diff --git a/api-reference/server/services/s2s/zoom-ai.mdx b/api-reference/server/services/s2s/zoom-ai.mdx
new file mode 100644
index 000000000..096b3e2cd
--- /dev/null
+++ b/api-reference/server/services/s2s/zoom-ai.mdx
@@ -0,0 +1,242 @@
+---
+title: "Zoom AI Voice Agent"
+sidebarTitle: "Zoom AI"
+description: "Speech-to-speech conversation with ZoomAIRealtimeLLMService over the Zoom AI Services Voice Agent API, with function calling."
+---
+
+## Overview
+
+`ZoomAIRealtimeLLMService` runs a speech-to-speech conversation with the [Voice Agent API](https://developers.zoom.us/docs/ai-services/voice-agent/) from Zoom AI Services, over a WebSocket [live session](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/). The Voice Agent detects turns, transcribes the caller, runs the model, and speaks the reply, so it replaces STT, the LLM, and TTS in a pipeline. It supports function calling, voice selection, custom vocabulary, and models from your own OpenAI or Anthropic account.
+
+
+
+ Pipecat's API methods for Zoom AI Voice Agent integration
+
+
+ Voice agent with function calling
+
+
+ Official Zoom Voice Agent API documentation
+
+
+ Get credentials and manage your Zoom AI Services plan
+
+
+
+## Installation
+
+To use Zoom AI services, install the required dependency:
+
+```bash
+uv add "pipecat-ai[zoom-ai]"
+```
+
+## Prerequisites
+
+### Zoom AI Account Setup
+
+Before using the Zoom AI Voice Agent, you need:
+
+1. **Zoom AI Services access**: Sign up through [Zoom AI Services](https://www.zoom.ai/)
+2. **API key**: Create a Zoom AI Services API key. See [Get credentials](https://developers.zoom.us/docs/ai-services/auth/).
+
+### Required Environment Variables
+
+- `ZOOM_AI_API_KEY`: Your Zoom AI Services API key
+
+## Configuration
+
+### ZoomAIRealtimeLLMService
+
+```python
+from pipecat.services.zoom_ai.realtime.llm import ZoomAIRealtimeLLMService
+```
+
+
+ Zoom AI Services API key. It is sent on every connection and reconnect, so it
+ must stay valid for as long as the service runs.
+
+
+
+ Voice Agent WebSocket URL.
+
+
+
+ Runtime-configurable settings. See [Settings](#settings) below.
+
+
+
+ Whether to start with audio input paused.
+
+
+### Settings
+
+Runtime-configurable settings passed via the `settings` constructor argument using `ZoomAIRealtimeLLMService.Settings(...)`. They map to the Voice Agent's [session settings](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#configure-the-session), and can be updated mid-conversation with `LLMUpdateSettingsFrame`. See [Service Settings](/pipecat/fundamentals/service-settings) for how settings work in Pipecat.
+
+| Parameter | Type | Default | Description |
+| -------------------- | ------------------- | --------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `model` | `str` | `"gemma-4"` | Conversational model: a Zoom model, or a model from your own `provider`. Set when a session starts, so a change mid-conversation logs a warning and takes effect only after a reconnect. See [Think settings](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#think-settings). _(Inherited from base settings.)_ |
+| `provider` | `Provider \| None` | `None` | Where the model runs. `None` uses Zoom's own models. Set it to use your own OpenAI or Anthropic account. Set when a session starts, like `model`. See [Provider](#provider). |
+| `system_instruction` | `str \| None` | `None` | Instructions for the agent. `None` uses the Voice Agent's default instructions for spoken conversation. _(Inherited from base settings.)_ |
+| `listen` | `Listen \| None` | `None` | How the agent listens: the caller's language and custom vocabulary. Set when a session starts, like `model`. `None` uses the Voice Agent's defaults. See [Listen](#listen). |
+| `speak` | `Speak \| None` | `None` | How the agent speaks: its voice. `None` uses the Voice Agent's default voice. See [Speak](#speak). |
+| `session_properties` | `SessionProperties` | `SessionProperties()` | OpenAI Realtime session properties. The service sends their `tools`, `tool_choice`, `max_output_tokens`, and `reasoning`. Set the voice with `speak`. |
+| `extra` | `dict[str, Any]` | `{}` | Voice Agent session fields that have no setting, shaped like the `session` of [`session.update`](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#configure-the-session) and merged into it. |
+
+`provider`, `listen`, and `speak` also accept plain dicts in the same shape, such as `{"voice": "ava"}`.
+
+### Provider
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Provider
+```
+
+| Parameter | Type | Default | Description |
+| ---------- | ---------------------------------------- | -------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `type` | `Literal["zoom", "openai", "anthropic"]` | `"zoom"` | `"zoom"` for Zoom's own models, or `"openai"` or `"anthropic"` for your own account with that provider. |
+| `model` | `str \| None` | `None` | Model ID from the provider. `None` uses the service's `model` setting. |
+| `headers` | `dict[str, str] \| None` | `None` | Your provider API key, as `{"X-Api-Key": "..."}`. Required for `"openai"` and `"anthropic"`. This is your OpenAI or Anthropic key, not your Zoom AI Services key. |
+| `metadata` | `dict[str, Any] \| None` | `None` | The provider model's limits, such as `context_window` and `max_output_tokens`. |
+
+See [Bring your own key](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#bring-your-own-key) for the providers and models the Voice Agent supports.
+
+### Listen
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Listen, Transcription, Vocabulary
+```
+
+| Parameter | Type | Default | Description |
+| -------------------------- | -------------------- | ------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `transcription.language` | `str \| None` | `None` | Language of the caller. See [Listen settings](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#listen-settings) for supported languages. |
+| `transcription.vocabulary` | `Vocabulary \| None` | `None` | Custom vocabulary: `Vocabulary(phrases=[...])`, names and terms to recognize, such as product names. |
+
+The service sets the input audio format from the pipeline's `audio_in_sample_rate`.
+
+### Speak
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Speak
+```
+
+| Parameter | Type | Default | Description |
+| --------- | ------------- | ------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `voice` | `str \| None` | `None` | Voice ID, such as `"jenna"` or `"ava"`. `None` uses the Voice Agent's default voice. See [Available voices](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#available-voices). |
+
+## Usage
+
+### Basic Setup
+
+```python
+import os
+
+from pipecat.services.zoom_ai.realtime.llm import ZoomAIRealtimeLLMService
+
+llm = ZoomAIRealtimeLLMService(
+ api_key=os.getenv("ZOOM_AI_API_KEY"),
+ settings=ZoomAIRealtimeLLMService.Settings(
+ system_instruction="You are a friendly assistant. Keep your replies to one or two sentences.",
+ ),
+)
+```
+
+Register tools on the `LLMContext`, as for other LLM services. See [Function Calling](/pipecat/learn/function-calling).
+
+### Choosing a Voice and Vocabulary
+
+This example sets the agent's voice and adds product names the Voice Agent should recognize:
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Listen, Speak, Transcription, Vocabulary
+
+llm = ZoomAIRealtimeLLMService(
+ api_key=os.getenv("ZOOM_AI_API_KEY"),
+ settings=ZoomAIRealtimeLLMService.Settings(
+ system_instruction="You are the returns assistant for Zoom Cafe Shop.",
+ listen=Listen(
+ transcription=Transcription(
+ vocabulary=Vocabulary(phrases=["Zoom Cafe Shop", "house blend", "milk frother"])
+ )
+ ),
+ speak=Speak(voice="ava"),
+ ),
+)
+```
+
+### Using Your Own OpenAI or Anthropic Model
+
+To run the conversation on your own OpenAI or Anthropic account, set `provider` and name the provider's model in `model`:
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Provider
+
+llm = ZoomAIRealtimeLLMService(
+ api_key=os.getenv("ZOOM_AI_API_KEY"),
+ settings=ZoomAIRealtimeLLMService.Settings(
+ model="gpt-4.1-mini",
+ provider=Provider(type="openai", headers={"X-Api-Key": os.getenv("OPENAI_API_KEY")}),
+ system_instruction="You are a friendly assistant.",
+ ),
+)
+```
+
+The Voice Agent still transcribes the caller and speaks the reply; only the model's turn uses your provider account.
+
+### Updating Settings Mid-Conversation
+
+Send an `LLMUpdateSettingsFrame` to change the instructions, tools, or voice during a conversation:
+
+```python
+from pipecat.frames.frames import LLMUpdateSettingsFrame
+
+await worker.queue_frame(
+ LLMUpdateSettingsFrame(
+ delta=ZoomAIRealtimeLLMService.Settings(
+ system_instruction="You are the shipping assistant for Zoom Cafe Shop.",
+ speak=Speak(voice="mark"),
+ )
+ )
+)
+```
+
+## Notes
+
+- **No local VAD**: The Voice Agent detects turns and handles barge-in itself, so the pipeline needs no VAD analyzer. Turn detection isn't configurable.
+- **Audio**: Input audio is sent at the pipeline's `audio_in_sample_rate`. The agent's audio is pushed as 24 kHz mono PCM. Audio that arrives while the session is connecting or reconnecting is held, up to the latest 3 seconds, and sent once the session is ready, so the caller can speak first.
+- **Transcripts**: The caller's speech is pushed upstream as `TranscriptionFrame`s, and the agent's speech downstream as `TTSTextFrame`s.
+- **Reconnects**: A session lasts up to 60 minutes. When a session ends unexpectedly, the service reconnects, up to 3 times in a row, and restores the conversation from the context. If reconnecting fails, it pushes a permanent error.
+- **Errors**: If Zoom refuses the connection (for example, an invalid API key), the error includes Zoom's reason. If the Voice Agent rejects the session's settings (an unknown voice, for example), or the provider rejects your own provider key, the session ends with a permanent error. Other errors the Voice Agent reports without ending the session are pushed as non-fatal errors. Errors include the session's tracking ID (`x-zm-trackingid`), which Zoom support uses to find a session. See [Error Handling](/pipecat/fundamentals/error-handling).
+- **System instruction precedence**: The `system_instruction` from service settings takes precedence over an initial system message in the LLM context. A warning is logged when both are set.
+
+## Event Handlers
+
+| Event | Description |
+| ------------------------------ | -------------------------------------------------------------------------------------------------------------------- |
+| `on_conversation_item_created` | Called when a new conversation item is created in the session |
+| `on_conversation_item_updated` | Called when a conversation item is updated or completed. `item` is `None` when a user audio transcription completes. |
+
+```python
+@llm.event_handler("on_conversation_item_created")
+async def on_item_created(service, item_id, item):
+ print(f"New conversation item: {item_id}")
+```
diff --git a/api-reference/server/services/supported-services.mdx b/api-reference/server/services/supported-services.mdx
index 6523ba6d8..918fac993 100644
--- a/api-reference/server/services/supported-services.mdx
+++ b/api-reference/server/services/supported-services.mdx
@@ -157,6 +157,7 @@ Realtime LLMs (speech-to-speech) are multi-modal models that take in audio, vide
| [OpenAI Realtime](/api-reference/server/services/s2s/openai) | `uv add "pipecat-ai[openai]"` | Pipecat |
| [ThunderPhone](/api-reference/server/services/s2s/thunderphone) | `uv add pipecat-thunderphone` | Community |
| [Ultravox](/api-reference/server/services/s2s/ultravox) | `uv add "pipecat-ai[ultravox]"` | Pipecat |
+| [Zoom AI Voice Agent](/api-reference/server/services/s2s/zoom-ai) | `uv add "pipecat-ai[zoom-ai]"` | Pipecat |
## Serializers
diff --git a/docs.json b/docs.json
index 0f793e21b..51dad0c44 100644
--- a/docs.json
+++ b/docs.json
@@ -565,7 +565,8 @@
"api-reference/server/services/s2s/openai-live",
"api-reference/server/services/s2s/openai",
"api-reference/server/services/s2s/thunderphone",
- "api-reference/server/services/s2s/ultravox"
+ "api-reference/server/services/s2s/ultravox",
+ "api-reference/server/services/s2s/zoom-ai"
]
},
{
diff --git a/llms-full.txt b/llms-full.txt
index d9b424049..1d40d1b34 100644
--- a/llms-full.txt
+++ b/llms-full.txt
@@ -36550,6 +36550,7 @@ Realtime LLMs (speech-to-speech) are multi-modal models that take in audio, vide
| [OpenAI Realtime](/api-reference/server/services/s2s/openai) | `uv add "pipecat-ai[openai]"` | Pipecat |
| [ThunderPhone](/api-reference/server/services/s2s/thunderphone) | `uv add pipecat-thunderphone` | Community |
| [Ultravox](/api-reference/server/services/s2s/ultravox) | `uv add "pipecat-ai[ultravox]"` | Pipecat |
+| [Zoom AI Voice Agent](/api-reference/server/services/s2s/zoom-ai) | `uv add "pipecat-ai[zoom-ai]"` | Pipecat |
## Serializers
@@ -72653,6 +72654,248 @@ await worker.queue_frame(
- **Async tool support**: Functions registered with `cancel_on_interruption=False` are supported. When an async function is invoked, Ultravox ships a placeholder `client_tool_result` immediately to unfreeze the conversation (since Ultravox's API freezes between tool invocation and result). The actual result is injected as user-side text once the tool completes. Intermediate results (`is_final=False`) are injected as user-side text too.
- **No user turn frames**: The service emits no `UserStartedSpeakingFrame` or `UserStoppedSpeakingFrame`, so processors that depend on them (RTVI user speech events, `UserIdleController`, user mute strategies) don't activate unless you add a `vad_analyzer` to `LLMUserAggregatorParams`.
+# Zoom AI Voice Agent
+Source: https://docs.pipecat.ai/api-reference/server/services/s2s/zoom-ai.md
+
+Speech-to-speech conversation with ZoomAIRealtimeLLMService over the Zoom AI Services Voice Agent API, with function calling.
+
+## Overview
+
+`ZoomAIRealtimeLLMService` runs a speech-to-speech conversation with the [Voice Agent API](https://developers.zoom.us/docs/ai-services/voice-agent/) from Zoom AI Services, over a WebSocket [live session](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/). The Voice Agent detects turns, transcribes the caller, runs the model, and speaks the reply, so it replaces STT, the LLM, and TTS in a pipeline. It supports function calling, voice selection, custom vocabulary, and models from your own OpenAI or Anthropic account.
+
+
+
+ Pipecat's API methods for Zoom AI Voice Agent integration
+
+
+ Voice agent with function calling
+
+
+ Official Zoom Voice Agent API documentation
+
+
+ Get credentials and manage your Zoom AI Services plan
+
+
+
+## Installation
+
+To use Zoom AI services, install the required dependency:
+
+```bash
+uv add "pipecat-ai[zoom-ai]"
+```
+
+## Prerequisites
+
+### Zoom AI Account Setup
+
+Before using the Zoom AI Voice Agent, you need:
+
+1. **Zoom AI Services access**: Sign up through [Zoom AI Services](https://www.zoom.ai/)
+2. **API key**: Create a Zoom AI Services API key. See [Get credentials](https://developers.zoom.us/docs/ai-services/auth/).
+
+### Required Environment Variables
+
+- `ZOOM_AI_API_KEY`: Your Zoom AI Services API key
+
+## Configuration
+
+### ZoomAIRealtimeLLMService
+
+```python
+from pipecat.services.zoom_ai.realtime.llm import ZoomAIRealtimeLLMService
+```
+
+
+ Zoom AI Services API key. It is sent on every connection and reconnect, so it
+ must stay valid for as long as the service runs.
+
+
+
+ Voice Agent WebSocket URL.
+
+
+
+ Runtime-configurable settings. See [Settings](#settings) below.
+
+
+
+ Whether to start with audio input paused.
+
+
+### Settings
+
+Runtime-configurable settings passed via the `settings` constructor argument using `ZoomAIRealtimeLLMService.Settings(...)`. They map to the Voice Agent's [session settings](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#configure-the-session), and can be updated mid-conversation with `LLMUpdateSettingsFrame`. See [Service Settings](/pipecat/fundamentals/service-settings) for how settings work in Pipecat.
+
+| Parameter | Type | Default | Description |
+| -------------------- | ------------------- | --------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `model` | `str` | `"gemma-4"` | Conversational model: a Zoom model, or a model from your own `provider`. Set when a session starts, so a change mid-conversation logs a warning and takes effect only after a reconnect. See [Think settings](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#think-settings). _(Inherited from base settings.)_ |
+| `provider` | `Provider \| None` | `None` | Where the model runs. `None` uses Zoom's own models. Set it to use your own OpenAI or Anthropic account. Set when a session starts, like `model`. See [Provider](#provider). |
+| `system_instruction` | `str \| None` | `None` | Instructions for the agent. `None` uses the Voice Agent's default instructions for spoken conversation. _(Inherited from base settings.)_ |
+| `listen` | `Listen \| None` | `None` | How the agent listens: the caller's language and custom vocabulary. Set when a session starts, like `model`. `None` uses the Voice Agent's defaults. See [Listen](#listen). |
+| `speak` | `Speak \| None` | `None` | How the agent speaks: its voice. `None` uses the Voice Agent's default voice. See [Speak](#speak). |
+| `session_properties` | `SessionProperties` | `SessionProperties()` | OpenAI Realtime session properties. The service sends their `tools`, `tool_choice`, `max_output_tokens`, and `reasoning`. Set the voice with `speak`. |
+| `extra` | `dict[str, Any]` | `{}` | Voice Agent session fields that have no setting, shaped like the `session` of [`session.update`](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#configure-the-session) and merged into it. |
+
+`provider`, `listen`, and `speak` also accept plain dicts in the same shape, such as `{"voice": "ava"}`.
+
+### Provider
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Provider
+```
+
+| Parameter | Type | Default | Description |
+| ---------- | ---------------------------------------- | -------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `type` | `Literal["zoom", "openai", "anthropic"]` | `"zoom"` | `"zoom"` for Zoom's own models, or `"openai"` or `"anthropic"` for your own account with that provider. |
+| `model` | `str \| None` | `None` | Model ID from the provider. `None` uses the service's `model` setting. |
+| `headers` | `dict[str, str] \| None` | `None` | Your provider API key, as `{"X-Api-Key": "..."}`. Required for `"openai"` and `"anthropic"`. This is your OpenAI or Anthropic key, not your Zoom AI Services key. |
+| `metadata` | `dict[str, Any] \| None` | `None` | The provider model's limits, such as `context_window` and `max_output_tokens`. |
+
+See [Bring your own key](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#bring-your-own-key) for the providers and models the Voice Agent supports.
+
+### Listen
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Listen, Transcription, Vocabulary
+```
+
+| Parameter | Type | Default | Description |
+| -------------------------- | -------------------- | ------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `transcription.language` | `str \| None` | `None` | Language of the caller. See [Listen settings](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#listen-settings) for supported languages. |
+| `transcription.vocabulary` | `Vocabulary \| None` | `None` | Custom vocabulary: `Vocabulary(phrases=[...])`, names and terms to recognize, such as product names. |
+
+The service sets the input audio format from the pipeline's `audio_in_sample_rate`.
+
+### Speak
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Speak
+```
+
+| Parameter | Type | Default | Description |
+| --------- | ------------- | ------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `voice` | `str \| None` | `None` | Voice ID, such as `"jenna"` or `"ava"`. `None` uses the Voice Agent's default voice. See [Available voices](https://developers.zoom.us/docs/ai-services/voice-agent/live-mode/#available-voices). |
+
+## Usage
+
+### Basic Setup
+
+```python
+import os
+
+from pipecat.services.zoom_ai.realtime.llm import ZoomAIRealtimeLLMService
+
+llm = ZoomAIRealtimeLLMService(
+ api_key=os.getenv("ZOOM_AI_API_KEY"),
+ settings=ZoomAIRealtimeLLMService.Settings(
+ system_instruction="You are a friendly assistant. Keep your replies to one or two sentences.",
+ ),
+)
+```
+
+Register tools on the `LLMContext`, as for other LLM services. See [Function Calling](/pipecat/learn/function-calling).
+
+### Choosing a Voice and Vocabulary
+
+This example sets the agent's voice and adds product names the Voice Agent should recognize:
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Listen, Speak, Transcription, Vocabulary
+
+llm = ZoomAIRealtimeLLMService(
+ api_key=os.getenv("ZOOM_AI_API_KEY"),
+ settings=ZoomAIRealtimeLLMService.Settings(
+ system_instruction="You are the returns assistant for Zoom Cafe Shop.",
+ listen=Listen(
+ transcription=Transcription(
+ vocabulary=Vocabulary(phrases=["Zoom Cafe Shop", "house blend", "milk frother"])
+ )
+ ),
+ speak=Speak(voice="ava"),
+ ),
+)
+```
+
+### Using Your Own OpenAI or Anthropic Model
+
+To run the conversation on your own OpenAI or Anthropic account, set `provider` and name the provider's model in `model`:
+
+```python
+from pipecat.services.zoom_ai.realtime.events import Provider
+
+llm = ZoomAIRealtimeLLMService(
+ api_key=os.getenv("ZOOM_AI_API_KEY"),
+ settings=ZoomAIRealtimeLLMService.Settings(
+ model="gpt-4.1-mini",
+ provider=Provider(type="openai", headers={"X-Api-Key": os.getenv("OPENAI_API_KEY")}),
+ system_instruction="You are a friendly assistant.",
+ ),
+)
+```
+
+The Voice Agent still transcribes the caller and speaks the reply; only the model's turn uses your provider account.
+
+### Updating Settings Mid-Conversation
+
+Send an `LLMUpdateSettingsFrame` to change the instructions, tools, or voice during a conversation:
+
+```python
+from pipecat.frames.frames import LLMUpdateSettingsFrame
+
+await worker.queue_frame(
+ LLMUpdateSettingsFrame(
+ delta=ZoomAIRealtimeLLMService.Settings(
+ system_instruction="You are the shipping assistant for Zoom Cafe Shop.",
+ speak=Speak(voice="mark"),
+ )
+ )
+)
+```
+
+## Notes
+
+- **No local VAD**: The Voice Agent detects turns and handles barge-in itself, so the pipeline needs no VAD analyzer. Turn detection isn't configurable.
+- **Audio**: Input audio is sent at the pipeline's `audio_in_sample_rate`. The agent's audio is pushed as 24 kHz mono PCM. Audio that arrives while the session is connecting or reconnecting is held, up to the latest 3 seconds, and sent once the session is ready, so the caller can speak first.
+- **Transcripts**: The caller's speech is pushed upstream as `TranscriptionFrame`s, and the agent's speech downstream as `TTSTextFrame`s.
+- **Reconnects**: A session lasts up to 60 minutes. When a session ends unexpectedly, the service reconnects, up to 3 times in a row, and restores the conversation from the context. If reconnecting fails, it pushes a permanent error.
+- **Errors**: If Zoom refuses the connection (for example, an invalid API key), the error includes Zoom's reason. If the Voice Agent rejects the session's settings (an unknown voice, for example), or the provider rejects your own provider key, the session ends with a permanent error. Other errors the Voice Agent reports without ending the session are pushed as non-fatal errors. Errors include the session's tracking ID (`x-zm-trackingid`), which Zoom support uses to find a session. See [Error Handling](/pipecat/fundamentals/error-handling).
+- **System instruction precedence**: The `system_instruction` from service settings takes precedence over an initial system message in the LLM context. A warning is logged when both are set.
+
+## Event Handlers
+
+| Event | Description |
+| ------------------------------ | -------------------------------------------------------------------------------------------------------------------- |
+| `on_conversation_item_created` | Called when a new conversation item is created in the session |
+| `on_conversation_item_updated` | Called when a conversation item is updated or completed. `item` is `None` when a user audio transcription completes. |
+
+```python
+@llm.event_handler("on_conversation_item_created")
+async def on_item_created(service, item_id, item):
+ print(f"New conversation item: {item_id}")
+```
+
# Azure OpenAI Image Generation
Source: https://docs.pipecat.ai/api-reference/server/services/image-generation/azure.md
diff --git a/llms.txt b/llms.txt
index 36333e485..181963d20 100644
--- a/llms.txt
+++ b/llms.txt
@@ -496,6 +496,7 @@ LLM, transports, serializers), pipelines, frames, workers, and utilities.
- [OpenAI Realtime](https://docs.pipecat.ai/api-reference/server/services/s2s/openai.md): OpenAIRealtimeLLMService provides speech-to-speech conversation over the OpenAI Realtime API, replacing separate STT, LLM, and TTS.
- [ThunderPhone](https://docs.pipecat.ai/api-reference/server/services/s2s/thunderphone.md): Speech-to-speech service that runs a ThunderPhone voice agent (speech recognition, LLM, voice, turn-taking, 47 languages and tools) inside a Pipecat pipeline
- [Ultravox Realtime](https://docs.pipecat.ai/api-reference/server/services/s2s/ultravox.md): UltravoxRealtimeLLMService provides real-time conversational AI with Ultravox Realtime, with agent and one-shot input modes.
+- [Zoom AI Voice Agent](https://docs.pipecat.ai/api-reference/server/services/s2s/zoom-ai.md): Speech-to-speech conversation with ZoomAIRealtimeLLMService over the Zoom AI Services Voice Agent API, with function calling.
##### Image Generation