Hi Product Hunt, I'm Jacob. I run Prodular, a product design studio.

I kept seeing the same thing in products built with Claude Code and Cursor. The code arrived fast. The interface drifted: every prompt made its own calls on spacing, colour and components, so screen ten never quite matched screen one, and empty, loading and error states only existed when someone remembered to ask for them.

Prodular UI is the system I wanted underneath those products:

162 React components on shadcn/ui and Base UI. 36 of them are for AI: chat, reasoning, tool calls, sources, approvals, agent runs and voice.
21 SaaS screens, each with its empty, loading and error states.
10 themes, light and dark, three densities. The token build fails if any text misses WCAG AA contrast.
A Figma library on the same token names.
Rules for agents. Every component says how to use it: required props, which variant when, what accessibility needs. Your agent reads them over MCP (npx prodular-ui mcp init) or llms.txt, installs what it needs and checks the code it wrote.

Everything installs as source into your repo, so there is no package to work around.

The core, 100 components, is free under MIT. Pro is one payment, from $199 at launch for the first 100 licenses of each plan.

Before launch I built three products on it myself. The 14 newest components came straight from what those products were missing.

One question for you: what does your coding agent still get wrong in UI? That list is what I will work on next.