analoghealth

A diagnostics mod for the Elektron Analog Rytm MKII (OS 1.74) and Analog Four MKII (OS 1.56): irpina's digihealth for the Digitakt mk1, ported to the two analog machines. It adds two rows to SETTINGS:

• SYSTEM INFO shows CPU, DSP, RAM and (on the Rytm) free sample memory in the top bar. A read-only USB diagnostics channel lets tools/analogusb.py read the same figures, and more, from a computer.
• REMOTE KEYS (on at every power-on) lets tools/analogusb.py keys press the panel's buttons, and tools/analogusb.py turn turn its encoders, over USB: see Remote keys. Off, the unit refuses key commands and lets go of any key the computer holds; the diagnostics still answer.

digihealth's third row, FAST AUDIO, is not ported: see Why no FAST AUDIO.

It is an elekloader mod and needs the device's core: core-ar2 on the Rytm (built and checked with 0.4), core-a4 on the A4 (built and checked with 0.1). elekloader builds a custom OS file on your own machine, from your stock OS file and the mods you pick; nothing from Elektron is distributed.

Until 1.1 it was called rytmhealth and was for the Rytm only.

Status: passed on a unit on both machines (Rytm OS 1.74, A4 OS 1.56), and checked in the emulator and on the host (see How it was checked).

This repository holds no Elektron code or data: stock code sites are named by address, length and SHA-256, and their bytes, like the panel's button layout, come from your own stock file or your own unit.

SYSTEM INFO

• The top bar shows two pages, two seconds each:
  • CPU and DSP load (DSP now, and its peak in the last second);
  • free RAM (the heap), and on the Rytm free sample memory (the A4 has none).
• DSP is the ColdFire's audio interrupt: on the Rytm the sample voices, the effects and the mix; on the A4 the effects and the mix. The analog voices cost it nothing, so they are not in it.
• CPU is the share of time the processor is not idle.
• The USB channel is SysEx with Elektron's manufacturer header and a device byte (0x7D) that no Elektron machine uses, so a stock OS ignores it. Its diagnostics commands are read-only: HELLO, STATS (the once-a-second snapshot: render and idle time, the heap by block size, free sample memory) and PEEK (DDR and the SRAM only). The diagnostics protocol is digihealth's, byte for byte.

python tools/analogusb.py cfw          # which mod answers, and its uptime
python tools/analogusb.py stats 10     # ten readings, one a second
python tools/analogusb.py peek 0x47be0000 64   # the mods' RAM area
python tools/analogusb.py --port "Analog Four" stats   # with both units connected

analogusb.py runs on Windows (winmm, through ctypes; no packages) and on macOS/Linux (Mido + python-rtmidi: pip install mido python-rtmidi). Close Elektron Transfer and Overbridge apps first on Windows, which lets one program at a time open a MIDI port. It looks for one MIDI input and output whose name contains "Analog Rytm" or "Analog Four"; with both units connected, --port NAME (before the command) picks one.

Remote keys

Every key of the panel: 78 on the Rytm (16 trig keys, 12 pads, 8 bank keys, the encoder pushes and every other button), 89 on the A4 (the same kinds, the track keys and the 13-key keyboard). Names are the OS's own (FUNC, TRK, GLOBAL, PLAY, ...): see the button maps. The client asks the unit which device it is and uses that device's names.

python tools/analogusb.py keys GLOBAL                     # open SETTINGS
python tools/analogusb.py keys 'FUNC(PLAY)'               # hold FUNC, tap PLAY
python tools/analogusb.py keys 'GLOBAL,DOWN,DOWN,YES'

A bare name taps that button; FUNC(PLAY) holds FUNC, taps PLAY, then releases FUNC before the next step. Commas separate steps; quote the sequence so the shell keeps the parentheses. --tap-ms defaults to 100 (maximum 250) and --hold-ms to 2000 (maximum 2000); the hold is a safety lease, not a delay. All names are checked before any key action, each step waits for completion, and an error stops the sequence and lets go of every key. Completion means the stock button driver took the transition, not that a menu opened.

• Rytm pads: a pad name in keys is the pad's key event only. To strike a pad, with its velocity, pressure and release as the panel sends them, use analogusb.py pad (pads); padcap logs what the panel sends and padplay sends a logged performance again.
• Encoders: python tools/analogusb.py turn E -12 turns encoder E 12 counts left, one count per UI tick (20 a second), as a panel turn does: the value popup and the header's long name appear as for a hand turn. The encoders are A-H and LEVEL on the Rytm, A-J and LEVEL on the A4. As on the panel, several counts make one value step (on a Rytm, 60 counts at 4 a tick moved FRQ 12 steps, and 20 counts at 2 a tick moved the track level 2), so a turn of a few counts may change nothing. The OS also drops the first counts of a turn after a pause, so a small turn out and the same turn back need not return to the start: in the emulator, panel turns of +10 and then -10 counts on the Rytm's SRC page moved SWT and left it there, as remote turns do on a unit. --step N sends N counts a tick (faster, and the OS's acceleration applies); --hold FUNC holds a key through the turn. An encoder pushed within the last moment ignores turns, as on the panel.
• Physical keys win: a key held on the panel cannot be taken by the computer, and a panel press of a key the computer holds survives the computer's release.
• Watching the panel: python tools/analogusb.py watch 30 prints keys as they go down and up for 30 seconds, and watch --until FUNC waits for a key. It reads the button state the OS keeps with PEEK about 20 times a second, so it needs nothing more from the firmware; it sees keys the computer holds too, and a tap shorter than one read can be missed.

Python callers can reuse any port with send, receive and close:

from analogusb import Diag, RemoteKeys  # tools/ on the import path

with RemoteKeys(Diag(port)) as keys:
    keys.press('FUNC')
    keys.tap('PLAY')
    keys.release('FUNC')

The protocol is digihealth's revision 2 with 128-bit key maps, because these machines have more keys than revision 2's 64-bit maps hold, and a TURN action for the encoders (revision 4). See the protocol reference.

Why no FAST AUDIO

digihealth's FAST AUDIO runs the Digitakt's audio code from a stretch of the processor's 64 KB on-chip SRAM that the Digitakt's OS leaves empty. The Rytm's OS fills and uses all of that SRAM itself, so there is no room to run code from without taking room the OS needs. The A4's OS also fills the whole SRAM from its image at boot; it has not been surveyed further.

Install

You need:

• elekloader with the device's core (see elekloader); from source, build it with python -m elekloader.sdk.build mods/core-ar2 --stock Analog-Rytm_MKII_OS1.74.syx or mods/core-a4 --stock Analog-Four_MKII_OS1.56.syx.
• This mod, built from source (below).
• The stock OS file, Analog-Rytm_MKII_OS1.74.syx or Analog-Four_MKII_OS1.56.syx, from Elektron's downloads. elekloader recognises it by its hash.

python -m elekloader.patch --stock Analog-Rytm_MKII_OS1.74.syx \
    --mod core-0.4.elemod --mod out/analoghealth-1.3.elemod \
    --out Analog-Rytm_MKII_OS1.74-health.syx --version AH13
python -m elekloader.patch --stock Analog-Four_MKII_OS1.56.syx \
    --mod core-0.1.elemod --mod out/a4/analoghealth-1.2.elemod \
    --out Analog-Four_MKII_OS1.56-health.syx --version AH12

Flash it with Elektron Transfer, as for any OS update, and don't turn the unit off until the upgrade is done. Recovery: elekloader never changes the bootloader, so the stock OS file always restores the unit (hold FUNC while powering on for the startup menu, then OS UPGRADE).

Build it from source

The Digitakt mk1 cross toolchain (m68k binutils and gcc; on Debian or Ubuntu, apt install binutils-m68k-linux-gnu gcc-m68k-linux-gnu; on Windows, inside WSL) and elekloader, importable. The stock file picks the device:

python build.py --stock Analog-Rytm_MKII_OS1.74.syx     # -> out/analoghealth-1.3.elemod (Rytm)
python build.py --stock Analog-Four_MKII_OS1.56.syx --out out/a4   # (A4)
python -m elekloader.lint out/analoghealth-1.3.elemod --stock Analog-Rytm_MKII_OS1.74.syx --with core-0.4.elemod

How the port was made

digihealth's stock routines, found again in each OS, with that OS's addresses (ar174.inc, a4/a4156.inc). It hooks the same three places as digihealth: the task switch (idle time), the SysEx input (the USB channel) and the panel's button input (the remote keys). Encoder turns use the OS's own encoder input, with no further hook. On the Rytm, a fourth site, the entry of the panel's receive path, logs the panel's bytes and lets the pad player wait until the panel is between messages.

How it was checked

• Build: build.py and elekloader.lint against stock Rytm 1.74 with core-ar2 0.4 and stock A4 1.56 with core-a4 0.1: no overlaps, stock bytes as expected, code sites whole instructions; elekloader.patch builds and verifies both OSes (AH12).
• Host tests (tests/, 44): the real C protocol and bridge, built for each device, behind a fake port with a synthetic panel table: capabilities, a hold and tap, every key found and reaching its place, a key missing from the unit's table not offered, the CLI sequence, physical overlap, a busy queue, the REMOTE KEYS row off, the hold lease, name checks, encoder turns, and the client's selftest; and the host tools' tests. With ELEKLOADER_STOCK (Rytm 1.74) and ELEKLOADER_A4_STOCK (A4 1.56) set, it checks that each unit's table holds every key code in its keys-*.def.
• The Rytm in the emulator (trigcondition/analog-emu), from a cold boot (rytmhealth 1.0 to 1.1):
  • HELLO, STATS and CAPABILITIES sent on USB MIDI were answered: 1500 renders a second at 132.000 MHz, CPU 51 % and DSP 49 % at rest, 13.26 MB heap free, 64.00 MB sample memory free; 78 keys. The heap walk stopped one free list at a failed check, which digihealth also sees on a stock Digitakt (its README).
  • Remote GLOBAL opened SETTINGS; eight remote DOWNs reached SYSTEM INFO, after SYSTEM; YES ticked it; NO closed the menu and the top bar showed "CPU 51% DSP 49%/49%", then "RAM 13.2M SMP 64.0M".
  • A remote FLTR and then turn E -6 --step 2 showed FREQUENCY in the header and FRQ 121; a remote LEVEL turn showed TRK LEVEL 84.
  • A remote BD tap sent no MIDI note (a pad hit does).
  • The emulator stands in for the OS's SysEx sending, so the unit is the check for the USB replies.
• The Rytm on a unit (OS 1.74):
  • rytmhealth 1.0 (22 Oct 2026): every step of the hardware test passed: the rows, the readout stopped and playing, cfw and stats over USB, remote GLOBAL, the SETTINGS sequence, FUNC(PLAY), PLAY and STOP; REMOTE KEYS off refuses keys while diagnostics answer; a panel-held FUNC is refused as an overlap; a remote pad is silent; normal playing is unchanged; after a power cycle SYSTEM INFO is off and REMOTE KEYS on.
  • With key positions from the unit's own table (23 Oct 2026): the same remote keys, both rows and the readout in screen captures, and watch saw a held FUNC while keys 'FUNC(PLAY)' was refused as an overlap.
  • rytmhealth 1.1 (23 Oct 2026): on the FILTER page a remote turn E -60 --step 4 put FREQUENCY in the header and moved FRQ from 127 to 115, and turn E 60 --step 4 brought it back; turn LEVEL -20 --step 2 showed TRK LEVEL 98, and +20 brought it back to 100.
• The A4 in the emulator (the A4 build of rytm-emu, analoghealth 1.2, from a cold boot): HELLO answered "analoghealth 1.2"; STATS: 1500 renders a second at 132 MHz, CPU 49 % and DSP 48 % at rest, 13.80 MB heap free; CAPABILITIES: product 4, OS 1.56, 89 keys (each found in the unit's table) and 11 encoders. Remote GLOBAL opened SETTINGS, and SYSTEM INFO and REMOTE KEYS sit after SYSTEM; remote DOWNs and YES ticked SYSTEM INFO, and the top bar showed "CPU 49% DSP 48%/48%". A remote turn A -40 --step 4 on the OSC page showed COARSE TUNE in the header and the value -8, as an emulated panel turn does. The same build also went through the emulator's OS upgrade path: a stock A4 1.56 received it as elkflash sends it (no prompt), wrote its flash, restarted from it, and answered as above.
• The A4 on a unit (OS 1.56, analoghealth 1.2, 23 Oct 2026, flashed with elkflash): HELLO and STATS answered over USB (CPU 70 %, DSP 67 %, 13.77 MB heap free, 132 MHz, 1500 renders a second); remote GLOBAL and eight DOWNs reached SYSTEM INFO after SYSTEM; YES ticked it and the top bar showed "CPU 73% DSP 67%/72%", then "RAM 13.7M"; a remote turn LEVEL -20 --step 2 moved the SETTINGS cursor up a row; NO closed the menu with the readout still on. With FUNC held on the panel, watch saw it and a remote FUNC press was refused as a physical overlap; the panel's FUNC then came up normally. The unit then went back to stock 1.56.

Credits

• digihealth by irpina (GPL-2.0-or-later): this mod is its SYSTEM INFO and REMOTE KEYS, with the Rytm's and the A4's addresses; the protocol, the client and the input bridge are its.
• elekloader by irpina (GPL-2.0-or-later) and its core-ar2 and core-a4.
• analog-emu by trigcondition (MIT): the emulator the checks ran in.

Licence

GPL-2.0-or-later. analoghealth is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 2 of the License, or (at your option) any later version. It is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details: LICENSE holds version 2.

Not affiliated with Elektron. Analog Rytm, Analog Four and Digitakt are trademarks of Elektron. Custom firmware is at your own risk.