Sonogram is the audio line’s flagship and the catalog’s multiplier: a sound source that drives your whole shelf. It analyses audio with musically-correct DSP and routes the result through a mapping graph that can set CSS custom properties, paint a built-in canvas visualizer, and — the headline — call update() on any other Caustics engine. Buy it once and your shader-gradient hero pulses to the bass, your flowfield swirls on the beat, your fluid-canvas reacts to the room. A SonogramAPI is an AudioSignalSource, so wiring it to another instrument is a single prop.

The DSP is honest, not a bar-meter dressed up. Bands are A-weighted on a log frequency layout (IEC 61672); envelope followers compute their attack/release coefficients from the real frame dt; onsets use spectral flux against an adaptive median threshold with a refractory gate, so you get one hit per drum, not a splat per frame — and onset strength is measured against a decaying flux peak, so a soft hit reads soft. Tempo is an onset-novelty autocorrelation with a confidence score; loudness is perceptual, not raw RMS. Every channel — bands, loudness, RMS, BPM, BPM confidence, beat phase, onset — is exposed in the domain you want to render in, bindable directly through the mapping graph.

It is built on the shared @caustics/audio-core contract, the one typed SignalFrame every audio-aware engine in the catalog produces and consumes — so the hook is frozen and future-proof, and flowfield’s field: 'audio' already speaks it. Audio is gesture-gated for the browser autoplay policy; mic, <audio>/<video> element, and MediaStream sources are all supported, with the AudioContext resume handled for you. Denied-mic and unsupported states are genuinely inert (no spinning loop). Full lifecycle, SSR-safe, idempotent destroy() that releases the AudioContext and tracks. Ships for Svelte 5, React 18+, and vanilla JS through one mount().