# File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details. from __future__ import annotations import json import time import random import logging from types import TracebackType from typing import TYPE_CHECKING, Any, Union, Callable, Iterator, Awaitable, cast from typing_extensions import AsyncIterator import httpx from pydantic import BaseModel from .calls import ( Calls, AsyncCalls, CallsWithRawResponse, AsyncCallsWithRawResponse, CallsWithStreamingResponse, AsyncCallsWithStreamingResponse, ) from ..._types import Omit, Query, Headers, omit from ..._utils import ( is_azure_client, maybe_transform, strip_not_given, async_maybe_transform, is_async_azure_client, ) from ..._compat import cached_property from ..._models import construct_type_unchecked from ..._resource import SyncAPIResource, AsyncAPIResource from ..._exceptions import OpenAIError, WebSocketConnectionClosedError from ..._send_queue import SendQueue from ..._base_client import _merge_mappings from .client_secrets import ( ClientSecrets, AsyncClientSecrets, ClientSecretsWithRawResponse, AsyncClientSecretsWithRawResponse, ClientSecretsWithStreamingResponse, AsyncClientSecretsWithStreamingResponse, ) from ..._event_handler import EventHandlerRegistry from ...types.realtime import session_update_event_param from ...types.websocket_reconnection import ReconnectingEvent, ReconnectingOverrides, is_recoverable_close from ...types.websocket_connection_options import WebSocketConnectionOptions from ...types.realtime.realtime_error_event import RealtimeErrorEvent from ...types.realtime.realtime_client_event import RealtimeClientEvent from ...types.realtime.realtime_server_event import RealtimeServerEvent from ...types.realtime.conversation_item_param import ConversationItemParam from ...types.realtime.realtime_client_event_param import RealtimeClientEventParam from ...types.realtime.realtime_response_create_params_param import RealtimeResponseCreateParamsParam if TYPE_CHECKING: from websockets.sync.client import ClientConnection as WebSocketConnection from websockets.asyncio.client import ClientConnection as AsyncWebSocketConnection from ..._client import OpenAI, AsyncOpenAI __all__ = ["Realtime", "AsyncRealtime"] log: logging.Logger = logging.getLogger(__name__) class Realtime(SyncAPIResource): @cached_property def client_secrets(self) -> ClientSecrets: return ClientSecrets(self._client) @cached_property def calls(self) -> Calls: from ...lib._realtime import _Calls return _Calls(self._client) @cached_property def with_raw_response(self) -> RealtimeWithRawResponse: """ This property can be used as a prefix for any HTTP method call to return the raw response object instead of the parsed content. For more information, see https://www.github.com/openai/openai-python#accessing-raw-response-data-eg-headers """ return RealtimeWithRawResponse(self) @cached_property def with_streaming_response(self) -> RealtimeWithStreamingResponse: """ An alternative to `.with_raw_response` that doesn't eagerly read the response body. For more information, see https://www.github.com/openai/openai-python#with_streaming_response """ return RealtimeWithStreamingResponse(self) def connect( self, *, call_id: str | Omit = omit, model: str | Omit = omit, extra_query: Query = {}, extra_headers: Headers = {}, websocket_connection_options: WebSocketConnectionOptions = {}, on_reconnecting: Callable[[ReconnectingEvent], ReconnectingOverrides | None] | None = None, max_retries: int = 5, initial_delay: float = 0.5, max_delay: float = 8.0, max_queue_size: int = 1_048_576, ) -> RealtimeConnectionManager: """ The Realtime API enables you to build low-latency, multi-modal conversational experiences. It currently supports text and audio as both input and output, as well as function calling. Some notable benefits of the API include: - Native speech-to-speech: Skipping an intermediate text format means low latency and nuanced output. - Natural, steerable voices: The models have natural inflection and can laugh, whisper, and adhere to tone direction. - Simultaneous multimodal output: Text is useful for moderation; faster-than-realtime audio ensures stable playback. The Realtime API is a stateful, event-based API that communicates over a WebSocket. """ return RealtimeConnectionManager( client=self._client, extra_query=extra_query, extra_headers=extra_headers, websocket_connection_options=websocket_connection_options, on_reconnecting=on_reconnecting, max_retries=max_retries, initial_delay=initial_delay, max_delay=max_delay, max_queue_size=max_queue_size, call_id=call_id, model=model, ) class AsyncRealtime(AsyncAPIResource): @cached_property def client_secrets(self) -> AsyncClientSecrets: return AsyncClientSecrets(self._client) @cached_property def calls(self) -> AsyncCalls: from ...lib._realtime import _AsyncCalls return _AsyncCalls(self._client) @cached_property def with_raw_response(self) -> AsyncRealtimeWithRawResponse: """ This property can be used as a prefix for any HTTP method call to return the raw response object instead of the parsed content. For more information, see https://www.github.com/openai/openai-python#accessing-raw-response-data-eg-headers """ return AsyncRealtimeWithRawResponse(self) @cached_property def with_streaming_response(self) -> AsyncRealtimeWithStreamingResponse: """ An alternative to `.with_raw_response` that doesn't eagerly read the response body. For more information, see https://www.github.com/openai/openai-python#with_streaming_response """ return AsyncRealtimeWithStreamingResponse(self) def connect( self, *, call_id: str | Omit = omit, model: str | Omit = omit, extra_query: Query = {}, extra_headers: Headers = {}, websocket_connection_options: WebSocketConnectionOptions = {}, on_reconnecting: Callable[[ReconnectingEvent], ReconnectingOverrides | None] | None = None, max_retries: int = 5, initial_delay: float = 0.5, max_delay: float = 8.0, max_queue_size: int = 1_048_576, ) -> AsyncRealtimeConnectionManager: """ The Realtime API enables you to build low-latency, multi-modal conversational experiences. It currently supports text and audio as both input and output, as well as function calling. Some notable benefits of the API include: - Native speech-to-speech: Skipping an intermediate text format means low latency and nuanced output. - Natural, steerable voices: The models have natural inflection and can laugh, whisper, and adhere to tone direction. - Simultaneous multimodal output: Text is useful for moderation; faster-than-realtime audio ensures stable playback. The Realtime API is a stateful, event-based API that communicates over a WebSocket. """ return AsyncRealtimeConnectionManager( client=self._client, extra_query=extra_query, extra_headers=extra_headers, websocket_connection_options=websocket_connection_options, on_reconnecting=on_reconnecting, max_retries=max_retries, initial_delay=initial_delay, max_delay=max_delay, max_queue_size=max_queue_size, call_id=call_id, model=model, ) class RealtimeWithRawResponse: def __init__(self, realtime: Realtime) -> None: self._realtime = realtime @cached_property def client_secrets(self) -> ClientSecretsWithRawResponse: return ClientSecretsWithRawResponse(self._realtime.client_secrets) @cached_property def calls(self) -> CallsWithRawResponse: return CallsWithRawResponse(self._realtime.calls) class AsyncRealtimeWithRawResponse: def __init__(self, realtime: AsyncRealtime) -> None: self._realtime = realtime @cached_property def client_secrets(self) -> AsyncClientSecretsWithRawResponse: return AsyncClientSecretsWithRawResponse(self._realtime.client_secrets) @cached_property def calls(self) -> AsyncCallsWithRawResponse: return AsyncCallsWithRawResponse(self._realtime.calls) class RealtimeWithStreamingResponse: def __init__(self, realtime: Realtime) -> None: self._realtime = realtime @cached_property def client_secrets(self) -> ClientSecretsWithStreamingResponse: return ClientSecretsWithStreamingResponse(self._realtime.client_secrets) @cached_property def calls(self) -> CallsWithStreamingResponse: return CallsWithStreamingResponse(self._realtime.calls) class AsyncRealtimeWithStreamingResponse: def __init__(self, realtime: AsyncRealtime) -> None: self._realtime = realtime @cached_property def client_secrets(self) -> AsyncClientSecretsWithStreamingResponse: return AsyncClientSecretsWithStreamingResponse(self._realtime.client_secrets) @cached_property def calls(self) -> AsyncCallsWithStreamingResponse: return AsyncCallsWithStreamingResponse(self._realtime.calls) class AsyncRealtimeConnection: """Represents a live WebSocket connection to the Realtime API""" session: AsyncRealtimeSessionResource response: AsyncRealtimeResponseResource input_audio_buffer: AsyncRealtimeInputAudioBufferResource conversation: AsyncRealtimeConversationResource output_audio_buffer: AsyncRealtimeOutputAudioBufferResource _connection: AsyncWebSocketConnection def __init__( self, connection: AsyncWebSocketConnection, *, make_ws: Callable[[Query, Headers], Awaitable[AsyncWebSocketConnection]] | None = None, on_reconnecting: Callable[[ReconnectingEvent], ReconnectingOverrides | None] | None = None, max_retries: int = 5, initial_delay: float = 0.5, max_delay: float = 8.0, extra_query: Query = {}, extra_headers: Headers = {}, send_queue: SendQueue | None = None, ) -> None: self._connection = connection self._make_ws = make_ws self._on_reconnecting = on_reconnecting self._max_retries = max_retries self._initial_delay = initial_delay self._max_delay = max_delay self._extra_query = extra_query self._extra_headers = extra_headers self._intentionally_closed = False self._is_reconnecting = False self._send_queue = send_queue or SendQueue() self._event_handler_registry = EventHandlerRegistry(use_lock=False) self.session = AsyncRealtimeSessionResource(self) self.response = AsyncRealtimeResponseResource(self) self.input_audio_buffer = AsyncRealtimeInputAudioBufferResource(self) self.conversation = AsyncRealtimeConversationResource(self) self.output_audio_buffer = AsyncRealtimeOutputAudioBufferResource(self) async def __aiter__(self) -> AsyncIterator[RealtimeServerEvent]: """ An infinite-iterator that will continue to yield events until the connection is closed. """ from websockets.exceptions import ConnectionClosedOK, ConnectionClosedError while True: try: yield await self.recv() except ConnectionClosedOK: return except ConnectionClosedError as exc: if not await self._reconnect(exc): unsent = self._send_queue.drain() if unsent: raise WebSocketConnectionClosedError( "WebSocket connection closed with unsent messages", unsent_messages=unsent, ) from exc raise async def recv(self) -> RealtimeServerEvent: """ Receive the next message from the connection and parses it into a `RealtimeServerEvent` object. Canceling this method is safe. There's no risk of losing data. """ return self.parse_event(await self.recv_bytes()) async def recv_bytes(self) -> bytes: """Receive the next message from the connection as raw bytes. Canceling this method is safe. There's no risk of losing data. If you want to parse the message into a `RealtimeServerEvent` object like `.recv()` does, then you can call `.parse_event(data)`. """ message = await self._connection.recv(decode=False) log.debug(f"Received WebSocket message: %s", message) return message async def send(self, event: RealtimeClientEvent | RealtimeClientEventParam) -> None: data = ( event.to_json(use_api_names=True, exclude_defaults=True, exclude_unset=True) if isinstance(event, BaseModel) else json.dumps(await async_maybe_transform(event, RealtimeClientEventParam)) ) if self._is_reconnecting: self._send_queue.enqueue(data) return try: await self._connection.send(data) except Exception: self._send_queue.enqueue(data) raise async def send_raw(self, data: bytes | str) -> None: if self._is_reconnecting: raw = data if isinstance(data, str) else data.decode("utf-8") self._send_queue.enqueue(raw) return await self._connection.send(data) async def close(self, *, code: int = 1000, reason: str = "") -> None: self._intentionally_closed = True await self._connection.close(code=code, reason=reason) def parse_event(self, data: str | bytes) -> RealtimeServerEvent: """ Converts a raw `str` or `bytes` message into a `RealtimeServerEvent` object. This is helpful if you're using `.recv_bytes()`. """ return cast( RealtimeServerEvent, construct_type_unchecked(value=json.loads(data), type_=cast(Any, RealtimeServerEvent)) ) async def _reconnect(self, exc: Exception) -> bool: """Attempt to reconnect after a connection failure. Returns ``True`` if a new connection was established, ``False`` if the caller should re-raise the original exception. """ import asyncio if self._on_reconnecting is None or self._make_ws is None: return False from websockets.exceptions import ConnectionClosedError close_code = 1006 if isinstance(exc, ConnectionClosedError) and exc.rcvd is not None: close_code = exc.rcvd.code if not is_recoverable_close(close_code): return False self._is_reconnecting = True for attempt in range(1, self._max_retries + 1): base_delay = min(self._initial_delay * (2 ** (attempt - 1)), self._max_delay) jitter = 0.75 + random.random() * 0.25 delay = base_delay * jitter event = ReconnectingEvent( attempt=attempt, max_attempts=self._max_retries, delay=delay, close_code=close_code, extra_query=self._extra_query, extra_headers=self._extra_headers, ) try: result = self._on_reconnecting(event) except Exception: self._is_reconnecting = False return False if result is not None and result.get("abort"): self._is_reconnecting = False return False if result is not None: if "extra_query" in result: self._extra_query = result["extra_query"] if "extra_headers" in result: self._extra_headers = result["extra_headers"] log.info( "Reconnecting to WebSocket API (attempt %d/%d) after %.1fs delay", attempt, self._max_retries, delay, ) await asyncio.sleep(delay) if self._intentionally_closed: self._is_reconnecting = False return False try: self._connection = await self._make_ws(self._extra_query, self._extra_headers) log.info("Reconnected to WebSocket API") self._is_reconnecting = False await self._flush_send_queue() return True except Exception: pass self._is_reconnecting = False return False async def _flush_send_queue(self) -> None: """Send all queued messages over the current connection.""" async def _send(data: str) -> None: await self._connection.send(data) try: await self._send_queue.flush_async(_send) except Exception: log.warning("Failed to flush send queue after reconnect", exc_info=True) def on( self, event_type: str, handler: Callable[..., Any] | None = None ) -> Union[AsyncRealtimeConnection, Callable[[Callable[..., Any]], Callable[..., Any]]]: """Adds the handler to the end of the handlers list for the given event type. No checks are made to see if the handler has already been added. Multiple calls passing the same combination of event type and handler will result in the handler being added, and called, multiple times. Can be used as a method (returns ``self`` for chaining):: connection.on("conversation.created", my_handler) Or as a decorator:: @connection.on("conversation.created") async def my_handler(event): ... """ if handler is not None: self._event_handler_registry.add(event_type, handler) return self def decorator(fn: Callable[..., Any]) -> Callable[..., Any]: self._event_handler_registry.add(event_type, fn) return fn return decorator def off(self, event_type: str, handler: Callable[..., Any]) -> AsyncRealtimeConnection: """Remove a previously registered event handler.""" self._event_handler_registry.remove(event_type, handler) return self def once( self, event_type: str, handler: Callable[..., Any] | None = None ) -> Union[AsyncRealtimeConnection, Callable[[Callable[..., Any]], Callable[..., Any]]]: """Register a one-time event handler. Automatically removed after first invocation. """ if handler is not None: self._event_handler_registry.add(event_type, handler, once=True) return self def decorator(fn: Callable[..., Any]) -> Callable[..., Any]: self._event_handler_registry.add(event_type, fn, once=True) return fn return decorator async def dispatch_events(self) -> None: """Run the event loop, dispatching received events to registered handlers. Blocks until the connection is closed. This is the push-based alternative to iterating with ``async for event in connection``. If an ``"error"`` event arrives and no handler is registered for ``"error"`` or ``"event"``, an ``OpenAIError`` is raised. """ import asyncio async for event in self: event_type = event.type specific = self._event_handler_registry.get_handlers(event_type) generic = self._event_handler_registry.get_handlers("event") if event_type == "error" and not specific and not generic: if isinstance(event, RealtimeErrorEvent): raise OpenAIError(f"WebSocket error: {event}") for handler in specific: result = handler(event) if asyncio.iscoroutine(result): await result for handler in generic: result = handler(event) if asyncio.iscoroutine(result): await result class AsyncRealtimeConnectionManager: """ Context manager over a `AsyncRealtimeConnection` that is returned by `realtime.connect()` This context manager ensures that the connection will be closed when it exits. --- Note that if your application doesn't work well with the context manager approach then you can call the `.enter()` method directly to initiate a connection. **Warning**: You must remember to close the connection with `.close()`. ```py connection = await client.realtime.connect(...).enter() # ... await connection.close() ``` """ def __init__( self, *, client: AsyncOpenAI, call_id: str | Omit = omit, model: str | Omit = omit, extra_query: Query, extra_headers: Headers, websocket_connection_options: WebSocketConnectionOptions, on_reconnecting: Callable[[ReconnectingEvent], ReconnectingOverrides | None] | None = None, max_retries: int = 5, initial_delay: float = 0.5, max_delay: float = 8.0, max_queue_size: int = 1_048_576, ) -> None: self.__client = client self.__call_id = call_id self.__model = model self.__connection: AsyncRealtimeConnection | None = None self.__extra_query = extra_query self.__extra_headers = extra_headers self.__websocket_connection_options = websocket_connection_options self.__on_reconnecting = on_reconnecting self.__max_retries = max_retries self.__initial_delay = initial_delay self.__max_delay = max_delay self.__send_queue = SendQueue(max_bytes=max_queue_size) self.__event_handler_registry = EventHandlerRegistry(use_lock=False) def send(self, event: RealtimeClientEvent | RealtimeClientEventParam) -> None: """Queue a message to be sent when the connection is established. This can be called before entering the context manager. Queued messages are automatically sent once the WebSocket connection opens. """ data = ( event.to_json(use_api_names=True, exclude_defaults=True, exclude_unset=True) if isinstance(event, BaseModel) else json.dumps(event) ) self.__send_queue.enqueue(data) def on( self, event_type: str, handler: Callable[..., Any] | None = None ) -> Union[AsyncRealtimeConnectionManager, Callable[[Callable[..., Any]], Callable[..., Any]]]: """Register an event handler before the connection is established. Handlers are transferred to the connection on enter. Supports the same method and decorator forms as ``AsyncRealtimeConnection.on``. """ if handler is not None: self.__event_handler_registry.add(event_type, handler) return self def decorator(fn: Callable[..., Any]) -> Callable[..., Any]: self.__event_handler_registry.add(event_type, fn) return fn return decorator def off(self, event_type: str, handler: Callable[..., Any]) -> AsyncRealtimeConnectionManager: """Remove a previously registered event handler.""" self.__event_handler_registry.remove(event_type, handler) return self def once( self, event_type: str, handler: Callable[..., Any] | None = None ) -> Union[AsyncRealtimeConnectionManager, Callable[[Callable[..., Any]], Callable[..., Any]]]: """Register a one-time event handler before the connection is established.""" if handler is not None: self.__event_handler_registry.add(event_type, handler, once=True) return self def decorator(fn: Callable[..., Any]) -> Callable[..., Any]: self.__event_handler_registry.add(event_type, fn, once=True) return fn return decorator async def __aenter__(self) -> AsyncRealtimeConnection: """ If your application doesn't work well with the context manager approach then you can call this method directly to initiate a connection. **Warning**: You must remember to close the connection with `.close()`. ```py connection = await client.realtime.connect(...).enter() # ... await connection.close() ``` """ ws = await self._connect_ws(self.__extra_query, self.__extra_headers) self.__connection = AsyncRealtimeConnection( ws, make_ws=self._connect_ws if self.__on_reconnecting is not None else None, on_reconnecting=self.__on_reconnecting, max_retries=self.__max_retries, initial_delay=self.__initial_delay, max_delay=self.__max_delay, extra_query=self.__extra_query, extra_headers=self.__extra_headers, send_queue=self.__send_queue, ) self.__event_handler_registry.merge_into(self.__connection._event_handler_registry) await self.__connection._flush_send_queue() return self.__connection enter = __aenter__ async def _connect_ws(self, extra_query: Query, extra_headers: Headers) -> AsyncWebSocketConnection: try: from websockets.asyncio.client import connect except ImportError as exc: raise OpenAIError("You need to install `openai[realtime]` to use this method") from exc await self.__client._refresh_api_key() auth_headers = self.__client.auth_headers if self.__call_id is not omit: extra_query = {**extra_query, "call_id": self.__call_id} if is_async_azure_client(self.__client): model = self.__model if not model: raise OpenAIError("`model` is required for Azure Realtime API") else: url, auth_headers = await self.__client._configure_realtime(model, extra_query) else: url = self._prepare_url().copy_with( params={ **self.__client.base_url.params, **({"model": self.__model} if self.__model is not omit else {}), **extra_query, }, ) log.debug("Connecting to %s", url) if self.__websocket_connection_options: log.debug("Connection options: %s", self.__websocket_connection_options) return await connect( str(url), user_agent_header=self.__client.user_agent, additional_headers=_merge_mappings( { **auth_headers, }, extra_headers, ), **self.__websocket_connection_options, ) def _prepare_url(self) -> httpx.URL: if self.__client.websocket_base_url is not None: base_url = httpx.URL(self.__client.websocket_base_url) else: scheme = self.__client._base_url.scheme ws_scheme = "ws" if scheme == "http" else "wss" base_url = self.__client._base_url.copy_with(scheme=ws_scheme) merge_raw_path = base_url.raw_path.rstrip(b"/") + b"/realtime" return base_url.copy_with(raw_path=merge_raw_path) async def __aexit__( self, exc_type: type[BaseException] | None, exc: BaseException | None, exc_tb: TracebackType | None ) -> None: if self.__connection is not None: await self.__connection.close() class RealtimeConnection: """Represents a live WebSocket connection to the Realtime API""" session: RealtimeSessionResource response: RealtimeResponseResource input_audio_buffer: RealtimeInputAudioBufferResource conversation: RealtimeConversationResource output_audio_buffer: RealtimeOutputAudioBufferResource _connection: WebSocketConnection def __init__( self, connection: WebSocketConnection, *, make_ws: Callable[[Query, Headers], WebSocketConnection] | None = None, on_reconnecting: Callable[[ReconnectingEvent], ReconnectingOverrides | None] | None = None, max_retries: int = 5, initial_delay: float = 0.5, max_delay: float = 8.0, extra_query: Query = {}, extra_headers: Headers = {}, send_queue: SendQueue | None = None, ) -> None: self._connection = connection self._make_ws = make_ws self._on_reconnecting = on_reconnecting self._max_retries = max_retries self._initial_delay = initial_delay self._max_delay = max_delay self._extra_query = extra_query self._extra_headers = extra_headers self._intentionally_closed = False self._is_reconnecting = False self._send_queue = send_queue or SendQueue() self._event_handler_registry = EventHandlerRegistry(use_lock=True) self.session = RealtimeSessionResource(self) self.response = RealtimeResponseResource(self) self.input_audio_buffer = RealtimeInputAudioBufferResource(self) self.conversation = RealtimeConversationResource(self) self.output_audio_buffer = RealtimeOutputAudioBufferResource(self) def __iter__(self) -> Iterator[RealtimeServerEvent]: """ An infinite-iterator that will continue to yield events until the connection is closed. """ from websockets.exceptions import ConnectionClosedOK, ConnectionClosedError while True: try: yield self.recv() except ConnectionClosedOK: return except ConnectionClosedError as exc: if not self._reconnect(exc): unsent = self._send_queue.drain() if unsent: raise WebSocketConnectionClosedError( "WebSocket connection closed with unsent messages", unsent_messages=unsent, ) from exc raise def recv(self) -> RealtimeServerEvent: """ Receive the next message from the connection and parses it into a `RealtimeServerEvent` object. Canceling this method is safe. There's no risk of losing data. """ return self.parse_event(self.recv_bytes()) def recv_bytes(self) -> bytes: """Receive the next message from the connection as raw bytes. Canceling this method is safe. There's no risk of losing data. If you want to parse the message into a `RealtimeServerEvent` object like `.recv()` does, then you can call `.parse_event(data)`. """ message = self._connection.recv(decode=False) log.debug(f"Received WebSocket message: %s", message) return message def send(self, event: RealtimeClientEvent | RealtimeClientEventParam) -> None: data = ( event.to_json(use_api_names=True, exclude_defaults=True, exclude_unset=True) if isinstance(event, BaseModel) else json.dumps(maybe_transform(event, RealtimeClientEventParam)) ) if self._is_reconnecting: self._send_queue.enqueue(data) return try: self._connection.send(data) except Exception: self._send_queue.enqueue(data) raise def send_raw(self, data: bytes | str) -> None: if self._is_reconnecting: raw = data if isinstance(data, str) else data.decode("utf-8") self._send_queue.enqueue(raw) return self._connection.send(data) def close(self, *, code: int = 1000, reason: str = "") -> None: self._intentionally_closed = True self._connection.close(code=code, reason=reason) def parse_event(self, data: str | bytes) -> RealtimeServerEvent: """ Converts a raw `str` or `bytes` message into a `RealtimeServerEvent` object. This is helpful if you're using `.recv_bytes()`. """ return cast( RealtimeServerEvent, construct_type_unchecked(value=json.loads(data), type_=cast(Any, RealtimeServerEvent)) ) def _reconnect(self, exc: Exception) -> bool: """Attempt to reconnect after a connection failure. Returns ``True`` if a new connection was established, ``False`` if the caller should re-raise the original exception. """ if self._on_reconnecting is None or self._make_ws is None: return False from websockets.exceptions import ConnectionClosedError close_code = 1006 if isinstance(exc, ConnectionClosedError) and exc.rcvd is not None: close_code = exc.rcvd.code if not is_recoverable_close(close_code): return False self._is_reconnecting = True for attempt in range(1, self._max_retries + 1): base_delay = min(self._initial_delay * (2 ** (attempt - 1)), self._max_delay) jitter = 0.75 + random.random() * 0.25 delay = base_delay * jitter event = ReconnectingEvent( attempt=attempt, max_attempts=self._max_retries, delay=delay, close_code=close_code, extra_query=self._extra_query, extra_headers=self._extra_headers, ) try: result = self._on_reconnecting(event) except Exception: self._is_reconnecting = False return False if result is not None and result.get("abort"): self._is_reconnecting = False return False if result is not None: if "extra_query" in result: self._extra_query = result["extra_query"] if "extra_headers" in result: self._extra_headers = result["extra_headers"] log.info( "Reconnecting to WebSocket API (attempt %d/%d) after %.1fs delay", attempt, self._max_retries, delay, ) time.sleep(delay) if self._intentionally_closed: self._is_reconnecting = False return False try: self._connection = self._make_ws(self._extra_query, self._extra_headers) log.info("Reconnected to WebSocket API") self._is_reconnecting = False self._flush_send_queue() return True except Exception: pass self._is_reconnecting = False return False def _flush_send_queue(self) -> None: """Send all queued messages over the current connection.""" try: self._send_queue.flush_sync(lambda data: self._connection.send(data)) except Exception: log.warning("Failed to flush send queue after reconnect", exc_info=True) def on( self, event_type: str, handler: Callable[..., Any] | None = None ) -> Union[RealtimeConnection, Callable[[Callable[..., Any]], Callable[..., Any]]]: """Adds the handler to the end of the handlers list for the given event type. No checks are made to see if the handler has already been added. Multiple calls passing the same combination of event type and handler will result in the handler being added, and called, multiple times. Can be used as a method (returns ``self`` for chaining):: connection.on("conversation.created", my_handler) Or as a decorator:: @connection.on("conversation.created") def my_handler(event): ... """ if handler is not None: self._event_handler_registry.add(event_type, handler) return self def decorator(fn: Callable[..., Any]) -> Callable[..., Any]: self._event_handler_registry.add(event_type, fn) return fn return decorator def off(self, event_type: str, handler: Callable[..., Any]) -> RealtimeConnection: """Remove a previously registered event handler.""" self._event_handler_registry.remove(event_type, handler) return self def once( self, event_type: str, handler: Callable[..., Any] | None = None ) -> Union[RealtimeConnection, Callable[[Callable[..., Any]], Callable[..., Any]]]: """Register a one-time event handler. Automatically removed after first invocation. """ if handler is not None: self._event_handler_registry.add(event_type, handler, once=True) return self def decorator(fn: Callable[..., Any]) -> Callable[..., Any]: self._event_handler_registry.add(event_type, fn, once=True) return fn return decorator def dispatch_events(self) -> None: """Run the event loop, dispatching received events to registered handlers. Blocks the current thread until the connection is closed. This is the push-based alternative to iterating with ``for event in connection``. If an ``"error"`` event arrives and no handler is registered for ``"error"`` or ``"event"``, an ``OpenAIError`` is raised. """ for event in self: event_type = event.type specific = self._event_handler_registry.get_handlers(event_type) generic = self._event_handler_registry.get_handlers("event") if event_type == "error" and not specific and not generic: if isinstance(event, RealtimeErrorEvent): raise OpenAIError(f"WebSocket error: {event}") for handler in specific: handler(event) for handler in generic: handler(event) class RealtimeConnectionManager: """ Context manager over a `RealtimeConnection` that is returned by `realtime.connect()` This context manager ensures that the connection will be closed when it exits. --- Note that if your application doesn't work well with the context manager approach then you can call the `.enter()` method directly to initiate a connection. **Warning**: You must remember to close the connection with `.close()`. ```py connection = client.realtime.connect(...).enter() # ... connection.close() ``` """ def __init__( self, *, client: OpenAI, call_id: str | Omit = omit, model: str | Omit = omit, extra_query: Query, extra_headers: Headers, websocket_connection_options: WebSocketConnectionOptions, on_reconnecting: Callable[[ReconnectingEvent], ReconnectingOverrides | None] | None = None, max_retries: int = 5, initial_delay: float = 0.5, max_delay: float = 8.0, max_queue_size: int = 1_048_576, ) -> None: self.__client = client self.__call_id = call_id self.__model = model self.__connection: RealtimeConnection | None = None self.__extra_query = extra_query self.__extra_headers = extra_headers self.__websocket_connection_options = websocket_connection_options self.__on_reconnecting = on_reconnecting self.__max_retries = max_retries self.__initial_delay = initial_delay self.__max_delay = max_delay self.__send_queue = SendQueue(max_bytes=max_queue_size) self.__event_handler_registry = EventHandlerRegistry(use_lock=True) def send(self, event: RealtimeClientEvent | RealtimeClientEventParam) -> None: """Queue a message to be sent when the connection is established. This can be called before entering the context manager. Queued messages are automatically sent once the WebSocket connection opens. """ data = ( event.to_json(use_api_names=True, exclude_defaults=True, exclude_unset=True) if isinstance(event, BaseModel) else json.dumps(event) ) self.__send_queue.enqueue(data) def on( self, event_type: str, handler: Callable[..., Any] | None = None ) -> Union[RealtimeConnectionManager, Callable[[Callable[..., Any]], Callable[..., Any]]]: """Register an event handler before the connection is established. Handlers are transferred to the connection on enter. Supports the same method and decorator forms as ``RealtimeConnection.on``. """ if handler is not None: self.__event_handler_registry.add(event_type, handler) return self def decorator(fn: Callable[..., Any]) -> Callable[..., Any]: self.__event_handler_registry.add(event_type, fn) return fn return decorator def off(self, event_type: str, handler: Callable[..., Any]) -> RealtimeConnectionManager: """Remove a previously registered event handler.""" self.__event_handler_registry.remove(event_type, handler) return self def once( self, event_type: str, handler: Callable[..., Any] | None = None ) -> Union[RealtimeConnectionManager, Callable[[Callable[..., Any]], Callable[..., Any]]]: """Register a one-time event handler before the connection is established.""" if handler is not None: self.__event_handler_registry.add(event_type, handler, once=True) return self def decorator(fn: Callable[..., Any]) -> Callable[..., Any]: self.__event_handler_registry.add(event_type, fn, once=True) return fn return decorator def __enter__(self) -> RealtimeConnection: """ If your application doesn't work well with the context manager approach then you can call this method directly to initiate a connection. **Warning**: You must remember to close the connection with `.close()`. ```py connection = client.realtime.connect(...).enter() # ... connection.close() ``` """ ws = self._connect_ws(self.__extra_query, self.__extra_headers) self.__connection = RealtimeConnection( ws, make_ws=self._connect_ws if self.__on_reconnecting is not None else None, on_reconnecting=self.__on_reconnecting, max_retries=self.__max_retries, initial_delay=self.__initial_delay, max_delay=self.__max_delay, extra_query=self.__extra_query, extra_headers=self.__extra_headers, send_queue=self.__send_queue, ) self.__event_handler_registry.merge_into(self.__connection._event_handler_registry) self.__connection._flush_send_queue() return self.__connection enter = __enter__ def _connect_ws(self, extra_query: Query, extra_headers: Headers) -> WebSocketConnection: try: from websockets.sync.client import connect except ImportError as exc: raise OpenAIError("You need to install `openai[realtime]` to use this method") from exc self.__client._refresh_api_key() auth_headers = self.__client.auth_headers if self.__call_id is not omit: extra_query = {**extra_query, "call_id": self.__call_id} if is_azure_client(self.__client): model = self.__model if not model: raise OpenAIError("`model` is required for Azure Realtime API") else: url, auth_headers = self.__client._configure_realtime(model, extra_query) else: url = self._prepare_url().copy_with( params={ **self.__client.base_url.params, **({"model": self.__model} if self.__model is not omit else {}), **extra_query, }, ) log.debug("Connecting to %s", url) if self.__websocket_connection_options: log.debug("Connection options: %s", self.__websocket_connection_options) return connect( str(url), user_agent_header=self.__client.user_agent, additional_headers=_merge_mappings( { **auth_headers, }, extra_headers, ), **self.__websocket_connection_options, ) def _prepare_url(self) -> httpx.URL: if self.__client.websocket_base_url is not None: base_url = httpx.URL(self.__client.websocket_base_url) else: scheme = self.__client._base_url.scheme ws_scheme = "ws" if scheme == "http" else "wss" base_url = self.__client._base_url.copy_with(scheme=ws_scheme) merge_raw_path = base_url.raw_path.rstrip(b"/") + b"/realtime" return base_url.copy_with(raw_path=merge_raw_path) def __exit__( self, exc_type: type[BaseException] | None, exc: BaseException | None, exc_tb: TracebackType | None ) -> None: if self.__connection is not None: self.__connection.close() class BaseRealtimeConnectionResource: def __init__(self, connection: RealtimeConnection) -> None: self._connection = connection class RealtimeSessionResource(BaseRealtimeConnectionResource): def update(self, *, session: session_update_event_param.Session, event_id: str | Omit = omit) -> None: """ Send this event to update the session’s configuration. The client may send this event at any time to update any field except for `voice` and `model`. `voice` can be updated only if there have been no other audio outputs yet. When the server receives a `session.update`, it will respond with a `session.updated` event showing the full, effective configuration. Only the fields that are present in the `session.update` are updated. To clear a field like `instructions`, pass an empty string. To clear a field like `tools`, pass an empty array. To clear a field like `turn_detection`, pass `null`. """ self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "session.update", "session": session, "event_id": event_id}), ) ) class RealtimeResponseResource(BaseRealtimeConnectionResource): def create(self, *, event_id: str | Omit = omit, response: RealtimeResponseCreateParamsParam | Omit = omit) -> None: """ This event instructs the server to create a Response, which means triggering model inference. When in Server VAD mode, the server will create Responses automatically. A Response will include at least one Item, and may have two, in which case the second will be a function call. These Items will be appended to the conversation history by default. The server will respond with a `response.created` event, events for Items and content created, and finally a `response.done` event to indicate the Response is complete. The `response.create` event includes inference configuration like `instructions` and `tools`. If these are set, they will override the Session's configuration for this Response only. Responses can be created out-of-band of the default Conversation, meaning that they can have arbitrary input, and it's possible to disable writing the output to the Conversation. Only one Response can write to the default Conversation at a time, but otherwise multiple Responses can be created in parallel. The `metadata` field is a good way to disambiguate multiple simultaneous Responses. Clients can set `conversation` to `none` to create a Response that does not write to the default Conversation. Arbitrary input can be provided with the `input` field, which is an array accepting raw Items and references to existing Items. """ self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "response.create", "event_id": event_id, "response": response}), ) ) def cancel(self, *, event_id: str | Omit = omit, response_id: str | Omit = omit) -> None: """Send this event to cancel an in-progress response. The server will respond with a `response.done` event with a status of `response.status=cancelled`. If there is no response to cancel, the server will respond with an error. It's safe to call `response.cancel` even if no response is in progress, an error will be returned the session will remain unaffected. """ self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "response.cancel", "event_id": event_id, "response_id": response_id}), ) ) class RealtimeInputAudioBufferResource(BaseRealtimeConnectionResource): def clear(self, *, event_id: str | Omit = omit) -> None: """Send this event to clear the audio bytes in the buffer. The server will respond with an `input_audio_buffer.cleared` event. """ self._connection.send( cast(RealtimeClientEventParam, strip_not_given({"type": "input_audio_buffer.clear", "event_id": event_id})) ) def commit(self, *, event_id: str | Omit = omit) -> None: """ Send this event to commit the user input audio buffer, which will create a new user message item in the conversation. This event will produce an error if the input audio buffer is empty. When in Server VAD mode, the client does not need to send this event, the server will commit the audio buffer automatically. Committing the input audio buffer will trigger input audio transcription (if enabled in session configuration), but it will not create a response from the model. The server will respond with an `input_audio_buffer.committed` event. """ self._connection.send( cast(RealtimeClientEventParam, strip_not_given({"type": "input_audio_buffer.commit", "event_id": event_id})) ) def append(self, *, audio: str, event_id: str | Omit = omit) -> None: """Send this event to append audio bytes to the input audio buffer. The audio buffer is temporary storage you can write to and later commit. A "commit" will create a new user message item in the conversation history from the buffer content and clear the buffer. Input audio transcription (if enabled) will be generated when the buffer is committed. If VAD is enabled the audio buffer is used to detect speech and the server will decide when to commit. When Server VAD is disabled, you must commit the audio buffer manually. Input audio noise reduction operates on writes to the audio buffer. The client may choose how much audio to place in each event up to a maximum of 15 MiB, for example streaming smaller chunks from the client may allow the VAD to be more responsive. Unlike most other client events, the server will not send a confirmation response to this event. """ self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "input_audio_buffer.append", "audio": audio, "event_id": event_id}), ) ) class RealtimeConversationResource(BaseRealtimeConnectionResource): @cached_property def item(self) -> RealtimeConversationItemResource: return RealtimeConversationItemResource(self._connection) class RealtimeConversationItemResource(BaseRealtimeConnectionResource): def delete(self, *, item_id: str, event_id: str | Omit = omit) -> None: """Send this event when you want to remove any item from the conversation history. The server will respond with a `conversation.item.deleted` event, unless the item does not exist in the conversation history, in which case the server will respond with an error. """ self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "conversation.item.delete", "item_id": item_id, "event_id": event_id}), ) ) def create( self, *, item: ConversationItemParam, event_id: str | Omit = omit, previous_item_id: str | Omit = omit ) -> None: """ Add a new Item to the Conversation's context, including messages, function calls, and function call responses. This event can be used both to populate a "history" of the conversation and to add new items mid-stream, but has the current limitation that it cannot populate assistant audio messages. If successful, the server will respond with a `conversation.item.created` event, otherwise an `error` event will be sent. """ self._connection.send( cast( RealtimeClientEventParam, strip_not_given( { "type": "conversation.item.create", "item": item, "event_id": event_id, "previous_item_id": previous_item_id, } ), ) ) def truncate(self, *, audio_end_ms: int, content_index: int, item_id: str, event_id: str | Omit = omit) -> None: """Send this event to truncate a previous assistant message’s audio. The server will produce audio faster than realtime, so this event is useful when the user interrupts to truncate audio that has already been sent to the client but not yet played. This will synchronize the server's understanding of the audio with the client's playback. Truncating audio will delete the server-side text transcript to ensure there is not text in the context that hasn't been heard by the user. If successful, the server will respond with a `conversation.item.truncated` event. """ self._connection.send( cast( RealtimeClientEventParam, strip_not_given( { "type": "conversation.item.truncate", "audio_end_ms": audio_end_ms, "content_index": content_index, "item_id": item_id, "event_id": event_id, } ), ) ) def retrieve(self, *, item_id: str, event_id: str | Omit = omit) -> None: """ Send this event when you want to retrieve the server's representation of a specific item in the conversation history. This is useful, for example, to inspect user audio after noise cancellation and VAD. The server will respond with a `conversation.item.retrieved` event, unless the item does not exist in the conversation history, in which case the server will respond with an error. """ self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "conversation.item.retrieve", "item_id": item_id, "event_id": event_id}), ) ) class RealtimeOutputAudioBufferResource(BaseRealtimeConnectionResource): def clear(self, *, event_id: str | Omit = omit) -> None: """**WebRTC/SIP Only:** Emit to cut off the current audio response. This will trigger the server to stop generating audio and emit a `output_audio_buffer.cleared` event. This event should be preceded by a `response.cancel` client event to stop the generation of the current response. [Learn more](https://platform.openai.com/docs/guides/realtime-conversations#client-and-server-events-for-audio-in-webrtc). """ self._connection.send( cast(RealtimeClientEventParam, strip_not_given({"type": "output_audio_buffer.clear", "event_id": event_id})) ) class BaseAsyncRealtimeConnectionResource: def __init__(self, connection: AsyncRealtimeConnection) -> None: self._connection = connection class AsyncRealtimeSessionResource(BaseAsyncRealtimeConnectionResource): async def update(self, *, session: session_update_event_param.Session, event_id: str | Omit = omit) -> None: """ Send this event to update the session’s configuration. The client may send this event at any time to update any field except for `voice` and `model`. `voice` can be updated only if there have been no other audio outputs yet. When the server receives a `session.update`, it will respond with a `session.updated` event showing the full, effective configuration. Only the fields that are present in the `session.update` are updated. To clear a field like `instructions`, pass an empty string. To clear a field like `tools`, pass an empty array. To clear a field like `turn_detection`, pass `null`. """ await self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "session.update", "session": session, "event_id": event_id}), ) ) class AsyncRealtimeResponseResource(BaseAsyncRealtimeConnectionResource): async def create( self, *, event_id: str | Omit = omit, response: RealtimeResponseCreateParamsParam | Omit = omit ) -> None: """ This event instructs the server to create a Response, which means triggering model inference. When in Server VAD mode, the server will create Responses automatically. A Response will include at least one Item, and may have two, in which case the second will be a function call. These Items will be appended to the conversation history by default. The server will respond with a `response.created` event, events for Items and content created, and finally a `response.done` event to indicate the Response is complete. The `response.create` event includes inference configuration like `instructions` and `tools`. If these are set, they will override the Session's configuration for this Response only. Responses can be created out-of-band of the default Conversation, meaning that they can have arbitrary input, and it's possible to disable writing the output to the Conversation. Only one Response can write to the default Conversation at a time, but otherwise multiple Responses can be created in parallel. The `metadata` field is a good way to disambiguate multiple simultaneous Responses. Clients can set `conversation` to `none` to create a Response that does not write to the default Conversation. Arbitrary input can be provided with the `input` field, which is an array accepting raw Items and references to existing Items. """ await self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "response.create", "event_id": event_id, "response": response}), ) ) async def cancel(self, *, event_id: str | Omit = omit, response_id: str | Omit = omit) -> None: """Send this event to cancel an in-progress response. The server will respond with a `response.done` event with a status of `response.status=cancelled`. If there is no response to cancel, the server will respond with an error. It's safe to call `response.cancel` even if no response is in progress, an error will be returned the session will remain unaffected. """ await self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "response.cancel", "event_id": event_id, "response_id": response_id}), ) ) class AsyncRealtimeInputAudioBufferResource(BaseAsyncRealtimeConnectionResource): async def clear(self, *, event_id: str | Omit = omit) -> None: """Send this event to clear the audio bytes in the buffer. The server will respond with an `input_audio_buffer.cleared` event. """ await self._connection.send( cast(RealtimeClientEventParam, strip_not_given({"type": "input_audio_buffer.clear", "event_id": event_id})) ) async def commit(self, *, event_id: str | Omit = omit) -> None: """ Send this event to commit the user input audio buffer, which will create a new user message item in the conversation. This event will produce an error if the input audio buffer is empty. When in Server VAD mode, the client does not need to send this event, the server will commit the audio buffer automatically. Committing the input audio buffer will trigger input audio transcription (if enabled in session configuration), but it will not create a response from the model. The server will respond with an `input_audio_buffer.committed` event. """ await self._connection.send( cast(RealtimeClientEventParam, strip_not_given({"type": "input_audio_buffer.commit", "event_id": event_id})) ) async def append(self, *, audio: str, event_id: str | Omit = omit) -> None: """Send this event to append audio bytes to the input audio buffer. The audio buffer is temporary storage you can write to and later commit. A "commit" will create a new user message item in the conversation history from the buffer content and clear the buffer. Input audio transcription (if enabled) will be generated when the buffer is committed. If VAD is enabled the audio buffer is used to detect speech and the server will decide when to commit. When Server VAD is disabled, you must commit the audio buffer manually. Input audio noise reduction operates on writes to the audio buffer. The client may choose how much audio to place in each event up to a maximum of 15 MiB, for example streaming smaller chunks from the client may allow the VAD to be more responsive. Unlike most other client events, the server will not send a confirmation response to this event. """ await self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "input_audio_buffer.append", "audio": audio, "event_id": event_id}), ) ) class AsyncRealtimeConversationResource(BaseAsyncRealtimeConnectionResource): @cached_property def item(self) -> AsyncRealtimeConversationItemResource: return AsyncRealtimeConversationItemResource(self._connection) class AsyncRealtimeConversationItemResource(BaseAsyncRealtimeConnectionResource): async def delete(self, *, item_id: str, event_id: str | Omit = omit) -> None: """Send this event when you want to remove any item from the conversation history. The server will respond with a `conversation.item.deleted` event, unless the item does not exist in the conversation history, in which case the server will respond with an error. """ await self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "conversation.item.delete", "item_id": item_id, "event_id": event_id}), ) ) async def create( self, *, item: ConversationItemParam, event_id: str | Omit = omit, previous_item_id: str | Omit = omit ) -> None: """ Add a new Item to the Conversation's context, including messages, function calls, and function call responses. This event can be used both to populate a "history" of the conversation and to add new items mid-stream, but has the current limitation that it cannot populate assistant audio messages. If successful, the server will respond with a `conversation.item.created` event, otherwise an `error` event will be sent. """ await self._connection.send( cast( RealtimeClientEventParam, strip_not_given( { "type": "conversation.item.create", "item": item, "event_id": event_id, "previous_item_id": previous_item_id, } ), ) ) async def truncate( self, *, audio_end_ms: int, content_index: int, item_id: str, event_id: str | Omit = omit ) -> None: """Send this event to truncate a previous assistant message’s audio. The server will produce audio faster than realtime, so this event is useful when the user interrupts to truncate audio that has already been sent to the client but not yet played. This will synchronize the server's understanding of the audio with the client's playback. Truncating audio will delete the server-side text transcript to ensure there is not text in the context that hasn't been heard by the user. If successful, the server will respond with a `conversation.item.truncated` event. """ await self._connection.send( cast( RealtimeClientEventParam, strip_not_given( { "type": "conversation.item.truncate", "audio_end_ms": audio_end_ms, "content_index": content_index, "item_id": item_id, "event_id": event_id, } ), ) ) async def retrieve(self, *, item_id: str, event_id: str | Omit = omit) -> None: """ Send this event when you want to retrieve the server's representation of a specific item in the conversation history. This is useful, for example, to inspect user audio after noise cancellation and VAD. The server will respond with a `conversation.item.retrieved` event, unless the item does not exist in the conversation history, in which case the server will respond with an error. """ await self._connection.send( cast( RealtimeClientEventParam, strip_not_given({"type": "conversation.item.retrieve", "item_id": item_id, "event_id": event_id}), ) ) class AsyncRealtimeOutputAudioBufferResource(BaseAsyncRealtimeConnectionResource): async def clear(self, *, event_id: str | Omit = omit) -> None: """**WebRTC/SIP Only:** Emit to cut off the current audio response. This will trigger the server to stop generating audio and emit a `output_audio_buffer.cleared` event. This event should be preceded by a `response.cancel` client event to stop the generation of the current response. [Learn more](https://platform.openai.com/docs/guides/realtime-conversations#client-and-server-events-for-audio-in-webrtc). """ await self._connection.send( cast(RealtimeClientEventParam, strip_not_given({"type": "output_audio_buffer.clear", "event_id": event_id})) )