01Practitioners' hub · Generative UI
We Build AI Interfaces That Ship.
Generative UI consulting and development for teams that need AI features in production — not demos. Streaming components, tool-use flows, agent workflows, end-to-end delivery.
02Principles
What is generative UI?
Generative UI is interfaces produced by language models at runtime. Components are streamed from LLM output and rendered as real UI, not text. The model picks the form alongside the content: a chart for numbers, a form for input, a list for choices.
The model picks the form
A chart for numbers, a form for input, a list for choices.
Components from your library
The model doesn't write HTML or JSX — it picks from your library.
Streaming, not text
Interactive elements stream in as real UI, not as text.
Changes through the prompt
Changes ship by editing the prompt or the catalog.
Traditional UI
Generative UI
Form chosen by the designer at build time
Form chosen by the model at inference time
Static templates, deterministic outcomes
Runtime components, probabilistic outcomes
Same screen for every user, every request
Different shape per user, per request
Changes ship through a release cycle
Changes ship by editing the prompt or the catalog
03New
Latest
What we published most recently.
What is Generative UI? The Complete Guide
Everything you need to know about AI systems that generate interactive UI components, not just text.Generative UI vs Traditional UI: Key Differences
How generative interfaces differ from conventional UIs and when each approach makes sense.Building Your First Generative UI with Vercel AI SDK
Step-by-step guide to creating your first AI-powered interface with streaming components.04Field guide
Field guide
Six tracks across the generative-UI landscape. Pick by where you are — engineer mid-build, designer mapping patterns, PM evaluating, or executive sizing the bet.
Runtimes & SDKs
Vercel AI SDK, CopilotKit, Tambo, Thesys C1, A2UI — how each one streams components, where the seams are, and what production looks like.
Patterns & components
Interaction patterns for LLM-rendered UI — disclosure, repair flows, tool confirmations, and the design questions GenUI raises.
Designers · EngineersProduction engineering
Observability, performance, testing, accessibility. What it takes to put a streaming, LLM-rendered interface in front of users and keep it up.
EngineersRAG & tool use
Retrieval-augmented generation, tool calling, agent workflows. Patterns that move generative UI past static catalogs into real action.
EngineersState of the field
Market state, vendor landscape, adoption curves. What is and isn't shipping in production right now — for evaluation, scoping, and bets.
PMs · Executives · ConsultantsGetting started
What generative UI is, who it's for, when it makes sense to adopt. The entry-level primer for anyone new to the term.
Written for engineers, designers, product managers, and executives working on or evaluating generative UI. Topic-organised navigation and per-role filters are on the roadmap.
05State
State of generative UI — runtimes at a glance
Five production-grade options, compared on stack, streaming model, openness, and maturity.
| Runtime | Maturity | Summary |
|---|---|---|
| Vercel AI SDK | Mature | Streaming React/Vue components with first-class tool use. The default for new Next.js and Nuxt apps. |
| CopilotKit | Maturing | Drop-in copilot hooks and UI primitives. Strong for adding an assistant to an existing app. |
| Tambo | Early | Component catalog the model picks from. Useful when the design system is the unit of generation. |
| Thesys C1 | Maturing | Hosted API that returns ready-to-render UI. Trades infra ownership for time-to-first-screen. |
| A2UI (Google) | Early | Open JSON schema for renderer-agnostic UI from agents. Cross-runtime, very early. |
Open Source
Streaming markdown that doesn't flicker
Naive innerHTML renderers tear down the DOM on every token — causing flicker, lost focus, and layout thrash. Generative DOM diffs the AST and patches only what changed, delivering smooth, accessible streaming output at up to 90× less DOM work.
usershow Q1 sales
modeltool · revenueChart({range:"Q1",currency:"RUB"})
dom<RevenueChart/> — mounted
dom<FilterPanel/> — mounted
diffonly changed nodes · no flicker
07Open Source
Web components LLMs can actually compose
100+ framework-agnostic tags — chat surfaces, charts, decision matrices, lesson widgets — each carrying a validated meta.json with props, events, slots, themes, and dependencies as machine-readable JSON. The model sees the catalog the same way a developer does, and picks the right primitive for the data it's about to render.
Catalog: 100+ tags · meta.json on each.