The print farm

This page describes the machines your part is printed on, who decides which one runs it, and what is watched while it runs. It is the farm as it actually works today, including the steps that are a person rather than a program. For throughput and lead time, read /docs/farm/capacity instead.

Where it is#

One site, in Austin, Texas. The printers belong to us. The person who slices your plate is the person who takes the part off it and packs the box, and no part of the job is handed to another farm. That is why this page can tell you which machine model ran your part.

The machines#

Every printer in the farm is a Bambu Lab machine, and the fleet is H2S and P2S. The build volumes below are the ones the plate validator measures a plate against, for the machine a job is queued on.

The validator also recognises H2D, X1C, X1E, P1S, P1P, A1 and A1 Mini, because a G-code file you upload may have been sliced for any of them. A plate sliced for a different Bambu model than the machine it is queued on is a hard failure: the file is refused before it reaches the printer rather than after the first layer.

Every machine carries a 4-tray AMS. A print using five to eight colours needs a second unit chained on and loaded by hand, which is what the $25 second-AMS fee pays for; eight is the most in one part, and we do not swap spools mid-print. While no printer has a second unit loaded, a five-to-eight colour order waits in the queue, flagged on the orders list, until one has; the queue, move and start steps refuse any printer with fewer AMS slots than the plate has colours, and any tray picked for two colours. Colour reaches a printer from a plate you sliced yourself, a Bambu Studio or OrcaSlicer project we slice in its own colours, or a model generated in the Studio. A multi-plate project prints as one job per plate. Ordering multi-colour is covered on /docs/studio/multi-color.

The live fleet count is deliberately not written down here, because it changes. GET /api/farm/capacity returns fleet.total along with how many machines are printing, idle, offline or in maintenance right now, and that endpoint is public.

What cuts the plate#

Plates are sliced by OrcaSlicer running headless in our own service, using Bambu's own machine, process and filament presets rather than hand-written settings. The presets carry the real flush volumes and tool-change behaviour, which is what multi-colour pricing is computed from. Default nozzle 0.4mm, default plate a textured PEI sheet - which is what is physically on the machines. This is why your quote is a weight rather than an estimate: where a slice succeeds, the grams you are charged for are the grams the slicer says will leave the nozzle, rounded up to the whole gram.

How a job reaches a printer#

1. 1
   The order is paid

   A website order record only exists after Stripe confirms payment, and it is created at the placed stage; that includes an order we agreed with you directly and sent as a payment link, which becomes an order when it is paid. A print API order exists from submission but is refused by the queue until it is paid. Nothing is queued before payment. Every paid order is given a ship-by day: 2 business days for Priority print, 5 for Normal, or the day the operator sets on an order entered by hand.

2. 2
   An operator picks the machine and the trays

   The operator opens the order in the console and chooses the printer and the AMS tray for each filament in a dialog. A single-filament part can instead print from the printer's external spool, which is how flexible TPU is run, since it cannot feed through an AMS. The server would have to guess which machine is free and which spool was meant, and a wrong guess is a wasted plate. The person is standing in front of the fleet; the server is not.

3. 3
   The record is checked against the metal

   Before slicing, the printer's own reported model is compared with what our database says it is. If they disagree the record is corrected and the plate is cut for the model the machine says it is. A plate cut for another machine's acceleration limits is how prints lose steps.

4. 4
   The model is sliced for that machine and that plastic

   The plate is specific to the printer it was assigned to. It is not a generic file dropped on whichever machine happens to be free. If the operator later moves a queued job to a printer of another model, the job is sliced again for that machine with the plate's own settings and validated for it, and a move that cannot keep those settings is refused rather than printing something different.

5. 5
   The plate is validated

   The G-code is scanned for dangerous commands, temperatures outside hardware and material limits, and a printer-model mismatch. Detail on /docs/farm/quality.

6. 6
   It joins the queue, and a person starts it

   The job waits at QUEUED or READY until an operator starts it. No scheduled task starts prints. The queue lists high-priority jobs first, which is what Priority print buys, and then the soonest ship-by. If the AMS trays changed since the job was queued, the start dialog reopens with a warning naming the change rather than running on the wrong spool.

7. 7
   The order follows its jobs

   A job queued from an order is linked to it, item by item. When every job on the order is done and every item has a completed plate, the order moves to printed by itself and you get the printed email. A failed or cancelled plate counts for nothing, so the order waits for its reprint.

What is watched while it runs#

Printers connect over the local network to a gateway box in the farm, which forwards their telemetry to the application. The operator console streams it live.

• Nozzle, bed and chamber temperature, where the model reports one.
• Progress, current layer, and the printer's own G-code state.
• AMS tray contents: the material and colour each tray reports.
• The external spool: the material and colour set for it on the printer. It is not metered, and a holder with nothing set is shown as unknown, never as empty.
• Fault codes, normalised into described faults with a remedy each, raised as work items and swept on a schedule. Advisory faults do not page anyone unless the printer has parked itself in a paused state.
• Whether the telemetry is fresh. The gateway serves a cached snapshot after its link to a printer dies, so readable state proves nothing on its own; staleness is tracked by watching the timestamp change, and a printer whose serial does not match its record has its readings withheld.
• Whether the printer is accepting files. A printer's file service can stop answering while its status and AMS stay live. The printer is then shown online, with "Printer file service not responding" in place of a lost connection, and a print start is refused with that reason before anything is sent, because the file is uploaded over that service. A single slow moment clears by itself within a couple of minutes; one that lasts raises an alert, and restarting the printer usually clears it. A print already running carries on, so let it finish before restarting.
• A live camera, with a WebRTC, MSE and MJPEG fallback chain.

After the plate finishes#

A finished print is not a collectable print. Parts release from a PEI plate when the bed contracts on cooling, so a part pulled warm either will not come off or deforms. The job is marked complete when the printer finishes, but it is not called ready to collect until the machine has cooled to a removal temperature.

Owning the farm versus subcontracting it#

The practical difference is that every number on /docs/farm/capacity is measured from the machines that will print your part, rather than promised on someone else's behalf.