> ## 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.

# Chat

> Your trained agent, over an OpenAI-compatible chat API. One URL swap.

Like chat completions - except the model is your trained agent. A playbook +
a small model trained on your use case, or any third-party model, served as
one OpenAI-compatible endpoint. Text in, text out.

## How it connects

| Your stack                          | You swap                              | Unpod runs              |
| ----------------------------------- | ------------------------------------- | ----------------------- |
| LiveKit · Pipecat · Vapi · chat app | the LLM base URL (or the `llm=` slot) | agent, sessions, memory |

## Go live in 3 steps

### 1. Get an endpoint

Build and publish a playbook in the [Playground](https://superdialog.unpod.ai/playground),
then open **Deploy as Endpoint → Manage API Keys**. You get the endpoint, a
model id (`public:PB_...`), and prefilled snippets. See
[Publish & Share](/playbook/publish-and-share).

### 2. Point your stack at it

<Tabs>
  <Tab title="Any OpenAI client">
    ```bash theme={null}
    curl -X POST "https://inference.unpod.ai/v1/chat/completions" \
      -H "Authorization: Bearer $UNPOD_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{"model":"public:PB_7ZRMzCA1ojQ9LlcK","messages":[{"role":"user","content":"hi"}],"user":"sess_abc"}'
    ```

    Any OpenAI-compatible SDK works - swap base URL, key, and model. `user` is
    your session id, so the conversation stays stateful across calls.
  </Tab>

  <Tab title="LiveKit">
    ```python theme={null}
    from livekit.agents import Agent, AgentSession
    from superdialog import DialogMachine
    from superdialog.adapters.livekit import DialogMachineLLM

    dm = DialogMachine("kyc.yaml", llm="anthropic/claude-haiku-4-5")

    async def entrypoint(ctx):
        await AgentSession().start(agent=Agent(llm=DialogMachineLLM(dm)), room=ctx.room)
    ```

    The agent runs in-process in the `llm=` slot. Full guide:
    [LiveKit embedding](/superdialog/embedding-guides/livekit).
  </Tab>

  <Tab title="Pipecat">
    ```python theme={null}
    from superdialog import DialogMachine
    from superdialog.adapters.pipecat import make_processor

    agent = DialogMachine("kyc.yaml", llm="anthropic/claude-haiku-4-5")
    processor = make_processor(agent)  # place between STT and TTS
    ```

    Full guide: [Pipecat embedding](/superdialog/embedding-guides/pipecat).
  </Tab>
</Tabs>

### 3. Test it

`curl` the endpoint, or talk to the same agent in the
[Playground Preview](https://superdialog.unpod.ai/playground?tab=preview).
Same brain, same behaviour.

## Go deeper

<CardGroup cols={2}>
  <Card title="Chat API reference" icon="terminal" href="/playbook/api">
    Full request/response reference.
  </Card>

  <Card title="Build a playbook" icon="wand-sparkles" href="/playbook/build-an-agent">
    Describe the job; the playbook writes itself.
  </Card>

  <Card title="Embedding guides" icon="plug" href="/superdialog/embedding-guides/overview">
    LiveKit, Pipecat, FastAPI, CLI.
  </Card>

  <Card title="What is a playbook?" icon="book-open" href="/playbook/what-is-a-playbook">
    Persona, checkpoints, slots - the format.
  </Card>
</CardGroup>