An LLM agent can call this like a tiny music tool: pass a game idea or seed, infer a vibe,
and ship a deterministic soundtrack immediately. Our game factory uses it in every HTML5 game:
the baseline engine is zero-asset - no licensed tracks, no samples, no API dependency
(flagship builds add code-rendered tracks). Tap once below and it plays; then open the game built on it.
1 · The live engine (tap once - it plays in this page)
This exact WebAudio engine ships inside every game we publish. Type a game brief and the
agent-facing mapper picks the vibe; a per-seed musical genome picks key, chord progression, melody contour and tempo,
so 170+ shipped games each play a different song from the same 300 lines of code.
loading engine...
tap anywhere to start the audio (browser autoplay rule)
2 · Bandlings: the engine became a game
Merge game where your monster collection IS the music.
Every creature plays a looping part on one shared tempo grid: tier 1 drums, tier 2 bass,
tier 3 chords, tier 4 lead, tier 5 choir. Merging upgrades both the monster and its
instrument, so the song literally grows as you play. Per-player seed: your song is yours.
You can export it as a WAV from inside the game.
3 · The code-DAW (offline renders, same idea, bigger sound)
For flagship builds a 500-line numpy/scipy "code-DAW" composes and masters a full track:
band-limited oscillators, FM e-piano, synthesized drums, convolution reverb, sidechain, bus
compression. No model, no API, free local CPU. These two are real in-game tracks:
synthwave · A minor · 116 BPM · ships in Crash Buggylofi · F major 7ths · 76 BPM swung · ships in Idle Noodle Bar
4 · Even this demo's voice is generated
The event announcer below is ElevenLabs TTS (Creator tier, courtesy of today's event) -
the one piece of audio here that a model made. Everything else: pure code.