openlamp-midi 0.1.1

MIDI + Ableton Link frontends for OpenLamp/LumiDeck smart lamps — drive lamps from a MIDI controller/DAW and flash them on the beat.

openlamp-midi — MIDI overlay for LumiDeck

A MIDI frontend for the LumiDeck — see the OpenLamp umbrella) engine: drive your smart lamps (and, soon, anything else) from Ableton, Bome, Logic, or any cheap physical MIDI controller — and from the Stream Deck via its MIDI plugin.

This repo is the top layer of a stack. It references the others:

Layer	Repo	Role
core	Beennnn/lumideck — see the OpenLamp umbrella) → core/	standardized LED interface + OpenLamp State (OLS) contract + engine
streamdeck	Beennnn/lumideck — see the OpenLamp umbrella) → streamdeck/	Elgato Stream Deck plugin
midi	this repo	MIDI → OLS overlay

The overlay never talks to a device directly. It opens a virtual MIDI port LumiDeck, translates incoming MIDI into OLS commands, and calls the engine's local API (http://127.0.0.1:8377/cmd). The engine (from the core) owns the persistent device connections, so MIDI-triggered changes are as instant as a key press — and stay in sync with the Stream Deck.

For musicians, not Stream Deck users

Controlling lamps over MIDI is not a Stream Deck feature (the Stream Deck plugin drives the engine directly). This overlay targets the MIDI musician community — people who already own physical MIDI controllers and want to fire lamp colors from them, live on stage. See ENCAPSULATION.md.

Why MIDI

MIDI is the cheapest, most ubiquitous physical control layer: €20-80 pads, faders and footswitches, real-time, recognized by every OS and DAW, no drivers. This overlay turns any of them — or a Stream Deck — into a lamp/show controller.

• One MIDI channel per lamp group (channels map in mapping.json).
• Notes → colors/power/animations, CC → brightness/temperature/continuous hue, Program Change → scenes/presets/snapshots, MIDI clock → tempo.
• Full mapping and coverage matrix: MIDI-PROTOCOL.md.

Beyond lamps — a universal encapsulation bus

The lamp bridge is one instance of a general pattern: MIDI → any local backend (Home Assistant, DMX/Art-Net, OSC, MQTT, Shelly…), so one cheap controller drives your whole show and home. See ENCAPSULATION.md.

Install & run

pip install openlamp-midi          # pulls python-rtmidi; gives two commands:
pip install "openlamp-midi[link]"  # + Ableton Link support for beatsync (native build)

lumideck-midi                      # opens the virtual "LumiDeck" MIDI port
beatsync --source midi --port Ableton --action flash --colors rouge

Or run the scripts straight from a checkout without installing:

pip install python-rtmidi
python3 lumideck_midi.py

Route your DAW/controller output to LumiDeck. Autostart via com.openlamp.lumideck-midi.plist (launchd). The LumiDeck engine (Stream Deck plugin from the core repo) must be running.

Tempo & beat sync — beatsync.py

A second frontend in this repo, focused on rhythm: it follows an external MIDI clock (24 ppqn, Start/Stop/Continue) or an Ableton Link session and flashes / pulses / colour-cycles the lamps on the beat — locked to the music, no cable needed for Link. Same target as lumideck_midi.py: it POSTs to the engine's local API on 127.0.0.1:8377, respecting the ~4 commands/second the WLED firmware can ack (it drops excess ticks rather than choke the lamps).

pip install python-rtmidi          # MIDI clock source
pip install aalink                 # optional — Ableton Link source (native build)

python3 beatsync.py --list-ports
python3 beatsync.py --source midi --port Ableton --action flash --colors rouge
python3 beatsync.py --source link --bpm 120 --action pulse --accent

Subdivisions (--sub 1|2|4), per-beat action (flash / cycle / pulse), accent on beat 1, lamp/group targeting. Ctrl-C restores the lamps. Full options in the file's header docstring.

Landing the flash on the beat — latency learning + anticipation

There is always a delay between "beatsync decides to flash" and "the lamp physically changes": the HTTP round-trip to the engine, plus the WLED hardware's reaction time (~45 ms). Left uncompensated the flash lands after the beat. beatsync learns that delay and fires early so the light change coincides with the beat:

1. Learn the network delay. Every command POST to the engine (127.0.0.1:8377) is timed and folded into an exponential moving average rtt = 0.2·sample + 0.8·rtt. This is the network + engine share of the delay (small on a local machine, larger over Wi-Fi or a loaded engine — so it's worth measuring, not hard-coding).
2. Add the hardware floor. A fixed lamp-reaction constant (--latency-bias, default 45 ms — the measured WLED floor) is added: L = rtt + bias. Tune the bias by eye: raise it if flashes still feel late, lower it if they feel early.
3. Fire L ms before each beat.
   • MIDI clock: L is converted to whole MIDI clocks (24 per quarter, so clocks = round(L / clock_period)), and the tick fires that many clocks before the beat boundary.
   • Ableton Link: the shared timeline is polled; the tick fires the instant the next subdivision is L ms away ((next_beat − now)·beat_duration ≤ L).
4. Guard rails. L is capped at 200 ms (a bad measurement can never throw the flash wildly early), and a hard ≤ 4 commands/second limiter protects the WLED firmware — a blink is two commands (bright + dark), so this is enforced per command, not per beat. Excess is dropped, never queued.

Link vs MIDI clock — why Ableton Link is the better protocol

Both sources drive the same flash, but they carry different information, and that difference decides whether the downbeat (the accent on beat 1) can land where it should. Prefer Link whenever the DAW offers it.

	MIDI clock (0xF8)	Ableton Link
What's on the wire	24 anonymous ticks per quarter — tempo only	a continuously shared timeline: tempo + phase within the bar
Where's the bar / beat 1	nowhere — a tick is just a tick	link.phase is exact; 0 is the bar's downbeat
Tempo changes	followed tick-by-tick, but jittery (each tick is a separate event, subject to USB/driver scheduling)	smooth — you read a precise fractional beat position, not a tick count
Joining mid-song	you see only ticks; you've missed the one Start that marked bar 1	you're instantly in phase — the timeline is absolute and shared
Multiple devices	each derives its own guess of "where beat 1 is" → they drift apart	all peers read the same phase → sample-accurate agreement
Cabling	a MIDI/USB route (here: a virtual port)	none — peers find each other on the LAN

The crux is the downbeat. MIDI clock is a metronome with no score: it tells you how fast but never where in the bar you are. We confirmed this by sniffing Ableton during steady playback — it sends only clock (288 × 0xF8 over 6 s at 120 BPM), no Start, Continue, Stop, or Song Position. So under plain clock the accent can only be guessed, and it drifts.

MIDI does carry the bar in two occasional messages, and beatsync uses both when they appear:

• Start (0xFA) — sent once when you press Play. beatsync treats it as bar 1, so the accent locks — but only if you press Play after arming beatsync, from the top of a bar. Join mid-song and there's no Start to catch.
• Song Position (0xF2) — a 16th-note offset from song start; beatsync re-phases the bar from it. Reliable in theory, but many DAWs (Ableton under plain clock included) never send it.

Ableton Link needs none of that. Because the phase is shared continuously, the downbeat is always known — link.quantum is set to the bar length and the accent counter is seeded from link.phase, so beat 1 falls on Ableton's bar 1 the instant you connect, with no Play/Stop dance and no drift between two lamps. That's why the demos use Link for anything where the accent matters, and keep MIDI clock for the simpler "just flash on every beat" case where the bar position is irrelevant.

Rule of thumb: flash-on-every-beat → either source. Accent on the downbeat → Link, or a fresh Play from bar 1 if you're stuck on MIDI clock.

Publishing to PyPI (maintainer)

Releases are published to PyPI by Trusted Publishing (OIDC) — no API token is stored in the repo. .github/workflows/publish.yml builds the sdist + wheel and publishes on each GitHub Release.

One-time setup (yours — I don't touch PyPI credentials):

1. Create the PyPI project via a Pending Publisher (works before the first upload). On https://pypi.org → Your account → Publishing → Add a pending publisher, fill in:
   • PyPI Project Name: openlamp-midi
   • Owner: openlamp · Repository: midi
   • Workflow name: publish.yml
   • Environment name: pypi
2. Add the pypi environment on GitHub: repo → Settings → Environments → New environment → pypi (optionally require a reviewer for extra safety).
3. Cut a release: bump version in pyproject.toml, commit, then create a GitHub Release (tag e.g. v0.1.0). The workflow builds and publishes — check the Actions tab. Done: pip install openlamp-midi works worldwide.

For TestPyPI first, duplicate the publisher on https://test.pypi.org and add a repository-url: https://test.pypi.org/legacy/ to the publish step.

Credits

Built by @Beennnn with the help of Claude (Anthropic). Part of the OpenLamp family. Not affiliated with Tuya, Elgato, the WLED project, or any MIDI vendor.

Works with any class-compliant MIDI controller

Any controller your Mac sees as a MIDI device works out of the box — route it to the LumiDeck virtual port (directly, or through your DAW). Typical stage picks:

• Foot controllers (hands stay on your instrument): Hotone Ampero Control (4 footswitches, ~80 EUR), Morningstar MC6/MC8, Behringer FCB1010 (10 switches
  • 2 expression pedals, the classic). Map switches to blackout / restore / scene recalls per song section.
• Pads: Novation Launchpad Mini, Akai APC mini (~80-100 EUR) — one pad per color per group; the 8x8 grid maps naturally to 8 colors x channels.
• Faders/knobs: Korg nanoKONTROL2 (~60 EUR) — hue / saturation / brightness on three faders (CC 3/4/1) = paint any color live with one hand.
• Keys: any MIDI keyboard — notes 60-67 are the color palette; velocity is ignored, so nothing fires by accident while playing softly... on another channel.
• Multi-FX pedals that send MIDI: Hotone Ampero II Stomp, Line 6 HX Stomp — a patch change on your guitar board can also switch the stage color.
• From the DAW: Ableton Live clips (one MIDI track per lamp-group channel), or MIDI clock for tempo-synced pulses.

No drivers, no config on the controller side: it just sends notes/CC — the mapping lives in mapping.json on the computer.

Family & the one-host rule

This overlay talks to the engine's local API (/cmd on 127.0.0.1:8377) served by openlamp-engine — either its headless daemon (no Stream Deck needed: run-headless.sh) or the lumideck plugin. Run ONE host at a time. Family map: openlamp.