t8live — Live Coding for the Roland AIRA Compact T-8

Version 0.1.0 · Nikša Barlović · 2026 · Fork of Strudel (AGPL-3.0-or-later)

---

AI READING INSTRUCTION

This document is written in HADS format. Read [SPEC] and [BUG] blocks for authoritative facts about programming t8live and running/developing this repo. Read [NOTE] blocks only if additional context is needed. [?] blocks are unverified — treat with lower confidence and confirm before relying on them.

---

1. What is this

[SPEC]

• t8live is a full fork of Strudel, a browser-based live-coding pattern language (a JS port of TidalCycles)
• Adds first-class support for the Roland AIRA Compact T-8 beat machine and the AIRA Compact S-1 synth over Web MIDI
• Primary repo: codeberg.org/catcam/t8live — GitHub is a read-only mirror (see [NOTE] below)
• License: AGPL-3.0-or-later, same as upstream Strudel
• Domain: t8live.fyi (hosting target not yet decided)

[NOTE] Strudel's own maintainers ask contributors not to fork the project back to GitHub ("we have moved from Microsoft's Github platform to Codeberg for ethical reasons"). This project respects that: codeberg.org/catcam/t8live is the canonical repo, github.com/catcam/t8live exists only as a convenience mirror. All credit for the underlying pattern engine, mini-notation language, and audio engine belongs to the Strudel/TidalCycles project and contributors.

---

2. Running locally

[SPEC]

corepack prepare pnpm@9 --activate   # or npm i -g pnpm; corepack's `latest` needs Node >=22.13
pnpm i
pnpm dev                              # serves the REPL, default port 4321

• Requires Node >=18 (developed against Node 20.20.2)
• Requires a checkout fully owned by the user running pnpm — see [BUG] below if pnpm i fails with EACCES

[?]

• Exact minimum Node/pnpm version compatibility beyond what's been tested here is unverified.

---

3. Writing T-8 patterns — the @strudel/t8 API

[SPEC] Package: packages/t8/t8.mjs, globally available in the REPL (registered via evalScope in packages/repl/prebake.mjs and website/src/repl/util.mjs) — no import needed in your pattern.

Function	Purpose	Channel
t8drum(pattern)	Maps voice names to T-8 rhythm notes	10
t8bass(pattern)	Sets T-8 bass channel (note names pass through unchanged)	2
t8select(bank, pattern)	Program Change to select a T-8 pattern slot (bank 1-4, pattern 1-16)	16
t8clock(ticksPerCycle=48)	Continuous MIDI Clock stream	n/a
t8transport(startStopPattern='<start stop>/2', ticksPerCycle=48)	Clock + Start/Stop, stacked	n/a

t8drum voice names → MIDI notes: bd=36, sd=38, clap=50, tom=47, ch=42, oh=46. An unknown voice name errors (logged, query returns no haps) instead of silently sending a wrong note.

All of the above still need .midi('T-8 MIDI IN') appended to actually send anything.

// trigger drum voices by name
t8drum("bd ~ sd ~ bd bd sd ~").midi('T-8 MIDI IN')

// bass line -- standard note names work directly, no mapping needed
note("c2 ~ eb3 ~").t8bass().midi('T-8 MIDI IN')

// select bank 4 pattern 15 on the device
t8select(4, 15).midi('T-8 MIDI IN')

// drive the T-8's own transport (device menu: SYnC must be set to AUTO)
t8transport("<start stop>/4").midi('T-8 MIDI IN')

[NOTE] The T-8 has 6 real rhythm voices. The .prm pattern-file format used by the companion roland-t8 project has extra LT/HT/CY columns that don't map to independent MIDI notes on this unit, so t8drum intentionally exposes only the 6 confirmed voices. Bass note numbering on the T-8 happens to line up with standard MIDI note numbers, which is why t8bass needs no name-mapping step, unlike t8drum.

---

4. Writing S-1 patterns — the @strudel/s1 API

[SPEC] Package: packages/s1/s1.mjs, globally available in the REPL (registered via evalScope in packages/repl/prebake.mjs and website/src/repl/util.mjs, same mechanism as @strudel/t8) — no import needed in your pattern. Full sourced research/implementation writeup, including hardware-test results and open questions, lives in docs/s1-implementation-plan.md.

Function	Purpose	Channel
s1note(pattern)	Sets the S-1's synth channel on a note pattern (standard note names/numbers pass through unchanged)	3
s1cc(name | ccNumber, valuePattern)	Named-parameter CC helper, e.g. s1cc('cutoff', ...) instead of .ccn(74).ccv(...) — full name table in S1_CC	n/a
s1polyMode(valuePattern)	Raw 0-1 passthrough for CC 80 (POLY MODE) — deliberately not a named enum, see caveat below	n/a
s1chord({voice2, voice3, voice4, voice2Shift, voice3Shift, voice4Shift})	Composite helper for the four chord-voice CCs (81-83 switches, 85-87 key-shifts) instead of six separate s1cc calls	n/a
s1select(patch)	Program Change to select an S-1 patch slot (flat 0-63 range, no bank split unlike t8select)	16
s1clock(ticksPerCycle=48)	Continuous MIDI Clock stream	n/a
s1transport(startStopPattern='<start stop>/2', ticksPerCycle=48)	Clock + Start/Stop, stacked	n/a

Unlike the T-8, the S-1 is up to 4-voice polyphonic (Poly/Mono/Unison/Chord, switchable via s1polyMode/CC 80) — stacked/chorded note patterns are a first-class case for s1note, not just a monophonic bassline. All of the above still need .midi('S-1 MIDI IN') appended to actually send anything.

// a chord, sent to the S-1's synth channel
note("<[c3,eb3,g3] [c3,f3,ab3]>").s1note().midi('S-1 MIDI IN')

// sweep the filter cutoff
s1cc('cutoff', sine.slow(4)).midichan(3).midi('S-1 MIDI IN')

// set chord-voice 2 on with a key shift, leave 3/4 alone
s1chord({ voice2: true, voice2Shift: 0.6 }).midichan(3).midi('S-1 MIDI IN')

// select patch slot 12
s1select(12).midi('S-1 MIDI IN')

// drive the S-1's own transport -- unlike the T-8, this needs no on-device
// mode toggle for Clock/Start/Stop to work (confirmed against real hardware)
s1transport("<start stop>/4").midi('S-1 MIDI IN')

[NOTE] Unlike t8bass, s1note needs no note-name-to-MIDI-note mapping table — the S-1's own MIDI chart confirms a true 0-127 voice range with no internal remapping. The bulk of the CC table (§1.3 of the plan doc) is directly sourced from Roland's official chart, cross-checked against the PDF version of the manual (the HTML version's Transmitted/Recognized columns didn't survive scraping cleanly; the PDF did).

[BUG-LIKE CAVEAT] s1cc('pulseWidth', ...) (CC 15, PWM depth) does not audibly work Roughly 31 of the S-1's ~54 documented CCs are only reachable on the physical panel via a [SHIFT]+knob/pad gesture — the MIDI chart lists them as ordinary CC numbers with no modifier byte, but a community Pure Data project's author previously reported difficulty getting SHIFT-combo CCs working at all. Hardware-tested here (2026-07-31, spectral analysis via the S-1 audio bridge): CC 15 (OSC PULSE WIDTH / PWM depth) produces no measurable audible change between depth=0 and depth=127, in two independent test configurations, even with the LFO explicitly routed toward the oscillator first to rule out an unrouted-modulation confound — while a known-good non-SHIFT CC (74, filter cutoff) and even a modest non-SHIFT modulation-depth CC (13, OSC LFO) both clearly worked in the same setup. Treat any other SHIFT-combo CC in S1_CC with the same suspicion until individually verified — see docs/s1-implementation-plan.md §9 item 4 for the full list and methodology. CC 5 (PORTAMENTO TIME, also SHIFT-combo) was also tested but the result was genuinely inconclusive (pitch-tracking limitations, not a hardware finding either way) — left honestly unresolved rather than guessed at.

---

5. Known bugs (fixed in this fork — don't re-break these)

[BUG] .midi() sends an unwanted Start message on every cycle-0 hap Symptom: pressing play on any pattern using .midi() — even one that never touches midicmd — makes the T-8 start playing its own currently-selected pattern at the same time as whatever Strudel is triggering live. Two patterns clash audibly. Cause: upstream packages/midi/midi.mjs's onTrigger handler unconditionally calls device.sendStart() whenever a hap's cycle begins at 0. On the T-8 (SYnC=AUTO), a bare Start with no accompanying clock stream makes the device fall back to its own internal clock and start its own sequencer. Fix: gated behind a new autostart option on .midi(port, options), defaulting to false in this fork. Intentional transport control still works via t8transport(), which sends a real continuous clock alongside Start/Stop. Pass { autostart: true } to restore the old unconditional behavior if some other (non-T-8) device ever needs it.

[BUG] "Your Browser does not support WebMIDI" in Chrome Symptom: Chrome — which does support Web MIDI — throws this error anyway. Cause: navigator.requestMIDIAccess is only exposed in a secure context (HTTPS or localhost). A plain http://<tailscale-ip>:4321 origin doesn't qualify, even though the connection works fine for everything else served from that origin. Fix: the browser must load the REPL from localhost (or a real HTTPS origin once deployed). During remote development, a reverse SSH tunnel from the dev-server host to the machine running the browser (ssh -N -R 4321:localhost:4321 user@browser-host) makes http://localhost:4321 on that machine reach the remote dev server.

[BUG] pnpm i fails with EACCES creating node_modules Symptom: pnpm i fails trying to symlink into a package's node_modules directory. Cause: a file or directory somewhere in the repo was created by a different OS user (e.g. root instead of whichever user owns the checkout), flipping ownership on that subtree. Fix: chown -R <repo-owner>:<repo-owner> the whole repo, then retry.

[BUG] Edited a component but the browser doesn't reflect the change Symptom: hot-reload log shows the file was picked up, but the running app (via a tunnel, a bookmark, or otherwise) still shows the old behavior. Cause: an earlier pnpm dev process was never actually killed (e.g. a pkill pattern didn't match the real command line), so a second instance started and silently bound to the next free port (4322) instead of 4321 — whatever's still listening on 4321 is the stale one. Fix: check lsof -i :4321 / ps aux | grep astro for more than one matching process before assuming a code change didn't take effect. Kill all of them, then start exactly one.

---

6. Workflow notes for this repo

[SPEC]

• Primary remote: origin → codeberg.org/catcam/t8live. Push here first, always.
• Mirror remote: github-mirror → github.com/catcam/t8live. Convenience only — push after origin, never instead of it.
• upstream remote → codeberg.org/uzu/strudel (the original project, for pulling updates).
• Full outstanding-work tracker: see TODO.md in this repo — not duplicated here.

[NOTE] This README stays focused on "how do I write T-8 patterns" and "what's already been fixed and why." TODO.md is the living list of what's left, since that changes faster than a README should.

---

Forked from Strudel (AGPL-3.0-or-later) — all credit for the pattern engine, mini-notation language, and audio engine goes to the Strudel/TidalCycles project and contributors.