The first thing most people do when they connect Colour Memory to an AI agent is ask it to generate a palette. That is fine. It does that. But it is like buying a Leica and using it to photograph your lunch.
This post is about what Colour Memory actually is, why it exists, and what it can do that no other colour tool does or can do. If you are evaluating it, read this first. If you are already using it, read this anyway.
What people think it does
Most colour APIs do one of three things. They match a hex to a name. They suggest harmonious colours. They check contrast ratios. They are, in essence, lookup tables with an API wrapper. Fast, cheap, useful for narrow problems.
When people see "colour intelligence API" they assume Colour Memory is a more sophisticated version of the same thing. A better palette generator. A smarter colour name lookup. A faster WCAG checker.
It is not. The palette generation is a side effect of something more fundamental.
What it actually is
Colour Memory is the knowledge layer that colour has always needed and never had. It is the answer to a question that RGB cannot ask: what does this colour mean?
Not aesthetically. Historically. Commercially. Materially. Legally. Across cultures, centuries, and contexts. With primary source citations.
The same green reads as growth in one market and mourning in another. The same red means danger in the West and luck in China. Tyrian Purple was punishable by death to wear without imperial authorisation. Scheele's Green killed rooms full of people for fifty years while the Victorians defended it in the press. None of this is in any hex value.
The archive contains 17,346 colours across 44 cultural and material traditions. Every colour is anchored to a documented source: a museum collection, a historical text, a chemical analysis, a court record. The API makes that archive queryable, combinable, and actionable.
The six things it does that no other tool does
1. Brand audit in one call
Send a palette. Get back colour roles with archive names, a full WCAG accessibility matrix, cultural risk per colour across global markets, CSS variables, Tailwind config, and a scored verdict with specific suggestions. No LLM cost. All computed. ChatGPT ran this against the Colour Memory brand palette in May 2026 and found, unprompted, that the palette scored 74/100 and needed a light neutral. It was right. Archive Parchment is now the fifth brand colour.
2. The image generation loop
Most AI image generation is colour-blind. You prompt for a luxury product shot. The model returns something that looks premium. But the colours are generic AI aesthetics, not your specified palette. There was no way to know the difference. Until now.
agent_brief takes your concept and returns an archive-grounded image prompt with specific hex values pulled from real documented sources. verify_agent_palette takes the generated image and scores fidelity with CIEDE2000 distance per colour. We ran this against a Midjourney banana product shot. Fidelity score: 71.2. The model used broadly the right gold register but missed the shadow depth. That is actionable feedback no other tool provides.
3. Accessibility that becomes design guidance
WCAG checkers give you pass/fail. That is not enough to make a decision. accessibility_matrix gives you every pair in your palette at once with all four grades. accessibility_usage_rules goes further: safe pairs, AA-only pairs, large-text-only pairs, decorative-only pairs, and the best text colour for every background in your palette. The output is a design system rule, not a boolean.
4. Developer tokens from archive names
colour_name_slugs takes a hex and returns the archive-grounded name in every format a developer needs: CSS variable, Tailwind class, TypeScript const, kebab-case, camelCase, SCSS variable. --cm-song-gold-sycee: #D4A829 is not just a variable name. It is a citation. Every token carries provenance.
5. Cultural risk before it becomes a problem
The gold in your palette is Song Gold Sycee. In China that reads as imperial. In Buddhist markets it signals the sacred. In Western signage it means warning. None of this stops you using it. It means you use it knowing what you are carrying. cultural_risk_assessment covers twelve major markets with per-colour, per-market warnings, positives, and context-dependent readings.
6. Archive search across 17,346 colours
Query by hex or concept and you get nearest-neighbour matches. But sometimes you want to search differently. archive_search does full-text keyword search across all colour names and notes. Search "prussian" and you get 104 results across six archives, including Prussian Blue Chamber Staining from Pressac's 1989 documentation of Auschwitz. The archive does not sanitise. That is the point.
What this means for you
If you are building an AI product that touches colour, you have two choices. You can let your AI invent colour reasoning from its training data, which means generic psychology, confident hallucination, and no citations. Or you can give it a knowledge layer that was built specifically for this problem.
Colour Memory is the second option. One MCP endpoint. 55 tools. Every answer cited.