Taste Engine · live

A brand isn't a PDF. It's a system.

The Taste Engine turns a brand into rules a renderer can execute: tokens, voice and constraints. Output stays on-brand at agent scale.

How it works

Three steps. The rules do the rest.

01

Encode the brand

Tokens, type, motion and voice become machine-readable rules in one file. The file above is that file.

02

Generate on-brand

The renderer produces interfaces that stay inside those rules, so the output can't drift off-brand.

03

Vary freely

Infinite layouts, one locked identity. The unit is the rule set, not a template.

Constraints

What the renderer refuses, and what it always does.

Never

  • N1Inter, Roboto, or system-font defaults
  • N2Purple-to-blue gradients, or any hue-crossing brand gradient
  • N3Centered-everything symmetry on marketing surfaces
  • N4Uniform border-radius across all elements
  • N5Generic three-column icon-card feature grids
  • N6Card-in-card nesting; every section a same-shadow card
  • N7Shadows above 8% opacity outside overlays
  • N8Gradient text for impact; competing accent hues
  • N9Hedging copy and aspirational vapor
  • N10Asymmetry in tables, forms, or financial confirmations

Always

  • A1Dark-first: near-black base, cream ink
  • A2One gold element per viewport, on the action that matters
  • A3Halo field + grain veil on marketing surfaces
  • A4One easing curve, enter slower than exit
  • A5Mono with tabular-nums for all numeric data
  • A6Respect prefers-reduced-motion
FAQ

What is the Taste Engine?

A set of machine-readable brand rules (tokens, voice and constraints) and a renderer that produces output only inside them. The rules live in one file, atlas/dna.json.

Who builds the Taste Engine?

Purple Squirrel Media, an AI-agent-operated creative and engineering studio.

Is this page generated?

Partly. The token sheet is read from dna.json when the page is built. The copy is written by hand.

Generic output is a choice. Break Free.