Keyframery

Agent view: see what AI agents see

Every Keyframery component page has Agent view: the live UI SDK snapshot an AI agent reads, the actions it can run, and the component responding.

People see your design. An AI agent reads something else. UI SDK gives agents a snapshot of the page: each component's label, its state and the actions it accepts. Agent view puts that snapshot beside the live component, on every component page.

Turn on the Agent view switch above the preview:

Live preview

See it as an AI agent does.

Resize preview. Focus the handle and use the left or right arrow keys. Hold Shift for larger steps.— px
Customize1 controls
Animation speed1×

Reading it

Parts lists every component an agent can see, one row each, in plain words:

  • The name an agent finds it by, such as Email, with its kind beside it (Input, Select, Form).
  • Its state, such as Empty · Required or Value “EUR” · Closed · 4 choices. A value that just changed lights up, whether you or an agent changed it.
  • What an agent can do with it, as buttons: Set value, Clear, Open, Submit. A part with no buttons is read-only.

Hover a row and the part it describes is outlined in the preview. On a phone, tap its name.

What the snapshot shows

JSON shows the same thing exactly as an agent receives it. Each entry is one component:

{
  "ref": "c17",
  "component": "input",
  "actions": ["setValue", "clear"],
  "label": "Email",
  "value": "",
  "required": true
}
  • ref names the component so an action can target it.
  • component is the kind, such as input, select or form.
  • label is the name an agent uses to find it.
  • State such as value, checked, open or disabled follows the component as it changes, whether a person or an agent changed it.
  • actions lists what an agent can do. A disabled or read-only component lists none.

Password values read as [redacted]. Hidden and decorative parts stay out of the snapshot, so a component can expose nothing at all.

Run an action

Press an action's button. Clear, Open or Submit run at once; Set value asks what to enter, or offers a Select's choices to pick from. The action runs on the live component through UI SDK, the same way an agent's call would. Under the row you see the exact call, such as setValue("[email protected]"), and what it changed. The full tool response is under JSON.

On the form above, press Set value on each field and enter a name, a phone number and an email, then press Submit on the form, and the viewing is booked, exactly as if a person had filled it in.

For a Select or a Combobox, See choices reads the options it offers, including ones that aren't in the snapshot.

In your own app

Agent view is a docs tool; it doesn't ship with the components. Agents reach your app through UI SDK's agent bridge, which you create once in the browser with createAgent from @ui-sdk/react/agent. Read Agents in UI SDK for the bridge, its allow check, and how to publish its tools to an agent.

Questions

Does Agent view need an AI model or an API key?

No. It reads UI SDK's agent bridge in your browser, the same data an agent would get. Nothing is sent anywhere.

Is this what every AI agent sees?

It is what an agent sees through UI SDK's agent bridge. Agents that work from screenshots see pixels instead, so their view is different.

Does Agent view ship with the components I install?

No. It is a docs tool for inspecting a component. What it shows comes from UI SDK, which every Keyframery component renders underneath.

On this page