# Actions

> Run speech, hang-ups and your own code before or after the LLM responds at a node.

Source: https://polvallverdu.github.io/pipedog/docs/flows/actions/



Actions are steps a node runs around the LLM's turn. List them in `preActions` (before the LLM
responds) or `postActions` (after the bot has finished speaking). Each is an `ActionConfig` with a
`type`.

| Type               | Does                                             | Fields                                 |
| ------------------ | ------------------------------------------------ | -------------------------------------- |
| `tts_say`          | Speaks a fixed line.                             | `text`, optional `appendTextToContext` |
| `end_conversation` | Ends the pipeline, optionally after a last line. | optional `text`                        |
| `function`         | Runs your handler inline in the pipeline.        | `handler`                              |
| anything else      | A custom action you register.                    | `handler` and any extra fields         |

```ts
const actionsNode: NodeConfig = {
  taskMessages: [{ role: "developer", content: "Confirm the booking." }],
  preActions: [{ type: "tts_say", text: "One moment while I check." }],
  postActions: [
    {
      // A custom action: the handler receives the action and the flow manager.
      type: "notify",
      channel: "#bookings",
      handler: async (action, fm) => {
        console.log(`notify ${String(action.channel)}: ${JSON.stringify(fm.state.slot)}`);
      },
    },
    { type: "end_conversation", text: "Goodbye!" },
  ],
};
```

A handler is `(action, flowManager) => void | Promise<void>` and receives the whole action object,
so extra fields such as `channel` above reach it.

## Registering custom actions [#registering-custom-actions]

A custom action needs a handler. Put it on the action, or register it once on the manager:

```ts
flowManager.registerAction("notify", async (action, fm) => {
  console.log(`notify ${String(action.channel)}: ${JSON.stringify(fm.state)}`);
});
```

Post-actions wait until the bot has finished speaking, so with a TTS service in the pipeline they run
after the response is heard. `end_conversation` stops the remaining actions.