> ## Documentation Index
> Fetch the complete documentation index at: https://docs.unpod.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Playground

> Pick an agent, talk to it in the browser, and watch the live transcript - test a playbook end to end before attaching it to a pipe or number.

## What the Playground is

The **Playground** is the no-code way to test an agent. You pick an agent,
talk to it in the browser, and watch the transcript update live as the
conversation advances - no phone number, no pipe, and no client code required.

It runs the same stack production does: the agent's flow is authored with
**SuperDialog** (playbooks), and the call is served by the **unpod**
`AgentRunner`. What you hear in the Playground is what a caller would hear.

<Card title="Open the Playground" icon="play" href="https://superdialog.unpod.ai/playground?tab=preview">
  Launch the live Playground at `superdialog.unpod.ai/playground` and talk to an agent in your browser.
</Card>

<Note>
  Use the Playground to iterate on a playbook (persona, checkpoints, advance
  rules) and confirm it behaves, *then* attach the agent to a
  [Speech Pipe](/speech-stack/pipes) and a [number](/speech-stack/numbers) to take
  it live.
</Note>

## How a Playground session works

<Steps>
  <Step title="Pick an agent">
    Select the agent (playbook) you want to test from the Playground screen.
  </Step>

  <Step title="Connect">
    The browser opens a session and streams your microphone audio to the
    speech stack (16 kHz up); synthesized replies stream back (24 kHz down).
  </Step>

  <Step title="Talk">
    Speak as a caller would. Audio rides the WebSocket bridge; the agent's
    `AgentRunner` handles each turn exactly as it would on a real call.
  </Step>

  <Step title="Watch the transcript">
    Per-turn transcript and the current checkpoint stream to the browser over a
    side channel fed by `Session` hooks - separate from the audio path, so the
    transcript stays accurate even under load.
  </Step>
</Steps>

## What you can test here

* The persona and voice style your callers hear
* Whether checkpoints advance on the conditions you wrote (`done_when` /
  `advance_when`)
* Slot collection - does the agent capture what it needs before moving on
* Interrupts (goodbye, busy) firing from any step

## What is not in the Playground yet

The Playground ships in milestones. The current build is single-agent over the
WebSocket transport: connect, talk, live transcript. Live metrics, the flow
graph view, in-call controls, the LiveKit transport, and side-by-side agent
compare arrive in later milestones.

## Next steps

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/superdialog/quickstart">
    Generate your first playbook, then test it here.
  </Card>

  <Card title="Playbooks" icon="book" href="/superdialog/playbooks">
    Author the persona, checkpoints, and advance rules you test in the Playground.
  </Card>

  <Card title="Speech Pipe" icon="bot" href="/speech-stack/pipes">
    Once it behaves, attach the agent to a pipe to go live.
  </Card>

  <Card title="Sessions" icon="database" href="/superdialog/sessions">
    Persist and multiplex conversations beyond a single test.
  </Card>

  <Card title="Open the Playground" icon="play" href="https://superdialog.unpod.ai/playground?tab=preview">
    Jump straight into the live Playground and start talking to an agent.
  </Card>
</CardGroup>