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

# Multiple PALs in one conversation

> Run two PALs in the same room by pointing a second conversation at the first conversation's URL, and use wake and sleep phrases to direct which PAL is speaking.

> [!NOTE]
>
> This page is a draft under review. It documents a deliberately controlled pattern for running more than one PAL in a single session. Reach out to your Tavus account team before using it in production.

## Overview

This guide covers a specific, human-directed setup: **one shared room, one PAL awake at a time, phrase-based handoffs between them.**

The pattern:

1. Create an **anchor conversation** the normal way. Tavus returns a `conversation_url` for the room the first PAL joined.
2. Create a **second conversation** for the second PAL that points at the same room:
   * For a Tavus-managed Daily room, pass the anchor's `conversation_url` as `properties.daily_room`.
   * For Google Meet, Zoom, or Microsoft Teams, pass the same `meeting_url` you used on the anchor.
3. Give each PAL distinct `wake_phrase` and `sleep_phrase` values so only one is answering at a time.
4. Have the human participant join the shared URL.
5. When the session ends, call [End Conversation](https://docs.tavus.io/api-reference/conversations/end-conversation.md) for **every** `conversation_id` you created.

The common example is a **rep-directed handoff**: a sales rep is on the call with a CFO PAL and a security stakeholder PAL, and switches between them by saying each PAL's wake phrase and then its sleep phrase before switching.

This pattern is equally supported across Tavus-managed Daily rooms and [Google Meet, Zoom, and Microsoft Teams](https://docs.tavus.io/sections/conversational-video-interface/pal/meetings.md#join-a-meeting-via-api) meetings. The only difference is which field points the second PAL at the first PAL's room.

## What this guide does not cover

The following are outside the setup demonstrated here:

* **Autonomous interjections.** PALs do not decide on their own to jump in. The human on the call directs turns with wake and sleep phrases.
* **A single merged transcript across PALs.** Each PAL's conversation has its own transcript. Tavus does not currently provide a global, reliably-attributed multi-speaker transcript that combines what every PAL heard and said.
* **Automatic choice of which PAL speaks next.** There is no built-in arbitration between two awake PALs in the same room. If both are awake and both hear a prompt they consider theirs, they can both respond. The wake and sleep phrases are what keep only one active at a time.

## Prerequisites

* Two PALs configured with the [Conversational Flow layer](https://docs.tavus.io/sections/conversational-video-interface/pal/conversational-flow.md), each with its own distinctive `wake_phrase` and `sleep_phrase`. Phrases with two or more distinctive words work best.
* For Google Meet, Zoom, or Microsoft Teams: each PAL must have a [conferencing layer](https://docs.tavus.io/sections/conversational-video-interface/pal/meetings.md) with `username` configured, matching the requirements for a single-PAL meeting join.

## Create the anchor conversation

Create the first conversation with the first PAL. Nothing special is required; Tavus returns the room's URL.

```shell cURL theme={null}
curl --request POST \
  --url https://tavusapi.com/v2/conversations \
  --header 'Content-Type: application/json' \
  --header 'x-api-key: <api_key>' \
  --data '{
  "pal_id": "p22f53b7b4b9",
  "face_id": "r4b03c679d42",
  "conversation_name": "CFO conversation (anchor)",
  "max_participants": 3
}'
```

Response:

```json theme={null}
{
  "conversation_id": "cAAAAAAAAAAA",
  "conversation_url": "https://tavus.daily.co/cAAAAAAAAAAA",
  "status": "active"
}
```

Keep both `conversation_id` and `conversation_url`. You will reuse the URL on the next request and end both conversations later.

For Google Meet / Zoom / Teams, create the anchor the same way you would for a single-PAL join: pass `meeting_url` on the first conversation (see [Join a meeting via API](https://docs.tavus.io/sections/conversational-video-interface/pal/meetings.md#join-a-meeting-via-api)). Keep the same `meeting_url` to reuse on the second conversation.

## Add the second PAL to the same room

Create a second conversation for the second PAL. Point it at the same room the anchor is in.

**Tavus-managed Daily room** — set `properties.daily_room` to the anchor's `conversation_url`:

```shell cURL theme={null}
curl --request POST \
  --url https://tavusapi.com/v2/conversations \
  --header 'Content-Type: application/json' \
  --header 'x-api-key: <api_key>' \
  --data '{
  "pal_id": "pc11223344ff",
  "face_id": "rc55667788aa",
  "conversation_name": "Security stakeholder conversation",
  "max_participants": 3,
  "properties": {
    "daily_room": "https://tavus.daily.co/cAAAAAAAAAAA"
  }
}'
```

Response:

```json theme={null}
{
  "conversation_id": "cBBBBBBBBBBB",
  "conversation_url": "https://tavus.daily.co/cAAAAAAAAAAA",
  "status": "active"
}
```

**Google Meet / Zoom / Microsoft Teams** — pass the **same `meeting_url`** you used on the anchor conversation:

```shell cURL theme={null}
curl --request POST \
  --url https://tavusapi.com/v2/conversations \
  --header 'Content-Type: application/json' \
  --header 'x-api-key: <api_key>' \
  --data '{
  "pal_id": "pc11223344ff",
  "face_id": "rc55667788aa",
  "conversation_name": "Security stakeholder conversation",
  "max_participants": 3,
  "meeting_url": "https://meet.google.com/xgq-epxn-ccp"
}'
```

Both PALs are now in the same room.

> [!NOTE]
>
> Count every participant in `max_participants` on **each** conversation. Two PALs plus one human is three participants, so set `max_participants: 3` on both conversations. The [Participant Limits](https://docs.tavus.io/sections/conversational-video-interface/conversation/customizations/participant-limits.md) page has more detail; faces count as participants.

## Configure wake and sleep phrases

On each PAL, set a distinct `wake_phrase` and `sleep_phrase` on the [Conversational Flow layer](https://docs.tavus.io/sections/conversational-video-interface/pal/conversational-flow.md). Only the awake PAL responds; the other PAL stays silent.

Example pairing:

| PAL | `wake_phrase` | `sleep_phrase` |
| - | - | - |
| CFO | `"Hey Casey"` | `"Thanks Casey"` |
| Security stakeholder | `"Hey Sam"` | `"Thanks Sam"` |

Choose phrases that are unique enough not to over-trigger, and that do not overlap with each other.

> [!NOTE]
>
> A **sleeping PAL continues to receive conversation context.** While asleep it still hears everything said in the room and records it into its own transcript, so when the rep wakes it back up it has the full context of what happened while the other PAL was speaking. See [Conversational Flow](https://docs.tavus.io/sections/conversational-video-interface/pal/conversational-flow.md) for how wake and sleep detection works.

## Join the shared URL

The human participant joins the shared URL — the same URL both PALs are in. For a Tavus-managed Daily room, open the anchor's `conversation_url` in a browser or pass it to the Daily SDK, exactly as you would for a single-PAL conversation. For Google Meet, Zoom, or Microsoft Teams, join the meeting the normal way.

A worked two-PAL, one-human example:

1. The rep joins the shared URL. The CFO PAL is awake by default; the security PAL is asleep.
2. The rep discusses pricing with the CFO PAL.
3. The rep says `"Thanks Casey"`. The CFO PAL goes to sleep.
4. The rep says `"Hey Sam"`. The security PAL wakes up with the full transcript of the pricing discussion already in its context.
5. The rep discusses SOC 2 with the security PAL.
6. The rep says `"Thanks Sam"`, then `"Hey Casey"` to return to the CFO PAL.

## End every conversation

When the session is over, call [End Conversation](https://docs.tavus.io/api-reference/conversations/end-conversation.md) for **each** `conversation_id` you created. Ending one conversation does not end the other; each PAL has its own conversation, its own usage, and its own concurrency slot.

```shell cURL theme={null}
curl --request POST \
  --url https://tavusapi.com/v2/conversations/cAAAAAAAAAAA/end \
  --header 'x-api-key: <api_key>'

curl --request POST \
  --url https://tavusapi.com/v2/conversations/cBBBBBBBBBBB/end \
  --header 'x-api-key: <api_key>'
```

> [!WARNING]
>
> **Usage note**
>
> Each active PAL has its own conversation and consumes its own minutes **plus one concurrency slot**. Two concurrent PALs in the same room consume minutes at roughly twice the rate of one and use two concurrency slots, subject to your account's terms. End every conversation you created as soon as the session ends so idle PALs do not keep accruing usage or holding slots.

## Related docs

* [Conversational Flow](https://docs.tavus.io/sections/conversational-video-interface/pal/conversational-flow.md) - `wake_phrase` and `sleep_phrase` configuration.
* [Google Meet / Zoom / Teams](https://docs.tavus.io/sections/conversational-video-interface/pal/meetings.md) - `meeting_url` and the conferencing layer.
* [Participant Limits](https://docs.tavus.io/sections/conversational-video-interface/conversation/customizations/participant-limits.md) - `max_participants` behavior. Faces count as participants.
* [End Conversation](https://docs.tavus.io/api-reference/conversations/end-conversation.md) - how to clean up each conversation.
* [Create Conversation](https://docs.tavus.io/api-reference/conversations/create-conversation.md) - full request schema, including `properties.daily_room` and `meeting_url`.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.