several
One Czar is a pool for a small floor: a single cold room and its dock, a kitting cell and its route, a modest sortation floor. Several Czars joined are a Czar Grid: one pool for a whole facility. The Grid is still Czar. Adding a node to the rack adds to the pool; it does not add a second system or a second thing to manage.
How many nodes a building needs depends on the floor and the fleet, and it is sized during installation rather than guessed from a brochure. Whatever the count, the site holds a single subscription for the Grid as configured, and more robots on the pool do not move it.
What is in the rack and why it is sealed
The node is a sealed enclosure running a sealed, signed, read-only software image. Before it joins a Grid or accepts a task, it attests what it is running. The integration, sealing, and software load of every node are on record, so when the security lead asks where this hardware came from and what is on it, the facilities engineer answers from a record rather than a reassurance.
What the node holds is the site's: the feeds it has read, the captures it has kept, the models it has adapted, and the record of every decision it has served. None of that leaves the room. Watchman verifies the integrity of what is running. The portal at e31.online is where the site's staff go for support, billing, and signed updates; the node never connects to it on its own, and every remote session is customer-initiated.
What the room does not need
It does not need a vendor's engineer with a key. It does not need a path to the internet. It does not need to be rebuilt for the pool, since the pool goes in a rack the site already has. And it does not need to be anywhere near the robots, which is the thing that surprises people the first time: the machines are on the floor and in the yard, and the thing that thinks for them is in a locked room by the dock office, on the site's own hardware, where the site can see it.
For how this plays out on a cold site, see An inference pool for a cold-storage warehouse. For a site that is mostly outdoors, see An inference pool for an intermodal yard. For the appliance itself, see Czar.
Questions this raises
- Does the pool need its own room?
- No. It goes in the building's own rack in the conditioned space that already holds the facility's network equipment and the warehouse or execution system. The site's existing physical access controls apply to it.
- How many nodes does a building need?
- It depends on the floor and the fleet, and it is sized during installation with the operator's automation team and the robot vendors. A small floor may run on one Czar; a large facility runs several joined as one Grid. The subscription is for the configured Grid and does not move when more robots join.
- What if the robots are outdoors, as on a terminal?
- The node stays indoors in the terminal's own building, and the terminal network carries inference to the vehicles, cranes, and gate. The sealed enclosure is for a rack, not for weather.
KEEP READING