COMMUNITY OBSERVEDMetric claims: UNVERIFIEDSOURCE AUDITED🤖 Jev Agent

MatchCN

Semantic component search for coding agents, using Jev probabilities to rank existing UI components before generation.

Overview & result

MatchCN lets a developer describe the UI they need, then uses Jev-backed classifications and confidence signals to search existing components and return an install command.

From the original post

Built matchcn: semantic component search for coding agents, powered by Jev.

Describe the UI you need. It finds the component and returns the install command. No component name needed. No browsing 9 different libraries.

1,783 components across 9 @shadcn registries, classified across six dimensions: purpose, motion, density, interaction, data, and decoration.

Jev returns calibrated probabilities instead of generated text, so every classification comes with confidence. Matching uses that confidence when scoring results. You can see exactly what matched, what didn't, and how much each dimension contributed.

Currently indexing: React Bits @davidhaz, Magic UI @dillionverma, Aceternity @aceternitylabs , Kokonut UI @dorianbaffier, Animate UI @animate_ui, Motion Primitives @Ibelick, assistant-ui @assistantui, bundui @bunduidotio, shadcn-dashboard. (your components, your registries, your install commands, straight from you)

The goal: give coding agents a way to search before they generate.

@matchcndev

http://matchcn.dev

— whosfranki

How Jev fits in the loop

  1. The user supplies a natural-language component request.
  2. Jev helps classify candidate components across bounded design dimensions.
  3. The search layer ranks candidates and returns an existing component for the coding agent to install.

How to reproduce

  1. Open the live demo and inspect the indexed component metadata and result explanations.
  2. Run the same UI request against several registries and record which dimensions drive the match.
  3. Compare retrieval quality with a name-only or keyword-only baseline.

Why this build matters

It demonstrates Jev as a semantic retrieval and ranking layer that helps coding agents reuse existing code instead of generating everything from scratch.

Limitations

  • The reviewed post does not provide a public repository or ground-truth retrieval set.
  • The registry inventory and classifications may change over time.
  • It should be treated as a demo and pattern example until a reproducible evaluation is available.

Patterns