Your whole identity in a fraction of the tokens.
Raw CSS is noisy and enormous. Livery packs a system into a compact, agent-native format and serves only the slots a task needs, so agents spend their context on your product.
Brand & Design System Management for AI Agents
Agents write most new UI now, and left alone they write it generically. Every session invents a new radius, a new green, a new gradient. Livery gives them your system before they write a line.
Point Livery at a live site, a GitHub repo, a folder of screenshots or a one-line brief. It reads every style, sorts them into roles and hands you a draft in about three minutes.
Every slot carries its source and a confidence score. The least certain come first. Press Return to accept, or take everything above 80% at once.
Change the brand colour and every component on the board follows. Contrast is solved as you move. Then ask the assistant for a dark mode and see it cut in across everything at once.
Livery serves the system over MCP, a CLI and plain files. The agent pulls only the tokens, components and rules the task needs, writes on brand, and checks its own work.
The system writes its own checks. Raw hex, off-scale spacing, the wrong radius, a missing focus state: caught in the agent's code and sent straight back with the fix.
Published versions never change, so a repo always knows what it runs. Release a fix, an update or a breaking change, and Tailwind, SwiftUI, Compose, CLAUDE.md, Cursor rules and tokens all move together.
A design system an agent can actually use is not a style guide. It is a format, a resolver and a set of checks, all built in-house for one job: the same on-brand result, in as few tokens as possible.
Raw CSS is noisy and enormous. Livery packs a system into a compact, agent-native format and serves only the slots a task needs, so agents spend their context on your product.
Our own extraction and clustering engine sorts raw colours into roles, fits ramps in OKLCH, infers the type scale and spacing grid, and solves contrast. Software built for this job, not a prompt around a screenshot.
Agents resolve tokens instead of improvising values, and checks generated from your system catch whatever slips: raw hex, raw pixels, the wrong radius, missing states.
Each token records where it came from and how sure Livery is. You review the uncertain ones first, and nothing guessed passes as certain.
Every system gets its own live field, generated from its own palette. Browse the gallery, fork one in a click, and make it yours.
Two or three apps live, designing by prompting. One system keeps every session on brand.
Spin one up from the client's site, then keep every contractor's agent on brand.
Define the identity once. Get Tailwind, SwiftUI and Compose that never drift apart.
A live website, a GitHub repo (Tailwind, CSS variables and theme files), a set of screenshots or a moodboard, or a short description. You can also start blank.
Claude Code, Cursor and Codex, over a hosted or local MCP server, a CLI, or plain files like DESIGN.md, CLAUDE.md, AGENTS.md, skills and Cursor rules.
Tailwind v4, shadcn, SwiftUI, Compose and W3C design tokens, generated from the same system, so every platform agrees.
Livery generates checks from your system. Agents call them while they work, and they flag raw hex, raw pixel values, the wrong radius and missing states, then send the fix back.
Published versions are immutable. You make a new version, see exactly what changed, and each repo either follows updates or stays pinned.
Early access goes out in batches, in waitlist order. Join now and you'll get one email when it's your turn.
Be first to give every agent you use the same design system. Early access goes out in batches.