The first design system built for
AI-native apps.
184 Web Components. 89 AI-native patterns. 6 cognitive bias primitives. Generate UI via LLMs with typed component trees and behavioral psychology baked in.
Six capabilities no other library ships
Tracks 1–5 complete. 5 innovation packages. 10/10 across code health, quality, architecture and innovation.
AI-Native Foundation
89 components designed for LLM UIs: streaming text, reasoning trees, citations, agent cards, cost dashboards, guardrails. Purpose-built — not adapted.
Generative UI Runtime
Zod schemas let LLMs output structured component trees. The MCP server exposes 6 tools to Claude, Cursor and Windsurf so agents can compose UI directly.
Cognitive Psychology Built-In
184 behavioral bias cards plus 6 composable bias wrappers — anchoring, scarcity, social proof, authority, commitment, reciprocity — auditable at build time.
3-Tier Design Tokens
1,800+ tokens across core → semantic → component. ESLint plugin blocks raw hex, rgba(), magic px and fake tokens. Zero drift, by construction.
SSR + Tree-Shaking
Ship 8 kB gzipped for a single component, 246 kB for the full library. @lit-labs/ssr with declarative Shadow DOM eliminates FOUC on first paint.
Theme On Demand
Run `npx cognivo-theme "warm ocean professional"` and get a production-ready 7.4 kB token JSON — palette, surfaces, dark-mode pairs, all validated.
Server-render a gen-UI tree in seven lines
The same Lit template runs in the browser, streams from an LLM, or renders to HTML on the server.
Runs where you already work
Standards-based Web Components. First-class adapters for every major framework and agent runtime.
- React
- Vue
- Next.js
- Astro
- Remix
- Lit
- Claude
- Cursor
- Windsurf
Why another component library?
Because none of them were built for what's shipping right now.
Every other library
- Generic buttons, inputs, modals
- Static tokens, no semantic layer enforcement
- No AI-specific primitives
- No streaming support
- No cognitive-bias model
- You build the AI layer yourself
Cognivo
- 89 AI-native components (chat, reasoning trees, agents, costs)
- 3-tier tokens, ESLint-enforced — zero drift
- Streaming gen-UI: JSON-in, live components out
- SSR + declarative Shadow DOM — no FOUC
- 184 biases + 6 composable bias wrappers
- MCP server: Claude / Cursor / Windsurf integrate natively
Production-ready. MIT. No lock-in.
Generate bias-aware UI with your favorite LLM today.