Generative interface

Build generative UI with Jev + Signals + json-render.

Your generative interface can adapt to what each visitor is doing right now. Signals serves that state in 6ms. Jev picks your components and json-render draws them.

  • 6ms p50 read, in-region
  • 1 prompt to your coding agent
Teams running on Snowplow
Experian
1Password
Strava
HelloFresh
AutoTrader
Supercell

The model below is deciding about your visit as you read.

A model, about you

connecting
What a System One model decides about you and your visit, right now

Signals is computing your visit from the events, we will ask Jev about it next.

Browse around the site, come back, and watch both answers move.

Who · Jev's confidence in each rolewaiting for the first answerWhat · Jev's confidence in each intentwaiting for the first answer
Keyed on an anonymous session id. Jev is asked again each time your visit changes.
Why generative UI needs live state

A page that composes itself needs to know who it is composing for.

Before

The spec describes the app, not the visitor

Route, props, cart, maybe a profile computed before the visit. The model picks components for the average visitor because that is all the state describes, and the page comes out the same for everyone, the way a template does.

state = { route, props }
With Signals

The visitor's session, in the state

Add a Snowplow SDK and events flow in. Name the activity the composition needs and Signals returns it as JSON, current as of this click, ready to sit beside the route in the state object.

getAgenticContext()
Result

The spec changes with the visitor

Jev picks the compare panel for someone weighing two products and the reassurance card for someone stalled on price. json-render draws whichever it chose, from components you already have.

spec = compose(answers)
What you build with it

Pages that compose themselves around the visitor.

Each slot in the spec is one typed question. The catalog stays yours, so the model can only pick from components you wrote, and every pick comes with a probability you can read.

Lead with what they came for

Which section leads: the code, the numbers, the comparison or the price. Picked against what this visitor has read and skipped on this visit, and drawn from your catalog.

choice: which section leads?

Show proof in the shape they trust

A code sample for the engineer, an outcome for the PM, an assurance for the person who has been on the security page. The same catalog, a different component in the slot.

choice: code, number, outcome or assurance?

Answer the objection before it is typed

Someone who opened pricing twice and went back is asking about cost. The spec gets the pricing answer as a callout, not a link they have to find.

choice: which objection, or none?

Change the catalog, not the model

Add a component, describe it in the catalog, and the model can pick it on the next render. No retrain, no prompt engineering, one pull request.

edit the catalog, reload
How it works

State, decision, spec, render.

Signals computes what this visitor is doing. Jev does the choosing. json-render does the drawing, from a catalog only you can add to. The one-liner at the bottom of this page teaches your coding agent the whole loop.

  1. 01Collect. Add a Snowplow SDK to your app and behavioral events start flowing in, validated on the way.
  2. 02Read. One call returns what this visitor has been doing as JSON, <1s after the event, in 6ms p50.
  3. 03Choose. Pass it as the state on your Jev call, one typed question per slot in the page.
  4. 04Render. Turn the answers into a json-render spec and let it draw the components from your catalog.
Signals → your app6ms p50
namepage_session
formatjson
eventslast 30 this session
sections_read4
pricing_views2
Trigger objection_visibledelivered to your app
A page_session state showing four sections read and two pricing views goes to Jev, which answers which objection to address with pricing at 0.86, and json-render draws the PricingCallout component from the team's own catalog.
Works with
  • Jev
  • json-render by Vercel
  • CopilotKit
Demo

See it in action.

A storefront with no page designed in advance. The shopper browses, Signals keeps the session, Jev picks what goes in each slot, and json-render draws it from the catalog.

Silent recording, captioned on screen. The panel on the right is the state Signals served and the choices the model made from it.1 min 19 sec
Questions

Before you try it

Is this a tool that designs screens for me?

No. If you want an app that turns a prompt into a mockup or a finished screen, this is not it. Signals is for teams whose own app composes its pages at request time, from components they wrote, around what each visitor is doing.

What is json-render?

A library from Vercel for generative UI. You describe a catalog of components and their props, a model produces a JSON spec that uses only those components, and json-render draws it. The model never writes markup, so the page cannot contain anything you did not ship.

Is Signals a generative UI framework?

No. Signals renders nothing and picks nothing. It gives the framework you use, whether json-render, CopilotKit or the Vercel AI SDK, what this visitor is doing right now as JSON. The model picks the components, from a catalog that stays yours. json-render is on this page because it is the shortest path from Jev's answers to a rendered page.

Does this only work with Jev?

No. The context comes back as JSON over the REST API, the Node.js or Python SDK, or the MCP server, and as a plain string in narrative form. Anything that takes a state object or a prompt can read it. Jev is the model this page is written for because one typed question per slot, on every render, is what it is built to answer.

Do I need an event stream already?

No. Collection is part of Signals. Add a Snowplow SDK to your web, mobile or server app and events start flowing in. If you already run Snowplow, Signals reads the pipeline you have.

What does the free tier include?

The full product against your own traffic, up to 5 million events a month, with no card and no sales call. You should see an attribute updating against your own traffic in the first session.

Let your next page compose itself in minutes.

Paste the one-liner and the prompt below into your coding agent. The panel at the top of this page is Jev deciding about you as you read, from the state Signals computed.

The real engine · no card · no sales call
Paste this into your coding agent
npx plugins add snowplow/skills
Then: Add Snowplow Signals to this app. Find the page that should compose itself around the visitor. Define an agentic context for what they have been doing this session, read it as JSON and pass it as the state on one Jev typed question per slot. Turn the answers into a json-render spec against our component catalog and render it.