HN Showcase
Back to browse
Daub – A rendering spec for AI-generated UIs (two files, no build step)

Daub – A rendering spec for AI-generated UIs (two files, no build step)

by kulesh·Mar 11, 2026·2 points·1 comment
Visit ProjectView on HN

AI Analysis

●●●BangerBig BrainZero to OneBold Bet

AI generates JSON specs instead of code — MCP server integration included.

Strengths
  • •JSON spec means AI doesn't hallucinate broken component imports
  • •MCP server lets any AI assistant generate complete UIs via conversation
  • •Zero build step with CDN delivery — two files and you're rendering
Weaknesses
  • •AI UI generation is crowded — v0, Bolt, and Builder.io already exist
  • •Adoption depends on AI tool integrations beyond the MCP server
Target Audience

Developers building AI agents that generate UIs

Similar To

v0.dev · Bolt.new · Builder.io

Post Description

Most UI frameworks solve the wrong problem — they help developers build interfaces faster. DAUB bets on the next problem: AI generating interfaces directly.

The core idea: instead of writing component code, you describe a UI in structured JSON. DAUB renders it. No compilation, no virtual DOM diffing. The spec is the UI.

What shipped so far (v3.13): - daub.css + daub.js: 76 components, 20 theme families, two CDN links, zero build step - JSON-Render spec: AI outputs structured JSON, DAUB renders a live interface — already working in the Playground - 7-stage generation pipeline: analyze → scaffold → generate → selfCheck → verify → repair → visual diff - 189 pre-made blocks across 34 categories with multimodal RAG retrieval - MCP server on Cloudflare edge: generate_ui, render_spec, validate_spec, get_component_catalog — any MCP client (Claude, Cursor, custom agents) can generate DAUB interfaces without writing a line of component code

What's coming: - v3.14: Action bindings — buttons that POST to real endpoints, forms with submit handlers, shareable URLs, version history. Generated UI becomes functional and outlives the session. - v4.0 (Intent Engine, in progress): "I need to track my expenses" → spending dashboard with charts, filters, input form. Context-aware: location, time, history inform the output. - v5.0 (Agent Runtime): multi-agent orchestration where one agent fetches data, another picks layout, a third wires actions. The interface is a byproduct of understanding, not a thing you build.

The underlying bet: as AI generates more of the interface layer, value accrues to whoever owns the rendering spec that agents target. DAUB wants that spec to be small, inspectable, text-diffable, and easy for both humans and models to work with — not locked inside a framework.

It's not competing with React or shadcn/ui. It's the layer underneath whatever generates the UI.

Playground: https://daub.dev/playground.html Roadmap: https://daub.dev/roadmap MCP server + GitHub: https://github.com/sliday/daub

Happy to get into the spec format, the complexity-routing pipeline, or how the MCP integration actually works.

Similar Projects