OpenAI

Channels

• conversation.created
  Emitted by the server when a new conversation is created on the session.
  OpenAI Realtime API
• conversation.item.create
  Send by the client to insert a conversation item (a message, function_call, or function_call_output) into the conversation history.
  OpenAI Realtime API
• conversation.item.created
  Emitted when a new conversation item has been added (either by the client or by the model generating a response).
  OpenAI Realtime API
• conversation.item.delete
  Send by the client to delete a conversation item by id.
  OpenAI Realtime API
• conversation.item.deleted
  Emitted after the server applies a `conversation.item.delete` request.
  OpenAI Realtime API
• conversation.item.input_audio_transcription.completed
  Emitted when input audio transcription for a user audio item has completed (requires `input_audio_transcription` enabled on the session).
  OpenAI Realtime API
• conversation.item.input_audio_transcription.failed
  Emitted when input audio transcription fails for a user audio item.
  OpenAI Realtime API
• conversation.item.truncate
  Send by the client to truncate the assistant audio of an in-progress response item. Used for interruption: audio after `audio_end_ms` is discarded and any text after that point is cleared.
  OpenAI Realtime API
• conversation.item.truncated
  Emitted after the server applies a `conversation.item.truncate` request from the client.
  OpenAI Realtime API
• error
  Server-emitted error envelope. Sent whenever a client event is invalid or the server encounters a problem processing a request.
  OpenAI Realtime API
• input_audio_buffer.append
  Send by the client to append base64-encoded audio bytes to the input audio buffer. The default audio format is `pcm16` at 24 kHz.
  OpenAI Realtime API
• input_audio_buffer.clear
  Send by the client to clear the input audio buffer without committing it.
  OpenAI Realtime API
• input_audio_buffer.cleared
  Emitted after the server clears the input audio buffer.
  OpenAI Realtime API
• input_audio_buffer.commit
  Send by the client to commit the input audio buffer to the conversation as a user message. Required in non-VAD modes before requesting a response.
  OpenAI Realtime API
• input_audio_buffer.committed
  Emitted when the input audio buffer is committed (either explicitly by the client via `input_audio_buffer.commit`, or implicitly by server VAD).
  OpenAI Realtime API
• input_audio_buffer.speech_started
  Emitted in server VAD mode when speech is detected starting in the input audio buffer.
  OpenAI Realtime API
• input_audio_buffer.speech_stopped
  Emitted in server VAD mode when speech is detected stopping in the input audio buffer.
  OpenAI Realtime API
• rate_limits.updated
  Emitted periodically with the current rate limit state for the connection (requests and tokens, remaining and reset_seconds).
  OpenAI Realtime API
• response.audio.delta
  Streaming base64-encoded audio delta for an `audio` content part on an assistant item.
  OpenAI Realtime API
• response.audio.done
  Emitted when an `audio` content part is fully generated. No final base64 payload is included; clients reassemble from the deltas.
  OpenAI Realtime API
• response.audio_transcript.delta
  Streaming transcript delta for an `audio` content part on an assistant item.
  OpenAI Realtime API
• response.audio_transcript.done
  Emitted when the transcript for an `audio` content part is fully generated.
  OpenAI Realtime API
• response.cancel
  Send by the client to cancel an in-progress response.
  OpenAI Realtime API
• response.content_part.added
  Emitted when a new content part (text, audio, or transcript) is added to an output item.
  OpenAI Realtime API
• response.content_part.done
  Emitted when a content part on an output item is complete.
  OpenAI Realtime API
• response.create
  Send by the client to instruct the model to generate a response. Optionally overrides the session configuration for this single response.
  OpenAI Realtime API
• response.created
  Emitted when the server begins generating a response after a `response.create` (explicit) or after server VAD commits a user turn.
  OpenAI Realtime API
• response.done
  Emitted when a response has finished (status `completed`, `cancelled`, `failed`, or `incomplete`). Carries usage and final output items.
  OpenAI Realtime API
• response.function_call_arguments.delta
  Streaming delta for a tool/function call's `arguments` string.
  OpenAI Realtime API
• response.function_call_arguments.done
  Emitted when the `arguments` string for a function call is complete.
  OpenAI Realtime API
• response.output_item.added
  Emitted when a new output item is added to a response.
  OpenAI Realtime API
• response.output_item.done
  Emitted when an output item on a response is complete.
  OpenAI Realtime API
• response.text.delta
  Streaming text delta for a `text` content part on an assistant item.
  OpenAI Realtime API
• response.text.done
  Emitted when a `text` content part is fully generated.
  OpenAI Realtime API
• session.created
  Emitted by the server immediately after the WebSocket connection is authenticated. Contains the initial session configuration.
  OpenAI Realtime API
• session.update
  Send by the client to update the session's default configuration (modalities, instructions, voice, audio formats, turn detection, tools, tool_choice, temperature, max_response_output_tokens).
  OpenAI Realtime API
• session.updated
  Emitted after the server applies a `session.update` from the client.
  OpenAI Realtime API