Before your first sort run

Five things stand between a set-up machine and a first sort run. All of them are in the UI and none of them takes long. Four are checks; the one you have to do is step 4, deploying a profile.

Before you start

• First setup in the UI is done: the UI opens on the dashboard, not on the wizard.
• Camera calibration is done.
• Chute calibration is done, and a test aim landed the chute centred over the bins you tried.

1. Check the machine setup matches your build

Open Settings → General → Machine setup. It names the shape of your machine: standard carousel, classification channel, or manual carousel. A new install is set to classification channel.

It has to match the machine you built. If it is wrong, change it here before you go any further, then go back through the Cameras step of the setup wizard: this setting decides which cameras and which endstops the machine asks for.

2. Set the storage layers

Open Settings → Storage Layers. Each layer needs four things: switched on, its number of sections, the number of bins in a section, and the servo that opens its doors.

The setup wizard assigned the servos, and you set the counts at chute calibration. This is the check: if your chute test aim landed centred every time, the counts are already right.

3. Check the detection model

Detection is how the machine sees that a piece is there and where it is on the channel. Each camera channel runs its own model.

There is nothing to choose on a new machine. When it first comes online it asks Hive for the default detection model for its hardware (on an Orange Pi 5, r5 C-Channel Full YOLO11s 320, a model built for the Orange Pi's NPU), downloads it, and puts it on every channel that has none. If the machine was offline, it tries again every five minutes.

To check, open Settings → Local Models. The model is in the Installed list, and its Active menu names every channel. To use a different model on a channel, pick it from that model's Activate for subsystem menu.

Three things about this page:

• Classification C-Channel and Carousel detect are one setting shown as two rows. Set either one and both change.
• The page lets you activate a model on a channel it was not trained for. It marks the row with a note and does not stop you.
• The change takes effect in a second or two. Nothing needs restarting.

4. Deploy a sorting profile

Open Profiles, then press activate on Presort. Presort has eight categories and a catch-all, so nothing falls through on a first run.

Activating asks how the bins should start. Pre-assign from rules fills the bins in order: the first category goes to the first bin of the first section of the first layer, the second category to the next bin, and so on. Read that order off the screen and put your bins where the machine expects them. Reset bins empties them instead and assigns each category to a bin the first time a piece needs one.

Presort is the profile that ships with the machine. When you want your own boxes, build your first sorting profile walks through making one in Hive and getting it onto this machine.

No bin is kept for pieces that match nothing. Those drop out of the bottom of the tower, so put a box or a tray under it before you start.

5. Check the machine is on the internet

Every piece is identified by a service on the internet. A machine with no connection still feeds, sees and moves pieces, and identifies none of them. What leaves the machine covers what is sent.

You need no account and no key for this. The OpenRouter key on the Settings page is a separate, optional thing.

Home the chute before every run

The aiming is saved. The homing is not. Home the chute from Settings → Chute → Home to Endstop before each run, and again after any stall. Until it is homed, the aiming numbers mean nothing.

The finished result

A profile active, its categories assigned to bins, and the chute homed.

Next

Preparing LEGO for a sort run, which is what to take out of a tub of bulk LEGO before it goes in the bulk bucket, then your first sort run.