OpenAI Realtime API

asyncapi: '2.6.0'
info:
  title: OpenAI Realtime API
  version: '2024-10-01'
  description: |
    The OpenAI Realtime API provides low-latency, bidirectional, event-driven
    communication with multimodal models that natively support speech-to-speech,
    text, and audio in a single conversation. This AsyncAPI document describes
    the **WebSocket** transport for the Realtime API, including all documented
    client-to-server events and server-to-client events.

    The Realtime API is currently in beta. Clients must include the
    `OpenAI-Beta: realtime=v1` header when connecting.

    Connection URL:

        wss://api.openai.com/v1/realtime?model={model}

    Events flow over a single full-duplex WebSocket connection. Every event has
    a top-level `type` and most events also carry an `event_id` correlation id.
  contact:
    name: OpenAI
    url: https://platform.openai.com/docs/api-reference/realtime
  license:
    name: Proprietary
    url: https://openai.com/policies/terms-of-use
  termsOfService: https://openai.com/policies/terms-of-use
  x-source:
    derivedFrom:
      - https://platform.openai.com/docs/guides/realtime
      - https://platform.openai.com/docs/api-reference/realtime-client-events
      - https://platform.openai.com/docs/api-reference/realtime-server-events
      - https://github.com/openai/openai-realtime-api-beta

defaultContentType: application/json

servers:
  production:
    url: api.openai.com/v1/realtime
    protocol: wss
    description: |
      OpenAI Realtime WebSocket endpoint. The `model` query parameter selects
      the underlying realtime-capable model (for example
      `gpt-4o-realtime-preview-2024-10-01`).
    variables:
      model:
        description: Realtime model identifier appended as a query string parameter.
        default: gpt-4o-realtime-preview-2024-10-01
        examples:
          - gpt-4o-realtime-preview
          - gpt-4o-realtime-preview-2024-10-01
          - gpt-4o-mini-realtime-preview
    security:
      - bearerAuth: []
        openAiBeta: []
    bindings:
      ws:
        bindingVersion: '0.1.0'
        headers:
          type: object
          required:
            - Authorization
            - OpenAI-Beta
          properties:
            Authorization:
              type: string
              description: Bearer token in the form `Bearer <OPENAI_API_KEY>`.
            OpenAI-Beta:
              type: string
              description: Beta opt-in header. Must be set to `realtime=v1`.
              enum:
                - realtime=v1
            OpenAI-Organization:
              type: string
              description: Optional organization identifier for billing.
            OpenAI-Project:
              type: string
              description: Optional project identifier for billing.
        query:
          type: object
          required:
            - model
          properties:
            model:
              type: string
              description: Realtime model identifier.

channels:

  # ---------------------------------------------------------------------------
  # Client -> Server events
  # ---------------------------------------------------------------------------

  session.update:
    description: |
      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).
    publish:
      operationId: sendSessionUpdate
      summary: Update session configuration.
      message:
        $ref: '#/components/messages/SessionUpdate'

  input_audio_buffer.append:
    description: |
      Send by the client to append base64-encoded audio bytes to the input
      audio buffer. The default audio format is `pcm16` at 24 kHz.
    publish:
      operationId: sendInputAudioBufferAppend
      summary: Append audio bytes to the input buffer.
      message:
        $ref: '#/components/messages/InputAudioBufferAppend'

  input_audio_buffer.commit:
    description: |
      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.
    publish:
      operationId: sendInputAudioBufferCommit
      summary: Commit the input audio buffer.
      message:
        $ref: '#/components/messages/InputAudioBufferCommit'

  input_audio_buffer.clear:
    description: |
      Send by the client to clear the input audio buffer without committing it.
    publish:
      operationId: sendInputAudioBufferClear
      summary: Clear the input audio buffer.
      message:
        $ref: '#/components/messages/InputAudioBufferClear'

  conversation.item.create:
    description: |
      Send by the client to insert a conversation item (a message,
      function_call, or function_call_output) into the conversation history.
    publish:
      operationId: sendConversationItemCreate
      summary: Insert a conversation item.
      message:
        $ref: '#/components/messages/ConversationItemCreate'

  conversation.item.truncate:
    description: |
      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.
    publish:
      operationId: sendConversationItemTruncate
      summary: Truncate an in-progress assistant item.
      message:
        $ref: '#/components/messages/ConversationItemTruncate'

  conversation.item.delete:
    description: |
      Send by the client to delete a conversation item by id.
    publish:
      operationId: sendConversationItemDelete
      summary: Delete a conversation item.
      message:
        $ref: '#/components/messages/ConversationItemDelete'

  response.create:
    description: |
      Send by the client to instruct the model to generate a response.
      Optionally overrides the session configuration for this single response.
    publish:
      operationId: sendResponseCreate
      summary: Trigger a model response.
      message:
        $ref: '#/components/messages/ResponseCreate'

  response.cancel:
    description: |
      Send by the client to cancel an in-progress response.
    publish:
      operationId: sendResponseCancel
      summary: Cancel an in-progress response.
      message:
        $ref: '#/components/messages/ResponseCancel'

  # ---------------------------------------------------------------------------
  # Server -> Client events
  # ---------------------------------------------------------------------------

  error:
    description: |
      Server-emitted error envelope. Sent whenever a client event is invalid
      or the server encounters a problem processing a request.
    subscribe:
      operationId: receiveError
      summary: Receive an error event.
      message:
        $ref: '#/components/messages/Error'

  session.created:
    description: |
      Emitted by the server immediately after the WebSocket connection is
      authenticated. Contains the initial session configuration.
    subscribe:
      operationId: receiveSessionCreated
      summary: Receive session.created.
      message:
        $ref: '#/components/messages/SessionCreated'

  session.updated:
    description: |
      Emitted after the server applies a `session.update` from the client.
    subscribe:
      operationId: receiveSessionUpdated
      summary: Receive session.updated.
      message:
        $ref: '#/components/messages/SessionUpdated'

  conversation.created:
    description: |
      Emitted by the server when a new conversation is created on the
      session.
    subscribe:
      operationId: receiveConversationCreated
      summary: Receive conversation.created.
      message:
        $ref: '#/components/messages/ConversationCreated'

  conversation.item.created:
    description: |
      Emitted when a new conversation item has been added (either by the
      client or by the model generating a response).
    subscribe:
      operationId: receiveConversationItemCreated
      summary: Receive conversation.item.created.
      message:
        $ref: '#/components/messages/ConversationItemCreated'

  conversation.item.input_audio_transcription.completed:
    description: |
      Emitted when input audio transcription for a user audio item has
      completed (requires `input_audio_transcription` enabled on the session).
    subscribe:
      operationId: receiveInputAudioTranscriptionCompleted
      summary: Receive input_audio_transcription.completed.
      message:
        $ref: '#/components/messages/InputAudioTranscriptionCompleted'

  conversation.item.input_audio_transcription.failed:
    description: |
      Emitted when input audio transcription fails for a user audio item.
    subscribe:
      operationId: receiveInputAudioTranscriptionFailed
      summary: Receive input_audio_transcription.failed.
      message:
        $ref: '#/components/messages/InputAudioTranscriptionFailed'

  conversation.item.truncated:
    description: |
      Emitted after the server applies a `conversation.item.truncate` request
      from the client.
    subscribe:
      operationId: receiveConversationItemTruncated
      summary: Receive conversation.item.truncated.
      message:
        $ref: '#/components/messages/ConversationItemTruncated'

  conversation.item.deleted:
    description: |
      Emitted after the server applies a `conversation.item.delete` request.
    subscribe:
      operationId: receiveConversationItemDeleted
      summary: Receive conversation.item.deleted.
      message:
        $ref: '#/components/messages/ConversationItemDeleted'

  input_audio_buffer.committed:
    description: |
      Emitted when the input audio buffer is committed (either explicitly by
      the client via `input_audio_buffer.commit`, or implicitly by server VAD).
    subscribe:
      operationId: receiveInputAudioBufferCommitted
      summary: Receive input_audio_buffer.committed.
      message:
        $ref: '#/components/messages/InputAudioBufferCommitted'

  input_audio_buffer.cleared:
    description: |
      Emitted after the server clears the input audio buffer.
    subscribe:
      operationId: receiveInputAudioBufferCleared
      summary: Receive input_audio_buffer.cleared.
      message:
        $ref: '#/components/messages/InputAudioBufferCleared'

  input_audio_buffer.speech_started:
    description: |
      Emitted in server VAD mode when speech is detected starting in the
      input audio buffer.
    subscribe:
      operationId: receiveSpeechStarted
      summary: Receive input_audio_buffer.speech_started.
      message:
        $ref: '#/components/messages/InputAudioBufferSpeechStarted'

  input_audio_buffer.speech_stopped:
    description: |
      Emitted in server VAD mode when speech is detected stopping in the
      input audio buffer.
    subscribe:
      operationId: receiveSpeechStopped
      summary: Receive input_audio_buffer.speech_stopped.
      message:
        $ref: '#/components/messages/InputAudioBufferSpeechStopped'

  response.created:
    description: |
      Emitted when the server begins generating a response after a
      `response.create` (explicit) or after server VAD commits a user turn.
    subscribe:
      operationId: receiveResponseCreated
      summary: Receive response.created.
      message:
        $ref: '#/components/messages/ResponseCreated'

  response.done:
    description: |
      Emitted when a response has finished (status `completed`, `cancelled`,
      `failed`, or `incomplete`). Carries usage and final output items.
    subscribe:
      operationId: receiveResponseDone
      summary: Receive response.done.
      message:
        $ref: '#/components/messages/ResponseDone'

  response.output_item.added:
    description: |
      Emitted when a new output item is added to a response.
    subscribe:
      operationId: receiveResponseOutputItemAdded
      summary: Receive response.output_item.added.
      message:
        $ref: '#/components/messages/ResponseOutputItemAdded'

  response.output_item.done:
    description: |
      Emitted when an output item on a response is complete.
    subscribe:
      operationId: receiveResponseOutputItemDone
      summary: Receive response.output_item.done.
      message:
        $ref: '#/components/messages/ResponseOutputItemDone'

  response.content_part.added:
    description: |
      Emitted when a new content part (text, audio, or transcript) is added
      to an output item.
    subscribe:
      operationId: receiveResponseContentPartAdded
      summary: Receive response.content_part.added.
      message:
        $ref: '#/components/messages/ResponseContentPartAdded'

  response.content_part.done:
    description: |
      Emitted when a content part on an output item is complete.
    subscribe:
      operationId: receiveResponseContentPartDone
      summary: Receive response.content_part.done.
      message:
        $ref: '#/components/messages/ResponseContentPartDone'

  response.text.delta:
    description: |
      Streaming text delta for a `text` content part on an assistant item.
    subscribe:
      operationId: receiveResponseTextDelta
      summary: Receive response.text.delta.
      message:
        $ref: '#/components/messages/ResponseTextDelta'

  response.text.done:
    description: |
      Emitted when a `text` content part is fully generated.
    subscribe:
      operationId: receiveResponseTextDone
      summary: Receive response.text.done.
      message:
        $ref: '#/components/messages/ResponseTextDone'

  response.audio_transcript.delta:
    description: |
      Streaming transcript delta for an `audio` content part on an
      assistant item.
    subscribe:
      operationId: receiveResponseAudioTranscriptDelta
      summary: Receive response.audio_transcript.delta.
      message:
        $ref: '#/components/messages/ResponseAudioTranscriptDelta'

  response.audio_transcript.done:
    description: |
      Emitted when the transcript for an `audio` content part is fully
      generated.
    subscribe:
      operationId: receiveResponseAudioTranscriptDone
      summary: Receive response.audio_transcript.done.
      message:
        $ref: '#/components/messages/ResponseAudioTranscriptDone'

  response.audio.delta:
    description: |
      Streaming base64-encoded audio delta for an `audio` content part on an
      assistant item.
    subscribe:
      operationId: receiveResponseAudioDelta
      summary: Receive response.audio.delta.
      message:
        $ref: '#/components/messages/ResponseAudioDelta'

  response.audio.done:
    description: |
      Emitted when an `audio` content part is fully generated. No final
      base64 payload is included; clients reassemble from the deltas.
    subscribe:
      operationId: receiveResponseAudioDone
      summary: Receive response.audio.done.
      message:
        $ref: '#/components/messages/ResponseAudioDone'

  response.function_call_arguments.delta:
    description: |
      Streaming delta for a tool/function call's `arguments` string.
    subscribe:
      operationId: receiveResponseFunctionCallArgumentsDelta
      summary: Receive response.function_call_arguments.delta.
      message:
        $ref: '#/components/messages/ResponseFunctionCallArgumentsDelta'

  response.function_call_arguments.done:
    description: |
      Emitted when the `arguments` string for a function call is complete.
    subscribe:
      operationId: receiveResponseFunctionCallArgumentsDone
      summary: Receive response.function_call_arguments.done.
      message:
        $ref: '#/components/messages/ResponseFunctionCallArgumentsDone'

  rate_limits.updated:
    description: |
      Emitted periodically with the current rate limit state for the
      connection (requests and tokens, remaining and reset_seconds).
    subscribe:
      operationId: receiveRateLimitsUpdated
      summary: Receive rate_limits.updated.
      message:
        $ref: '#/components/messages/RateLimitsUpdated'

components:

  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      bearerFormat: OpenAI API Key
      description: |
        Bearer token authentication using an OpenAI API key. The header
        `Authorization: Bearer <OPENAI_API_KEY>` must be sent on the
        WebSocket upgrade request.
    openAiBeta:
      type: apiKey
      in: header
      name: OpenAI-Beta
      description: |
        Beta opt-in header. Must be set to `realtime=v1` for the Realtime
        API while it remains in beta.

  messages:

    # ----- Client -> Server -------------------------------------------------

    SessionUpdate:
      name: SessionUpdate
      title: session.update
      summary: Update session configuration.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SessionUpdateEvent'

    InputAudioBufferAppend:
      name: InputAudioBufferAppend
      title: input_audio_buffer.append
      summary: Append audio bytes to the input buffer.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/InputAudioBufferAppendEvent'

    InputAudioBufferCommit:
      name: InputAudioBufferCommit
      title: input_audio_buffer.commit
      summary: Commit the input audio buffer.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/InputAudioBufferCommitEvent'

    InputAudioBufferClear:
      name: InputAudioBufferClear
      title: input_audio_buffer.clear
      summary: Clear the input audio buffer.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/InputAudioBufferClearEvent'

    ConversationItemCreate:
      name: ConversationItemCreate
      title: conversation.item.create
      summary: Insert a conversation item.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ConversationItemCreateEvent'

    ConversationItemTruncate:
      name: ConversationItemTruncate
      title: conversation.item.truncate
      summary: Truncate an assistant item's audio.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ConversationItemTruncateEvent'

    ConversationItemDelete:
      name: ConversationItemDelete
      title: conversation.item.delete
      summary: Delete a conversation item.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ConversationItemDeleteEvent'

    ResponseCreate:
      name: ResponseCreate
      title: response.create
      summary: Trigger a model response.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseCreateEvent'

    ResponseCancel:
      name: ResponseCancel
      title: response.cancel
      summary: Cancel an in-progress response.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseCancelEvent'

    # ----- Server -> Client -------------------------------------------------

    Error:
      name: Error
      title: error
      summary: Server error.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ErrorEvent'

    SessionCreated:
      name: SessionCreated
      title: session.created
      summary: Session has been created.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SessionCreatedEvent'

    SessionUpdated:
      name: SessionUpdated
      title: session.updated
      summary: Session configuration updated.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/SessionUpdatedEvent'

    ConversationCreated:
      name: ConversationCreated
      title: conversation.created
      summary: Conversation created.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ConversationCreatedEvent'

    ConversationItemCreated:
      name: ConversationItemCreated
      title: conversation.item.created
      summary: Conversation item created.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ConversationItemCreatedEvent'

    InputAudioTranscriptionCompleted:
      name: InputAudioTranscriptionCompleted
      title: conversation.item.input_audio_transcription.completed
      summary: Input audio transcription completed.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/InputAudioTranscriptionCompletedEvent'

    InputAudioTranscriptionFailed:
      name: InputAudioTranscriptionFailed
      title: conversation.item.input_audio_transcription.failed
      summary: Input audio transcription failed.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/InputAudioTranscriptionFailedEvent'

    ConversationItemTruncated:
      name: ConversationItemTruncated
      title: conversation.item.truncated
      summary: Conversation item truncated.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ConversationItemTruncatedEvent'

    ConversationItemDeleted:
      name: ConversationItemDeleted
      title: conversation.item.deleted
      summary: Conversation item deleted.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ConversationItemDeletedEvent'

    InputAudioBufferCommitted:
      name: InputAudioBufferCommitted
      title: input_audio_buffer.committed
      summary: Input audio buffer committed.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/InputAudioBufferCommittedEvent'

    InputAudioBufferCleared:
      name: InputAudioBufferCleared
      title: input_audio_buffer.cleared
      summary: Input audio buffer cleared.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/InputAudioBufferClearedEvent'

    InputAudioBufferSpeechStarted:
      name: InputAudioBufferSpeechStarted
      title: input_audio_buffer.speech_started
      summary: VAD speech started.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/InputAudioBufferSpeechStartedEvent'

    InputAudioBufferSpeechStopped:
      name: InputAudioBufferSpeechStopped
      title: input_audio_buffer.speech_stopped
      summary: VAD speech stopped.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/InputAudioBufferSpeechStoppedEvent'

    ResponseCreated:
      name: ResponseCreated
      title: response.created
      summary: Response generation started.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseCreatedEvent'

    ResponseDone:
      name: ResponseDone
      title: response.done
      summary: Response generation finished.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseDoneEvent'

    ResponseOutputItemAdded:
      name: ResponseOutputItemAdded
      title: response.output_item.added
      summary: New output item added to response.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseOutputItemAddedEvent'

    ResponseOutputItemDone:
      name: ResponseOutputItemDone
      title: response.output_item.done
      summary: Output item on response complete.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseOutputItemDoneEvent'

    ResponseContentPartAdded:
      name: ResponseContentPartAdded
      title: response.content_part.added
      summary: Content part added to output item.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseContentPartAddedEvent'

    ResponseContentPartDone:
      name: ResponseContentPartDone
      title: response.content_part.done
      summary: Content part on output item complete.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseContentPartDoneEvent'

    ResponseTextDelta:
      name: ResponseTextDelta
      title: response.text.delta
      summary: Text delta for assistant message.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseTextDeltaEvent'

    ResponseTextDone:
      name: ResponseTextDone
      title: response.text.done
      summary: Text content part complete.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseTextDoneEvent'

    ResponseAudioTranscriptDelta:
      name: ResponseAudioTranscriptDelta
      title: response.audio_transcript.delta
      summary: Transcript delta for audio content part.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseAudioTranscriptDeltaEvent'

    ResponseAudioTranscriptDone:
      name: ResponseAudioTranscriptDone
      title: response.audio_transcript.done
      summary: Transcript for audio content part complete.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseAudioTranscriptDoneEvent'

    ResponseAudioDelta:
      name: ResponseAudioDelta
      title: response.audio.delta
      summary: Base64 audio delta for audio content part.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseAudioDeltaEvent'

    ResponseAudioDone:
      name: ResponseAudioDone
      title: response.audio.done
      summary: Audio content part complete.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseAudioDoneEvent'

    ResponseFunctionCallArgumentsDelta:
      name: ResponseFunctionCallArgumentsDelta
      title: response.function_call_arguments.delta
      summary: Function-call arguments delta.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseFunctionCallArgumentsDeltaEvent'

    ResponseFunctionCallArgumentsDone:
      name: ResponseFunctionCallArgumentsDone
      title: response.function_call_arguments.done
      summary: Function-call arguments complete.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/ResponseFunctionCallArgumentsDoneEvent'

    RateLimitsUpdated:
      name: RateLimitsUpdated
      title: rate_limits.updated
      summary: Current rate limit state.
      contentType: application/json
      payload:
        $ref: '#/components/schemas/RateLimitsUpdatedEvent'

  schemas:

    # ----- Shared resources ------------------------------------------------

    AudioFormat:
      type: string
      enum:
        - pcm16
        - g711_ulaw
        - g711_alaw
      description: Supported realtime audio codecs.

    Voice:
      type: string
      enum:
        - alloy
        - ash
        - ballad
        - coral
        - echo
        - sage
        - shimmer
        - verse
      description: Realtime model voice.

    Modality:
      type: string
      enum:
        - text
        - audio

    TurnDetection:
      type: object
      description: Server-side voice activity detection config. Set to `null` to disable.
      nullable: true
      properties:
        type:
          type: string
          enum:
            - server_vad
        threshold:
          type: number
          minimum: 0
          maximum: 1
          description: Activation threshold (default 0.5).
        prefix_padding_ms:
          type: integer
          description: Audio (ms) before speech start to include (default 300).
        silence_duration_ms:
          type: integer
          description: Silence (ms) before a turn is considered ended (default 200).
      required:
        - type

    InputAudioTranscription:
      type: object
      description: |
        Input audio transcription config. Set to `null` to disable. When
        enabled, the server emits `conversation.item.input_audio_transcription.completed`
        for each user audio item.
      nullable: true
      properties:
        model:
          type: string
          enum:
            - whisper-1
      required:
        - model

    ToolDefinition:
      type: object
      properties:
        type:
          type: string
          enum:
            - function
        name:
          type: string
        description:
          type: string
        parameters:
          type: object
          description: JSON Schema describing the tool's parameters.
      required:
        - name
        - parameters

    ToolChoice:
      oneOf:
        - type: string
          enum:
            - auto
            - none
            - required
        - type: object
          properties:
            type:
              type: string
              enum:
                - function
            name:
              type: string
          required:
            - type
            - name

    MaxResponseOutputTokens:
      oneOf:
        - type: integer
          minimum: 1
        - type: string
          enum:
            - inf
      description: Max output tokens per response, or `inf` for unlimited.

    Session:
      type: object
      description: Server-side session configuration.
      properties:
        id:
          type: string
        object:
          type: string
          enum:
            - realtime.session
        model:
          type: string
        modalities:
          type: array
          items:
            $ref: '#/components/schemas/Modality'
        instructions:
          type: string
        voice:
          $ref: '#/components/schemas/Voice'
        input_audio_format:
          $ref: '#/components/schemas/AudioFormat'
        output_audio_format:
          $ref: '#/components/schemas/AudioFormat'
        input_audio_transcription:
          $ref: '#/components/schemas/InputAudioTranscription'
        turn_detection:
          $ref: '#/components/schemas/TurnDetection'
        tools:
          type: array
          items:
            $ref: '#/components/schemas/ToolDefinition'
        tool_choice:
          $ref: '#/components/schemas/ToolChoice'
        temperature:
          type: number
        max_response_output_tokens:
          $ref: '#/components/schemas/MaxResponseOutputTokens'

    SessionPatch:
      type: object
      description: |
        Subset of session fields that may be supplied on `session.update`.
        Only included properties are modified.
      properties:
        modalities:
          type: array
          items:
            $ref: '#/components/schemas/Modality'
        instructions:
          type: string
        voice:
          $ref: '#/components/schemas/Voice'
        input_audio_format:
          $ref: '#/components/schemas/AudioFormat'
        output_audio_format:
          $ref: '#/components/schemas/AudioFormat'
        input_audio_transcription:
          $ref: '#/components/schemas/InputAudioTranscription'
        turn_detection:
          $ref: '#/components/schemas/TurnDetection'
        tools:
          type: array
          items:
            $ref: '#/components/schemas/ToolDefinition'
        tool_choice:
          $ref: '#/components/schemas/ToolChoice'
        temperature:
          type: number
        max_response_output_tokens:
          $ref: '#/components/schemas/MaxResponseOutputTokens'

    Conversation:
      type: object
      properties:
        id:
          type: string
        object:
          type: string
          enum:
            - realtime.conversation

    ItemStatus:
      type: string
      enum:
        - in_progress
        - completed
        - incomplete

    InputTextContent:
      type: object
      properties:
        type:
          type: string
          enum:
            - input_text
        text:
          type: string
      required:
        - type
        - text

    InputAudioContent:
      type: object
      properties:
        type:
          type: string
          enum:
            - input_audio
        audio:
          type: string
          description: Base64-encoded audio bytes in the session's `input_audio_format`.
        transcript:
          type: string
          nullable: true
      required:
        - type

    TextContent:
      type: object
      properties:
        type:
          type: string
          enum:
            - text
        text:
          type: string
      required:
        - type
        - text

    AudioContent:
      type: object
      properties:
        type:
          type: string
          enum:
            - audio
        audio:
          type: string
          description: Base64-encoded audio bytes in the session's `output_audio_format`.
        transcript:
          type: string
          nullable: true
      required:
        - type

    ContentPart:


# --- truncated at 32 KB (53 KB total) ---
# Full source: https://raw.githubusercontent.com/api-evangelist/openai/refs/heads/main/asyncapi/openai-realtime-asyncapi.yml