They home first, which the factory equivalents do not: a calibration run from an assumed position measures the wrong thing.

CALIBRATE_SHAPER_NEW stops the sweep at 130 Hz. The stock macro runs to 150, and on this machine nothing above 130 adds anything worth having — the extra range is only time on the accelerometer. It is not a typo, and it should not be “fixed”.

FILA_STATUS reports what the stock firmware keeps to itself: whether filament is present, and whether runout protection is actually armed. Those are two different questions, and the factory display answers neither.

• Phrozen-update protection — pressing Update on the Phrozen display overwrites the Klipper core + printer.cfg and halts the printer (Katapult keeps the F103 safe). The smart way is to stay ahead of it — this option has three parts:
  • Back up once, right after setup (recommended) — a “golden” snapshot of your patched stack + config.
  • Pre-patch USB — before every Phrozen update (recommended) — it bakes your fixes (from that backup) onto the update stick, so Phrozen’s update installs them along with itself. Each new update incrementally carries the fixes — nothing gets clobbered and you never need to recover.
  • Restore — the fallback only if an update ever slips through un-patched: one click puts back the v0.13 core, patched module and your config (calibration preserved) and the printer comes back up.
• Save / restore SETTINGS — the guards restore the Klipper core and the phrozen_dev module on their own, so the software needs no preparation any more. What no guard can do is give back the numbers your machine measured. Backup captures your printer configuration and calibration, the web interface’s own settings (theme, presets, macro groups, history — none of it in printer.cfg), your WiFi and the phrozen_dev module; Restore puts them back. It can also write the backup to a USB stick, which is the copy that survives a reflash — the local one lives on the very eMMC it is protecting. Pre-patch additionally writes your config onto a Phrozen update stick, so the installer deploys your printer.cfg instead of Phrozen’s.
• Emergency repair — the one to reach for when the printer halted, the display is dead, or an update button keeps failing. It runs every repair in order without asking what went wrong, because at that moment nobody knows whether Phrozen’s firmware, a Klipper update, a Moonraker update or a “hard recover” caused it. Every step is idempotent, so it is a no-op on a healthy printer, and it reports what actually needed fixing. It also covers two things no automatic guard catches. First: an AddOn.cfg that lost its [arco_mcu_timing] section, which silently stops the MCU timing from being applied. Second: root-owned files in ~/klipper or ~/moonraker, which make Moonraker’s update button fail with a git error that points nowhere near the actual cause.
• Check self-heal guards — the entry that replaced “Re-apply Klipper patches”, which had become redundant: those patches are re-applied by a guard on every Klipper start and by Emergency repair. What nothing checked is whether the guards are wired at all. Guards are installed when the image is built, not when the kit updates itself. So a printer that has been running a while can be missing a guard the current kit assumes is there, and there is no symptom until the failure that guard exists to catch. This compares klipper.service against what the kit’s own installer writes, so the expected list cannot drift, and offers to install anything missing. You do not have to do anything about that: klipper.service runs a series of ExecStartPre guards before klippy loads, so the restart that follows any update puts everything straight back. This menu option is the manual equivalent, idempotent and safe anytime. > Moonraker’s three actions are not equally harmless. Update refuses outright on a modified repo > and otherwise just pulls. Soft recover resets tracked files and leaves untracked ones alone. > Hard recover runs git reset --hard + git clean -d -f, so everything untracked is deleted — > phrozen_dev and our modules with it. Slow and unnecessary, but not fatal: the guards put our modules > back automatically, and a copy of your phrozen_dev is kept outside the Klipper tree for exactly this > case. If it ever does go wrong, Emergency repair above is the one action to run. > Klipper is not pinned: it sits on master with HEAD attached, the repository is clean and valid, > and Moonraker offers new versions normally — nothing is greyed out or held back. > (scripts/pin-klipper-updates.sh can hold a version deliberately, if you want that.)
• PhrozenGo / Cloud (scripts/phrozengo.sh) — Privacy closes the frpc SSH tunnel to the vendor’s server; the app still works via TUTK. OTA toggles Phrozen’s auto-update — turn it OFF to protect your v0.13 from a hostile Phrozen firmware update. Disable stops the cloud app entirely (no phone-home), frees the webcam + resources, and stops Phrozen from deleting Obico on boot. Defaults are as Phrozen ship them (all on); local display + light always stay. > ⚠️ Installing Obico? Turn PhrozenGo off first. Phrozen’s own KlipperScreen launcher runs > rm -rf ~/moonraker-obico and ~/moonraker-obico-env on every boot, and PhrozenGo ships > on, so those lines are live on a fresh printer. Obico installs cleanly, works until the next > restart, and is then simply gone — with nothing in any log to explain it. Disable here comments > the lines out. Other Moonraker add-ons (Spoolman, Mobileraker) are not touched.
• AMS — nothing to switch. The unit enumerates as a USB serial device, so the printer detects it and keeps a macro-readable ams flag in step, within seconds of plugging in or unplugging. That flag steers OrcaSlicer automatically — the print mode is picked at print start, see OrcaSlicer. Note: a factory-NEW AMS additionally needs one-time provisioning (firmware flash/pairing).

scripts/beacon_toggle.sh status|on|off · setup menu → b

Swaps Phrozen’s piezo probe for a Beacon eddy-current probe: Z is homed by a virtual endstop, the bed mesh is scanned instead of poked, and a 15×15 mesh costs less time than the stock 6×6 did. The config comes from a real conversion contributed by Philippe Humeau (unPhrozen) — including the gotchas he paid for, which are written into beacon.cfg where you need them.

Toggling is reversible: off puts the piezo probe sections back exactly as they were (verified) and keeps your calibrated beacon.cfg. It edits only the lines it marked as its own, so the rest of printer.cfg survives both directions untouched. on walks five gates before it changes anything, and each one aborts cleanly:

1. Not mid-print — switching restarts Klipper. Checked, not asked.
2. Is the machine physically ready? — Beacon mounted, wired and plugged in; rigid mount; bed clear. Asked before anything is downloaded, so declining leaves no half-finished conversion.
3. Is a Beacon actually on USB? — the one thing the script can verify by itself. No probe, no change.
4. The module. Klipper does not ship one: mainline v0.13 has generic eddy-current support (probe_eddy_current + ldc1612, for LDC1612 probes like BTT Eddy), but Beacon speaks its own protocol over its own MCU, so it needs the vendor’s out-of-tree module. If it is missing, you are told what will be downloaded, from where, under which licence and to which path — and asked. Nothing is redistributed by this kit.
5. The config change — type BEACON. A backup is written, and it rolls back if Klipper does not come up.

⚠️ Read this before switching

1. It is experimental in the literal sense. No Arco Unleashed developer has run it on a machine. Every number in beacon.cfg — y_offset, z_positions, mesh bounds — is one other printer’s value, not an Arco constant. Treat them as a starting point and verify each one.

2. Do not use the Arco display for anything Z-related afterwards. Not Z-calibration, not auto-levelling, not “probe” or mesh from the touchscreen. Its calibration flow was written for the piezo probe: it declares Z positions instead of measuring them (SET_KINEMATIC_POSITION), and it loads a stored mesh that was probed against a different Z reference. On a Beacon machine that can drive Z out of safe bounds — it was tested, and it did. Use Mainsail or Fluidd for homing, probing, mesh and Z-offset. Printing from the display is unaffected.

3. Verify z_positions before you trust Z_TILT_ADJUST. The order must match the physical Z motors, not the order of the config sections. Swapped, z_tilt diverges instead of converging and the retries make it worse. Check with STEPPER_BUZZ STEPPER=stepper_z / stepper_z1 and watch which side of the gantry moves.

What it changes. Four things have to disappear from printer.cfg, which no include can do. They are: [probe] (claims the same !PB9); [homing_override] (claims G28 and drives Z against the piezo); stepper_z’s position_endstop (Klipper rejects it as unused once the endstop is virtual); and stepper_z1’s endstop_pin (a second endstop on a probe-homed rail). Each is commented with a #:beacon: marker, so off is a prefix strip. Everything additive lives in beacon.cfg, included last so its section merges win.

Update safety. The Beacon module is untracked in Klipper’s tree, so Klipper updates leave it alone; Moonraker is not involved. A Phrozen update is the dangerous one — it replaces printer.cfg wholesale, which would put the piezo config back while a Beacon sits on the toolhead, and the next G28 would drive Z down waiting for a trigger that cannot come. So Beacon mode is re-applied by the ExecStartPre config guard on every Klipper start, keyed off a marker file, before klippy parses anything. A copy of your beacon.cfg is kept outside printer_data/ for the same reason.

First steps after switching (in Mainsail/Fluidd, not on the display): STEPPER_BUZZ both Z steppers → G28 → Z_TILT_ADJUST → BEACON_CAL (contact auto-calibration) → BEACON_MESH (re-scan the mesh — the saved one was probed with the piezo).

scripts/sensorless_toggle.sh status|on|off · setup menu → s

X and Y stop by detecting motor load (Trinamic StallGuard) instead of by a microswitch. Z is untouched and keeps its load-cell probe.

The microswitches remain the default and the recommendation. They stop at the same physical place every time, for free, with nothing to tune. If your switches work, leave this off.

What makes it worth shipping is one specific failure: X’s microswitch hangs off the toolhead MCU while the driver’s DIAG line goes to the main MCU. A broken toolhead cable therefore kills the switch but not the sensorless path — the printer homes again without waiting for a spare part.

Proven on hardware here: G28 X, G28 Y and a full G28 all home on StallGuard, and the disconnected switches still click at the end of the move. Note that G28 X brings Y to its endstop and clear first, on switches and on StallGuard alike — see Homing a single axis below. The click is the point — the stall happens at the mechanical stop, so the zero does not shift and the filament cutter (X=319), the wipe position (Y=322) and any saved mesh stay valid.

on equalises the two homing speeds, installs a sensorless.cfg that gates StallGuard by velocity, and rolls everything back if Klipper does not come up. It levels the speeds down, never up: StallGuard’s reading is velocity-dependent, so two axes at different speeds cannot share one sensitivity. off puts the endstop and homing-speed lines back exactly as they were, and removes the include — verified against the toggle’s own backup. It edits those lines and nothing else, so anything else in printer.cfg is carried through both directions untouched. A Phrozen update replaces printer.cfg. So while sensorless mode is enabled, the kit re-applies it before every Klipper start. Otherwise the update would hand you back the dead switch you switched away from.

⚠️ Read this before switching

1. Watch the first few homings, with M112 in reach. Wrong sensitivity means the carriage grinds into the rail instead of stopping, and on CoreXY that drags the other axis along. Listen for the switch to click: Klipper reports position 0 whether it reached the wall or tripped 2 mm in, so the sound is the only honest signal you have.

2. Sensitivity is driver_SGT, and the scale is backwards from intuition — LOWER is MORE sensitive. Grinds without stopping → lower it. Stops early with no click → raise it. The shipped 1 is what this hardware wanted at 30 mm/s and 1.2 A.

3. Calibrate at your running current. A value found at a reduced current does not transfer. With too little current the motor slips instead of building the load StallGuard measures, so nothing triggers at all — which looks exactly like “sensorless does not work on this machine”.

G28 X moves Y first. That is deliberate, and it is worth knowing before it surprises you: Y travels to its endstop and then 50 mm clear, and only then does X home. G28 Y homes Y alone and leaves X where it is. A plain G28 is unchanged — same full sequence, same probe.

The reason is a crash reported from a printer. Park the toolhead at the back, restart the firmware, send G28 X, and the head is dragged sideways through the wipe unit at Y=322 on its way to the X endstop — confirmed on the machine it happened to. Two things make that possible. Phrozen’s homing section is declared axes: z, so Klipper never routed a single-axis home through it and the Y-first order it already contained was skipped. And a firmware restart leaves the printer looking homed. Instead of clearing the position, it declares one, so Klipper believes the head sits at the middle of the bed no matter where it really is, and nothing refuses the move.

Y is homed rather than nudged forward a fixed distance because a homing move watches the endstop while it travels. A plain move does not, and no fixed distance can know how far back the head started.

Same behaviour on microswitches and on StallGuard.

One thing this does not fix: Z. After a firmware restart the declared height is an assumption, not a measurement, so home before you move Z. The declaration is deliberately set to the bottom of the travel, so Klipper refuses anything more than 5 mm downward until a real home. It gives you no protection upward: a large upward move is still yours to get wrong. G28 works from anywhere and settles it.

An electric-cobalt comic theme matching the Arco Unleashed branding — navy glass panels, halftone-burst background, the bookworm logo + a Burst “U” favicon.

• Mainsail — mainsail-theme/, light / dark + a switcher. Full details in its README.
• Fluidd — fluidd-theme/, the same look ported (both apps are Vuetify, so it is the same design, not a lookalike). Install with bash fluidd-theme/setup-fluidd-theme.sh, then reload with Ctrl+F5 and pick the dark theme. Fluidd has no hook for a custom logo or favicon, so those stay its own.

Mainsail can’t add named themes to its dropdown (that needs a Mainsail fork/rebuild, which breaks on every update), so the kit ships two .theme variants + a switcher instead — update-safe:

Switch over SSH with sh ~/printer_data/config/unleashed-theme.sh next, which cycles light → dark → stock. Or switch with one click in the Macros panel via SWITCH_THEME (needs the gcode_shell_command extension; setup-theme-macros.sh wires it up). Hard-reload after switching (Ctrl+F5). The active state is kept in .theme-state and survives reboots. Note: the .theme overlay is global, so it also tints the built-in themes — keeping those pristine would need a Mainsail fork (out of scope, not update-safe).

The Arco is a single-nozzle multicolor printer (filament-swap, like a Bambu AMS).

You need exactly one OrcaSlicer profile. Not one for AMS and one without, and nothing to switch before you slice. Slice single-colour or multicolour as the model needs, send it, and the printer picks the right AMS mode at print start by itself.

That works because the printer makes the decision, not the slicer. It decides from two things: what you sliced — one filament or several, which the start G-code passes through — and whether an AMS is attached, which the printer remembers.

Nothing to set, not even when the hardware changes. Attaching or removing the AMS used to need a command; it does not any more. The unit is a USB serial device, the printer watches for it, and the flag OrcaSlicer reads follows within seconds — no restart, no menu, nothing to remember. FILA_STATUS shows what the printer currently thinks, if you want to check.

The flag lives in [save_variables], so it survives reboots and updates, and arco_tool_gate keeps it honest: T1–T15 are registered exactly while an AMS is attached. There is one case it will not act on: a running print. If a unit drops off the bus mid-job, the tools and the flag stay as they are until the job has finished. Taking either away underneath a multicolour print is worse than a stale reading.

Profile. Stock OrcaSlicer already ships the official Phrozen Arco profile (vendor Phrozen, no fork needed). The kit adds an enhanced variant in orca/ with the Machine G-code already filled in — import it via File → Import → Import Configs…. To set the fields by hand instead, or to check an older profile against the current one, go to MANUAL › Step 9. It gives all four G-code fields in full, with a screenshot of the tab each belongs in.

All of that happens inside the start G-code: PHROZEN_AMS_START reads the flag and the print begins in the right mode. Nothing to pick, nothing to remember — and if you slice multicolour with no AMS attached, it stops and says so rather than starting something it cannot finish.

One thing does not announce itself, though: with the AMS fitted and a single-colour slice, the printer runs an endless spool and refills from the other slots when one runs out. Load them with the same colour — otherwise a long print changes shade partway through, and nothing warns you.

The start G-code heats the bed straight to print temperature and holds the nozzle at a probe-safe 140 °C through home and probe, so it cannot ooze onto the load cell. G30 then loads your saved phrozen mesh instantly — calibrating and saving it the first time, if none exists yet. Homing is automatic. It needs the auto_mode and ams features in AddOn.cfg, both on by default, and the gcode_shell_command extension.

Instead of loading the saved full-bed mesh, Klipper can probe only the print area on each print. You trade the instant G30 load for 30–60 s of probing per print, and get a mesh measured where the part actually sits. Adaptive meshes are transient (adaptive-XXXX) and never overwrite your saved phrozen profile, so switching back is just switching back. It needs OrcaSlicer’s Label objects checkbox — that is what emits the object bounding boxes adaptive meshing reads.

How to switch, and the one thing to check on your first slice: MANUAL › Step 9.

You do not need mode-specific Pause or Change filament G-code. One profile covers both cases, because the printer-side macros read the ams flag themselves. With an AMS, a colour change purges the slice’s flush volume through it. Without one, it runs the full manual retract → load → purge → resume. Pause behaves the same either way. Stock Klipper has no M601, so the kit maps Orca’s M601 onto Klipper’s PAUSE and a pause-at-layer works instead of hitting Unknown command.

Runout protection depends on the start G-code. The toolhead sensor is watched by Phrozen’s module, but only once a print mode has been set — which is what PHROZEN_AMS_START does. A start G-code without it prints unprotected, and stock firmware never says so. This kit does: a console warning once the job is past its start G-code, FILA_STATUS on demand, and printer['arco_fila_status'] (filament_present, adc, threshold, protection_active) for your own macros. It is read-only — detection and pausing stay with Phrozen’s module.

Three macros for the mechanics, all of them in AddOn.cfg and all toggleable from the setup menu. They derive their limits from your own configuration rather than carrying hard-coded coordinates, so a different bed or a different mesh does not silently inherit numbers chosen for one machine.

BELT_TENSION — homes, then parks where both belt spans are equal, which is where you pluck them. That is not the middle of the bed and not the middle of the travel: behind the Y rail there is a further 60 mm of belt to the idler (measured on this machine), so the middle of the span is Y190 where the middle of the travel would be Y160. Park at the wrong place and two perfectly matched belts read as mismatched. Aim for equal pitch between the two, not for any particular frequency — the absolute number depends on span length. Leave the steppers energised; M84 lets the belts go slack and the reading is void.

BELT_WARMUP — runs the gantry through its range to warm the belts and steppers up. BELT_WARMUP ACCEL=5000 MARGIN=40 SPEED=200 CYCLES=10. It bounds its own acceleration rather than inheriting the machine’s 40000 ceiling. At the low end it keeps MARGIN off the frame; at the high end it stays clear of the purge/wipe unit. Those two ends are not symmetric, so it derives them separately. Previous limits are restored, and it re-homes at the end rather than trusting the step count, because a cold-belt warm-up at speed is precisely what can lose steps.

CLEAN_IDLERS — turns the idler pulleys a measured amount so you can wipe the deposits off them. Run it with no arguments and it opens a dialog in Mainsail or Fluidd: pick the left or right pulleys, and the toolhead parks at the opposite end so it is not travelling past the hand holding the cotton bud. A second dialog with a Stop button replaces it while it runs.

One turn of a 20-tooth GT2 pulley is 40 mm of belt, so the default 1.5 turns is 60 mm — and on CoreXY that is also 60 mm of carriage travel, with no ratio to correct for. Everything is adjustable: CLEAN_IDLERS SIDE=left TURNS=1.5 TEETH=20 SPEED=45 PAUSE=3 PAUSE_TURN=1. It pauses before each forward pass, which is when you hold the bud against the pulley — not while it moves. At the far end it pauses only briefly. It refuses while a print is running, and refuses if the requested turns would carry Y past the purge unit.

config-templates/    printer.cfg + includes + AddOn.cfg + wpa  (templates; secrets/serial removed)
scripts/  unleashed_setup.sh   the setup MENU; _arco-lib.sh = shared helpers/actions
          fetch-phrozen-fw · usb-fw   <- installs Phrozen's module: user's USB zip first, else a confirmed fetch from Phrozen's own repo)
          apply-phrozen-patches  (3 v0.13 sed fixes + the SHAPER_END/BED_PROBE_END cal-handshake macros — only our rules)
          apply-phrozen-restore  (keeps a copy of YOUR phrozen_dev outside the Klipper tree; puts it back if an update deleted it)
          emergency-repair  (one action for "something broke" — runs every repair, reports what was wrong)
          apt-hold (mask auto-apt/unattended-upgrades + hold kernel/DTB/fw) · fix-gpio-led · resize-emmc · optimize-fs (noatime)
          optimize-boot  -> 40k real-time tuning: performance governor · klippy CPU3 + F407-IRQ affinity · numpy=1 · F407 USB no-suspend · ImageId
          flash_mcus [f407|f103|host] · install-katapult · set-mcu-serial
          phrozengo · addon-toggle · addon-features · phrozen-recover · selfupdate
          beacon_toggle  (EXPERIMENTAL: piezo probe <-> Beacon; reversible printer.cfg surgery)
          arco_*.py  own Klipper extras, untracked + self-healed by apply-arco-extras (ExecStartPre):
                     mcu_timing (MCU timing without patching mcu.py) · tool_gate (hide T1-T15 without an AMS)
                     sdcard_select (SDCARD_SELECT_FILE for the display) · fila_status (filament state, read-only)
          arco-firstrun (portal→fetch→install) · wifi-portal/ · tft.sh (boot progress screen)
          arco-splash (loading-screen branding)
system/                   USB automount (udev rule + mount service) for the gcodes/USB stick path
collect_data_arco.sh grab phrozen_master + ~/hdlDat (Phrozen gateway, NOT in the zip) from YOUR
printer -> USB
mainsail-theme/           optional Arco Unleashed Mainsail theme (Voron Light/Dark + switcher + mainsail-seed)
orca/                     OrcaSlicer profile (flag-driven AMS auto-mode; import or copy the start-gcode line)