About vs. Demo: You are on the interactive demo. Use the About Demo page for learning objectives, framework comparisons, and discussion prompts.
Teacher cue: Watch the Agent Trace panel — it shows the same observe → decide → act loop that production browser-agent frameworks use. Try the same command in Rule-based mode and then On-device AI mode to see the gap directly.

Agentic UI Control Lab

AI Page Agent

Type a command in plain English and watch an in-page agent read the support ticket queue as text, decide which action matches, and act on the real UI — the same "text-based DOM control, no screenshots" pattern used by frameworks like page-agent.

Browser-only No API keys Rule-based by default — optional on-device AI
Agent state Idle

Type a command and click Run Agent.

Interpreter Rule-based

Regex parser — switch to on-device AI below to try a real model.

Actions taken 0

Successful actions executed on the ticket queue this session.

Guardrail On

Destructive actions (resolve, escalate) pause for confirmation before they run.

Command the Agent

Require confirmation before destructive actions
On-device AI status
Not loaded — select "On-device AI" above to check availability.

Support Ticket Queue

These controls are real — you could click them yourself. The point of the demo is that the agent can operate them for you, reading exactly what's below as its only source of truth.

Agent Trace

Every run logs the same four steps a real browser-agent framework follows: observe the page as text, decide on an action, act on the DOM, and report the result.

Ready. Run a command to see the agent's observe → decide → act → result trace here.