Client Secrets

Create client secret

client.Realtime.ClientSecrets.New(ctx, body) (*ClientSecretNewResponse, error)

post /realtime/client_secrets

Create a Realtime client secret with an associated session configuration.

Client secrets are short-lived tokens that can be passed to a client app, such as a web frontend or mobile client, which grants access to the Realtime API without leaking your main API key. You can configure a custom TTL for each client secret.

You can also attach session configuration options to the client secret, which will be applied to any sessions created using that client secret, but these can also be overridden by the client connection.

Learn more about authentication with client secrets over WebRTC.

Returns the created client secret and the effective session object. The client secret is a string that looks like ek_1234.

Parameters

body ClientSecretNewParams
- ExpiresAfter param.Field[ClientSecretNewParamsExpiresAfter]
  
  Configuration for the client secret expiration. Expiration refers to the time after which a client secret will no longer be valid for creating sessions. The session itself may continue after that time once started. A secret can be used to create multiple sessions until it expires.
  - Anchor string
    
    The anchor point for the client secret expiration, meaning that seconds will be added to the created_at time of the client secret to produce an expiration timestamp. Only created_at is currently supported.
    - const ClientSecretNewParamsExpiresAfterAnchorCreatedAt ClientSecretNewParamsExpiresAfterAnchor = "created_at"
  - Seconds int64
    
    The number of seconds from the anchor point to the expiration. Select a value between 10 and 7200 (2 hours). This default to 600 seconds (10 minutes) if not specified.
- Session param.Field[ClientSecretNewParamsSessionUnion]
  
  Session configuration to use for the client secret. Choose either a realtime session or a transcription session.
  - type RealtimeSessionCreateRequest struct{…}
    
    Realtime session object configuration.
    - Type Realtime
      
      The type of session to create. Always realtime for the Realtime API.
      - const RealtimeRealtime Realtime = "realtime"
    - Audio RealtimeAudioConfig
      
      Configuration for input and output audio.
      - Input RealtimeAudioConfigInput
        
        Format RealtimeAudioFormatsUnion
        
        The format of the input audio.
        
        type RealtimeAudioFormatsAudioPCM struct{…}
        
        The PCM audio format. Only a 24kHz sample rate is supported.
        
        Rate int64
        
        The sample rate of the audio. Always 24000.
        
        const RealtimeAudioFormatsAudioPCMRate24000 RealtimeAudioFormatsAudioPCMRate = 24000
        
        Type string
        
        The audio format. Always audio/pcm.
        
        const RealtimeAudioFormatsAudioPCMTypeAudioPCM RealtimeAudioFormatsAudioPCMType = "audio/pcm"
        
        type RealtimeAudioFormatsAudioPCMU struct{…}
        
        The G.711 μ-law format.
        
        Type string
        
        The audio format. Always audio/pcmu.
        
        const RealtimeAudioFormatsAudioPCMUTypeAudioPCMU RealtimeAudioFormatsAudioPCMUType = "audio/pcmu"
        
        type RealtimeAudioFormatsAudioPCMA struct{…}
        
        The G.711 A-law format.
        
        Type string
        
        The audio format. Always audio/pcma.
        
        const RealtimeAudioFormatsAudioPCMATypeAudioPCMA RealtimeAudioFormatsAudioPCMAType = "audio/pcma"
        
        NoiseReduction RealtimeAudioConfigInputNoiseReduction
        
        Configuration for input audio noise reduction. This can be set to null to turn off. Noise reduction filters audio added to the input audio buffer before it is sent to VAD and the model. Filtering the audio can improve VAD and turn detection accuracy (reducing false positives) and model performance by improving perception of the input audio.
        
        Type NoiseReductionType
        
        Type of noise reduction. near_field is for close-talking microphones such as headphones, far_field is for far-field microphones such as laptop or conference room microphones.
        
        const NoiseReductionTypeNearField NoiseReductionType = "near_field"
        
        const NoiseReductionTypeFarField NoiseReductionType = "far_field"
        
        Transcription AudioTranscription
        
        Configuration for input audio transcription, defaults to off and can be set to null to turn off once on. Input audio transcription is not native to the model, since the model consumes audio directly. Transcription runs asynchronously through the /audio/transcriptions endpoint and should be treated as guidance of input audio content rather than precisely what the model heard. The client can optionally set the language and prompt for transcription, these offer additional guidance to the transcription service.
        
        Delay AudioTranscriptionDelay
        
        Controls how long the model waits before emitting transcription text. Higher values can improve transcription accuracy at the cost of latency. Only supported with gpt-realtime-whisper in GA Realtime sessions.
        
        const AudioTranscriptionDelayMinimal AudioTranscriptionDelay = "minimal"
        
        const AudioTranscriptionDelayLow AudioTranscriptionDelay = "low"
        
        const AudioTranscriptionDelayMedium AudioTranscriptionDelay = "medium"
        
        const AudioTranscriptionDelayHigh AudioTranscriptionDelay = "high"
        
        const AudioTranscriptionDelayXhigh AudioTranscriptionDelay = "xhigh"
        
        Language string
        
        The language of the input audio. Supplying the input language in ISO-639-1 (e.g. en) format will improve accuracy and latency.
        
        Model AudioTranscriptionModel
        
        The model to use for transcription. Current options are whisper-1, gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, and gpt-realtime-whisper. Use gpt-4o-transcribe-diarize when you need diarization with speaker labels.
        
        string
        
        type AudioTranscriptionModel string
        
        The model to use for transcription. Current options are whisper-1, gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, and gpt-realtime-whisper. Use gpt-4o-transcribe-diarize when you need diarization with speaker labels.
        
        const AudioTranscriptionModelWhisper1 AudioTranscriptionModel = "whisper-1"
        
        const AudioTranscriptionModelGPT4oMiniTranscribe AudioTranscriptionModel = "gpt-4o-mini-transcribe"
        
        const AudioTranscriptionModelGPT4oMiniTranscribe2025_12_15 AudioTranscriptionModel = "gpt-4o-mini-transcribe-2025-12-15"
        
        const AudioTranscriptionModelGPT4oTranscribe AudioTranscriptionModel = "gpt-4o-transcribe"
        
        const AudioTranscriptionModelGPT4oTranscribeDiarize AudioTranscriptionModel = "gpt-4o-transcribe-diarize"
        
        const AudioTranscriptionModelGPTRealtimeWhisper AudioTranscriptionModel = "gpt-realtime-whisper"
        
        Prompt string
        
        An optional text to guide the model's style or continue a previous audio segment. For whisper-1, the prompt is a list of keywords. For gpt-4o-transcribe models (excluding gpt-4o-transcribe-diarize), the prompt is a free text string, for example "expect words related to technology". Prompt is not supported with gpt-realtime-whisper in GA Realtime sessions.
        
        TurnDetection RealtimeAudioInputTurnDetectionUnion
        
        Configuration for turn detection, ether Server VAD or Semantic VAD. This can be set to null to turn off, in which case the client must manually trigger model response.
        
        Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech.
        
        Semantic VAD is more advanced and uses a turn detection model (in conjunction with VAD) to semantically estimate whether the user has finished speaking, then dynamically sets a timeout based on this probability. For example, if user audio trails off with "uhhm", the model will score a low probability of turn end and wait longer for the user to continue speaking. This can be useful for more natural conversations, but may have a higher latency.
        
        For gpt-realtime-whisper transcription sessions, turn detection must be set to null; VAD is not supported.
        
        RealtimeAudioInputTurnDetectionServerVad
        
        Type ServerVad
        
        Type of turn detection, server_vad to turn on simple Server VAD.
        
        const ServerVadServerVad ServerVad = "server_vad"
        
        CreateResponse bool
        
        Whether or not to automatically generate a response when a VAD stop event occurs. If interrupt_response is set to false this may fail to create a response if the model is already responding.
        
        If both create_response and interrupt_response are set to false, the model will never respond automatically but VAD events will still be emitted.
        
        IdleTimeoutMs int64
        
        Optional timeout after which a model response will be triggered automatically. This is useful for situations in which a long pause from the user is unexpected, such as a phone call. The model will effectively prompt the user to continue the conversation based on the current context.
        
        The timeout value will be applied after the last model response's audio has finished playing, i.e. it's set to the response.done time plus audio playback duration.
        
        An input_audio_buffer.timeout_triggered event (plus events associated with the Response) will be emitted when the timeout is reached. Idle timeout is currently only supported for server_vad mode.
        
        InterruptResponse bool
        
        Whether or not to automatically interrupt (cancel) any ongoing response with output to the default conversation (i.e. conversation of auto) when a VAD start event occurs. If true then the response will be cancelled, otherwise it will continue until complete.
        
        If both create_response and interrupt_response are set to false, the model will never respond automatically but VAD events will still be emitted.
        
        PrefixPaddingMs int64
        
        Used only for server_vad mode. Amount of audio to include before the VAD detected speech (in milliseconds). Defaults to 300ms.
        
        SilenceDurationMs int64
        
        Used only for server_vad mode. Duration of silence to detect speech stop (in milliseconds). Defaults to 500ms. With shorter values the model will respond more quickly, but may jump in on short pauses from the user.
        
        Threshold float64
        
        Used only for server_vad mode. Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A higher threshold will require louder audio to activate the model, and thus might perform better in noisy environments.
        
        RealtimeAudioInputTurnDetectionSemanticVad
        
        Type SemanticVad
        
        Type of turn detection, semantic_vad to turn on Semantic VAD.
        
        const SemanticVadSemanticVad SemanticVad = "semantic_vad"
        
        CreateResponse bool
        
        Whether or not to automatically generate a response when a VAD stop event occurs.
        
        Eagerness string
        
        Used only for semantic_vad mode. The eagerness of the model to respond. low will wait longer for the user to continue speaking, high will respond more quickly. auto is the default and is equivalent to medium. low, medium, and high have max timeouts of 8s, 4s, and 2s respectively.
        
        const RealtimeAudioInputTurnDetectionSemanticVadEagernessLow RealtimeAudioInputTurnDetectionSemanticVadEagerness = "low"
        
        const RealtimeAudioInputTurnDetectionSemanticVadEagernessMedium RealtimeAudioInputTurnDetectionSemanticVadEagerness = "medium"
        
        const RealtimeAudioInputTurnDetectionSemanticVadEagernessHigh RealtimeAudioInputTurnDetectionSemanticVadEagerness = "high"
        
        const RealtimeAudioInputTurnDetectionSemanticVadEagernessAuto RealtimeAudioInputTurnDetectionSemanticVadEagerness = "auto"
        
        InterruptResponse bool
        
        Whether or not to automatically interrupt any ongoing response with output to the default conversation (i.e. conversation of auto) when a VAD start event occurs.
      - Output RealtimeAudioConfigOutput
        
        Format RealtimeAudioFormatsUnion
        
        The format of the output audio.
        
        Speed float64
        
        The speed of the model's spoken response as a multiple of the original speed. 1.0 is the default speed. 0.25 is the minimum speed. 1.5 is the maximum speed. This value can only be changed in between model turns, not while a response is in progress.
        
        This parameter is a post-processing adjustment to the audio after it is generated, it's also possible to prompt the model to speak faster or slower.
        
        Voice RealtimeAudioConfigOutputVoiceUnion
        
        The voice the model uses to respond. Supported built-in voices are alloy, ash, ballad, coral, echo, sage, shimmer, verse, marin, and cedar. You may also provide a custom voice object with an id, for example { "id": "voice_1234" }. Voice cannot be changed during the session once the model has responded with audio at least once. We recommend marin and cedar for best quality.
        
        string
        
        string
        
        const RealtimeAudioConfigOutputVoiceString2Alloy RealtimeAudioConfigOutputVoiceString2 = "alloy"
        
        const RealtimeAudioConfigOutputVoiceString2Ash RealtimeAudioConfigOutputVoiceString2 = "ash"
        
        const RealtimeAudioConfigOutputVoiceString2Ballad RealtimeAudioConfigOutputVoiceString2 = "ballad"
        
        const RealtimeAudioConfigOutputVoiceString2Coral RealtimeAudioConfigOutputVoiceString2 = "coral"
        
        const RealtimeAudioConfigOutputVoiceString2Echo RealtimeAudioConfigOutputVoiceString2 = "echo"
        
        const RealtimeAudioConfigOutputVoiceString2Sage RealtimeAudioConfigOutputVoiceString2 = "sage"
        
        const RealtimeAudioConfigOutputVoiceString2Shimmer RealtimeAudioConfigOutputVoiceString2 = "shimmer"
        
        const RealtimeAudioConfigOutputVoiceString2Verse RealtimeAudioConfigOutputVoiceString2 = "verse"
        
        const RealtimeAudioConfigOutputVoiceString2Marin RealtimeAudioConfigOutputVoiceString2 = "marin"
        
        const RealtimeAudioConfigOutputVoiceString2Cedar RealtimeAudioConfigOutputVoiceString2 = "cedar"
        
        RealtimeAudioConfigOutputVoiceID
        
        ID string
        
        The custom voice ID, e.g. voice_1234.
    - Include []string
      
      Additional fields to include in server outputs.
      
      item.input_audio_transcription.logprobs: Include logprobs for input audio transcription.
      - const RealtimeSessionCreateRequestIncludeItemInputAudioTranscriptionLogprobs RealtimeSessionCreateRequestInclude = "item.input_audio_transcription.logprobs"
    - Instructions string
      
      The default system instructions (i.e. system message) prepended to model calls. This field allows the client to guide the model on desired responses. The model can be instructed on response content and format, (e.g. "be extremely succinct", "act friendly", "here are examples of good responses") and on audio behavior (e.g. "talk quickly", "inject emotion into your voice", "laugh frequently"). The instructions are not guaranteed to be followed by the model, but they provide guidance to the model on the desired behavior.
      
      Note that the server sets default instructions which will be used if this field is not set and are visible in the session.created event at the start of the session.
    - MaxOutputTokens RealtimeSessionCreateRequestMaxOutputTokensUnion
      
      Maximum number of output tokens for a single assistant response, inclusive of tool calls. Provide an integer between 1 and 4096 to limit output tokens, or inf for the maximum available tokens for a given model. Defaults to inf.
      - int64
      - Inf
        
        const InfInf Inf = "inf"
    - Model RealtimeSessionCreateRequestModel
      
      The Realtime model used for this session.
      - string
      - RealtimeSessionCreateRequestModel
        
        const RealtimeSessionCreateRequestModelGPTRealtime RealtimeSessionCreateRequestModel = "gpt-realtime"
        
        const RealtimeSessionCreateRequestModelGPTRealtime1_5 RealtimeSessionCreateRequestModel = "gpt-realtime-1.5"
        
        const RealtimeSessionCreateRequestModelGPTRealtime2 RealtimeSessionCreateRequestModel = "gpt-realtime-2"
        
        const RealtimeSessionCreateRequestModelGPTRealtime2025_08_28 RealtimeSessionCreateRequestModel = "gpt-realtime-2025-08-28"
        
        const RealtimeSessionCreateRequestModelGPT4oRealtimePreview RealtimeSessionCreateRequestModel = "gpt-4o-realtime-preview"
        
        const RealtimeSessionCreateRequestModelGPT4oRealtimePreview2024_10_01 RealtimeSessionCreateRequestModel = "gpt-4o-realtime-preview-2024-10-01"
        
        const RealtimeSessionCreateRequestModelGPT4oRealtimePreview2024_12_17 RealtimeSessionCreateRequestModel = "gpt-4o-realtime-preview-2024-12-17"
        
        const RealtimeSessionCreateRequestModelGPT4oRealtimePreview2025_06_03 RealtimeSessionCreateRequestModel = "gpt-4o-realtime-preview-2025-06-03"
        
        const RealtimeSessionCreateRequestModelGPT4oMiniRealtimePreview RealtimeSessionCreateRequestModel = "gpt-4o-mini-realtime-preview"
        
        const RealtimeSessionCreateRequestModelGPT4oMiniRealtimePreview2024_12_17 RealtimeSessionCreateRequestModel = "gpt-4o-mini-realtime-preview-2024-12-17"
        
        const RealtimeSessionCreateRequestModelGPTRealtimeMini RealtimeSessionCreateRequestModel = "gpt-realtime-mini"
        
        const RealtimeSessionCreateRequestModelGPTRealtimeMini2025_10_06 RealtimeSessionCreateRequestModel = "gpt-realtime-mini-2025-10-06"
        
        const RealtimeSessionCreateRequestModelGPTRealtimeMini2025_12_15 RealtimeSessionCreateRequestModel = "gpt-realtime-mini-2025-12-15"
        
        const RealtimeSessionCreateRequestModelGPTAudio1_5 RealtimeSessionCreateRequestModel = "gpt-audio-1.5"
        
        const RealtimeSessionCreateRequestModelGPTAudioMini RealtimeSessionCreateRequestModel = "gpt-audio-mini"
        
        const RealtimeSessionCreateRequestModelGPTAudioMini2025_10_06 RealtimeSessionCreateRequestModel = "gpt-audio-mini-2025-10-06"
        
        const RealtimeSessionCreateRequestModelGPTAudioMini2025_12_15 RealtimeSessionCreateRequestModel = "gpt-audio-mini-2025-12-15"
    - OutputModalities []string
      
      The set of modalities the model can respond with. It defaults to ["audio"], indicating that the model will respond with audio plus a transcript. ["text"] can be used to make the model respond with text only. It is not possible to request both text and audio at the same time.
      - const RealtimeSessionCreateRequestOutputModalityText RealtimeSessionCreateRequestOutputModality = "text"
      - const RealtimeSessionCreateRequestOutputModalityAudio RealtimeSessionCreateRequestOutputModality = "audio"
    - ParallelToolCalls bool
      
      Whether the model may call multiple tools in parallel. Only supported by reasoning Realtime models such as gpt-realtime-2.
    - Prompt ResponsePrompt
      
      Reference to a prompt template and its variables. Learn more.
      - ID string
        
        The unique identifier of the prompt template to use.
      - Variables map[string, ResponsePromptVariableUnion]
        
        Optional map of values to substitute in for variables in your prompt. The substitution values can either be strings, or other Response input types like images or files.
        
        string
        
        type ResponseInputText struct{…}
        
        A text input to the model.
        
        Text string
        
        The text input to the model.
        
        Type InputText
        
        The type of the input item. Always input_text.
        
        const InputTextInputText InputText = "input_text"
        
        type ResponseInputImage struct{…}
        
        An image input to the model. Learn about image inputs.
        
        Detail ResponseInputImageDetail
        
        The detail level of the image to be sent to the model. One of high, low, auto, or original. Defaults to auto.
        
        const ResponseInputImageDetailLow ResponseInputImageDetail = "low"
        
        const ResponseInputImageDetailHigh ResponseInputImageDetail = "high"
        
        const ResponseInputImageDetailAuto ResponseInputImageDetail = "auto"
        
        const ResponseInputImageDetailOriginal ResponseInputImageDetail = "original"
        
        Type InputImage
        
        The type of the input item. Always input_image.
        
        const InputImageInputImage InputImage = "input_image"
        
        FileID string
        
        The ID of the file to be sent to the model.
        
        ImageURL string
        
        The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL.
        
        type ResponseInputFile struct{…}
        
        A file input to the model.
        
        Type InputFile
        
        The type of the input item. Always input_file.
        
        const InputFileInputFile InputFile = "input_file"
        
        Detail ResponseInputFileDetail
        
        The detail level of the file to be sent to the model. Use low for the default rendering behavior, or high to render the file at higher quality. Defaults to low.
        
        const ResponseInputFileDetailLow ResponseInputFileDetail = "low"
        
        const ResponseInputFileDetailHigh ResponseInputFileDetail = "high"
        
        FileData string
        
        The content of the file to be sent to the model.
        
        FileID string
        
        The ID of the file to be sent to the model.
        
        FileURL string
        
        The URL of the file to be sent to the model.
        
        Filename string
        
        The name of the file to be sent to the model.
      - Version string
        
        Optional version of the prompt template.
    - Reasoning RealtimeReasoning
      
      Configuration for reasoning-capable Realtime models such as gpt-realtime-2.
      - Effort RealtimeReasoningEffort
        
        Constrains effort on reasoning for reasoning-capable Realtime models such as gpt-realtime-2.
        
        const RealtimeReasoningEffortMinimal RealtimeReasoningEffort = "minimal"
        
        const RealtimeReasoningEffortLow RealtimeReasoningEffort = "low"
        
        const RealtimeReasoningEffortMedium RealtimeReasoningEffort = "medium"
        
        const RealtimeReasoningEffortHigh RealtimeReasoningEffort = "high"
        
        const RealtimeReasoningEffortXhigh RealtimeReasoningEffort = "xhigh"
    - ToolChoice RealtimeToolChoiceConfigUnion
      
      How the model chooses tools. Provide one of the string modes or force a specific function/MCP tool.
      - type ToolChoiceOptions string
        
        Controls which (if any) tool is called by the model.
        
        none means the model will not call any tool and instead generates a message.
        
        auto means the model can pick between generating a message or calling one or more tools.
        
        required means the model must call one or more tools.
        
        const ToolChoiceOptionsNone ToolChoiceOptions = "none"
        
        const ToolChoiceOptionsAuto ToolChoiceOptions = "auto"
        
        const ToolChoiceOptionsRequired ToolChoiceOptions = "required"
      - type ToolChoiceFunction struct{…}
        
        Use this option to force the model to call a specific function.
        
        Name string
        
        The name of the function to call.
        
        Type Function
        
        For function calling, the type is always function.
        
        const FunctionFunction Function = "function"
      - type ToolChoiceMcp struct{…}
        
        Use this option to force the model to call a specific tool on a remote MCP server.
        
        ServerLabel string
        
        The label of the MCP server to use.
        
        Type Mcp
        
        For MCP tools, the type is always mcp.
        
        const McpMcp Mcp = "mcp"
        
        Name string
        
        The name of the tool to call on the server.
    - Tools RealtimeToolsConfig
      
      Tools available to the model.
      - type RealtimeFunctionTool struct{…}
        
        Description string
        
        The description of the function, including guidance on when and how to call it, and guidance about what to tell the user when calling (if anything).
        
        Name string
        
        The name of the function.
        
        Parameters any
        
        Parameters of the function in JSON Schema.
        
        Type RealtimeFunctionToolType
        
        The type of the tool, i.e. function.
        
        const RealtimeFunctionToolTypeFunction RealtimeFunctionToolType = "function"
      - RealtimeToolsConfigUnionMcp
        
        ServerLabel string
        
        A label for this MCP server, used to identify it in tool calls.
        
        Type Mcp
        
        The type of the MCP tool. Always mcp.
        
        const McpMcp Mcp = "mcp"
        
        AllowedTools RealtimeToolsConfigUnionMcpAllowedTools
        
        List of allowed tool names or a filter object.
        
        []string
        
        RealtimeToolsConfigUnionMcpAllowedToolsMcpToolFilter
        
        ReadOnly bool
        
        Indicates whether or not a tool modifies data or is read-only. If an MCP server is annotated with readOnlyHint, it will match this filter.
        
        ToolNames []string
        
        List of allowed tool names.
        
        Authorization string
        
        An OAuth access token that can be used with a remote MCP server, either with a custom MCP server URL or a service connector. Your application must handle the OAuth authorization flow and provide the token here.
        
        ConnectorID string
        
        Identifier for service connectors, like those available in ChatGPT. One of server_url or connector_id must be provided. Learn more about service connectors here.
        
        Currently supported connector_id values are:
        
        Dropbox: connector_dropbox
        
        Gmail: connector_gmail
        
        Google Calendar: connector_googlecalendar
        
        Google Drive: connector_googledrive
        
        Microsoft Teams: connector_microsoftteams
        
        Outlook Calendar: connector_outlookcalendar
        
        Outlook Email: connector_outlookemail
        
        SharePoint: connector_sharepoint
        
        const RealtimeToolsConfigUnionMcpConnectorIDConnectorDropbox RealtimeToolsConfigUnionMcpConnectorID = "connector_dropbox"
        
        const RealtimeToolsConfigUnionMcpConnectorIDConnectorGmail RealtimeToolsConfigUnionMcpConnectorID = "connector_gmail"
        
        const RealtimeToolsConfigUnionMcpConnectorIDConnectorGooglecalendar RealtimeToolsConfigUnionMcpConnectorID = "connector_googlecalendar"
        
        const RealtimeToolsConfigUnionMcpConnectorIDConnectorGoogledrive RealtimeToolsConfigUnionMcpConnectorID = "connector_googledrive"
        
        const RealtimeToolsConfigUnionMcpConnectorIDConnectorMicrosoftteams RealtimeToolsConfigUnionMcpConnectorID = "connector_microsoftteams"
        
        const RealtimeToolsConfigUnionMcpConnectorIDConnectorOutlookcalendar RealtimeToolsConfigUnionMcpConnectorID = "connector_outlookcalendar"
        
        const RealtimeToolsConfigUnionMcpConnectorIDConnectorOutlookemail RealtimeToolsConfigUnionMcpConnectorID = "connector_outlookemail"
        
        const RealtimeToolsConfigUnionMcpConnectorIDConnectorSharepoint RealtimeToolsConfigUnionMcpConnectorID = "connector_sharepoint"
        
        DeferLoading bool
        
        Whether this MCP tool is deferred and discovered via tool search.
        
        Headers map[string, string]
        
        Optional HTTP headers to send to the MCP server. Use for authentication or other purposes.
        
        RequireApproval RealtimeToolsConfigUnionMcpRequireApproval
        
        Specify which of the MCP server's tools require approval.
        
        RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalFilter
        
        Always RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalFilterAlways
        
        A filter object to specify which tools are allowed.
        
        ReadOnly bool
        
        Indicates whether or not a tool modifies data or is read-only. If an MCP server is annotated with readOnlyHint, it will match this filter.
        
        ToolNames []string
        
        List of allowed tool names.
        
        Never RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalFilterNever
        
        A filter object to specify which tools are allowed.
        
        ReadOnly bool
        
        Indicates whether or not a tool modifies data or is read-only. If an MCP server is annotated with readOnlyHint, it will match this filter.
        
        ToolNames []string
        
        List of allowed tool names.
        
        string
        
        const RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalSettingAlways RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalSetting = "always"
        
        const RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalSettingNever RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalSetting = "never"
        
        ServerDescription string
        
        Optional description of the MCP server, used to provide more context.
        
        ServerURL string
        
        The URL for the MCP server. One of server_url or connector_id must be provided.
    - Tracing RealtimeTracingConfigUnion
      
      Realtime API can write session traces to the Traces Dashboard. Set to null to disable tracing. Once tracing is enabled for a session, the configuration cannot be modified.
      
      auto will create a trace for the session with default values for the workflow name, group id, and metadata.
      - Auto
        
        const AutoAuto Auto = "auto"
      - RealtimeTracingConfigTracingConfiguration
        
        GroupID string
        
        The group id to attach to this trace to enable filtering and grouping in the Traces Dashboard.
        
        Metadata any
        
        The arbitrary metadata to attach to this trace to enable filtering in the Traces Dashboard.
        
        WorkflowName string
        
        The name of the workflow to attach to this trace. This is used to name the trace in the Traces Dashboard.
    - Truncation RealtimeTruncationUnion
      
      When the number of tokens in a conversation exceeds the model's input token limit, the conversation be truncated, meaning messages (starting from the oldest) will not be included in the model's context. A 32k context model with 4,096 max output tokens can only include 28,224 tokens in the context before truncation occurs.
      
      Clients can configure truncation behavior to truncate with a lower max token limit, which is an effective way to control token usage and cost.
      
      Truncation will reduce the number of cached tokens on the next turn (busting the cache), since messages are dropped from the beginning of the context. However, clients can also configure truncation to retain messages up to a fraction of the maximum context size, which will reduce the need for future truncations and thus improve the cache rate.
      
      Truncation can be disabled entirely, which means the server will never truncate but would instead return an error if the conversation exceeds the model's input token limit.
      - type RealtimeTruncationRealtimeTruncationStrategy string
        
        The truncation strategy to use for the session. auto is the default truncation strategy. disabled will disable truncation and emit errors when the conversation exceeds the input token limit.
        
        const RealtimeTruncationRealtimeTruncationStrategyAuto RealtimeTruncationRealtimeTruncationStrategy = "auto"
        
        const RealtimeTruncationRealtimeTruncationStrategyDisabled RealtimeTruncationRealtimeTruncationStrategy = "disabled"
      - type RealtimeTruncationRetentionRatio struct{…}
        
        Retain a fraction of the conversation tokens when the conversation exceeds the input token limit. This allows you to amortize truncations across multiple turns, which can help improve cached token usage.
        
        RetentionRatio float64
        
        Fraction of post-instruction conversation tokens to retain (0.0 - 1.0) when the conversation exceeds the input token limit. Setting this to 0.8 means that messages will be dropped until 80% of the maximum allowed tokens are used. This helps reduce the frequency of truncations and improve cache rates.
        
        Type RetentionRatio
        
        Use retention ratio truncation.
        
        const RetentionRatioRetentionRatio RetentionRatio = "retention_ratio"
        
        TokenLimits RealtimeTruncationRetentionRatioTokenLimits
        
        Optional custom token limits for this truncation strategy. If not provided, the model's default token limits will be used.
        
        PostInstructions int64
        
        Maximum tokens allowed in the conversation after instructions (which including tool definitions). For example, setting this to 5,000 would mean that truncation would occur when the conversation exceeds 5,000 tokens after instructions. This cannot be higher than the model's context window size minus the maximum output tokens.
  - type RealtimeTranscriptionSessionCreateRequest struct{…}
    
    Realtime transcription session object configuration.
    - Type Transcription
      
      The type of session to create. Always transcription for transcription sessions.
      - const TranscriptionTranscription Transcription = "transcription"
    - Audio RealtimeTranscriptionSessionAudio
      
      Configuration for input and output audio.
      - Input RealtimeTranscriptionSessionAudioInput
        
        Format RealtimeAudioFormatsUnion
        
        The PCM audio format. Only a 24kHz sample rate is supported.
        
        NoiseReduction RealtimeTranscriptionSessionAudioInputNoiseReduction
        
        Configuration for input audio noise reduction. This can be set to null to turn off. Noise reduction filters audio added to the input audio buffer before it is sent to VAD and the model. Filtering the audio can improve VAD and turn detection accuracy (reducing false positives) and model performance by improving perception of the input audio.
        
        Type NoiseReductionType
        
        Type of noise reduction. near_field is for close-talking microphones such as headphones, far_field is for far-field microphones such as laptop or conference room microphones.
        
        Transcription AudioTranscription
        
        Configuration for input audio transcription, defaults to off and can be set to null to turn off once on. Input audio transcription is not native to the model, since the model consumes audio directly. Transcription runs asynchronously through the /audio/transcriptions endpoint and should be treated as guidance of input audio content rather than precisely what the model heard. The client can optionally set the language and prompt for transcription, these offer additional guidance to the transcription service.
        
        TurnDetection RealtimeTranscriptionSessionAudioInputTurnDetectionUnion
        
        Configuration for turn detection, ether Server VAD or Semantic VAD. This can be set to null to turn off, in which case the client must manually trigger model response.
        
        Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech.
        
        Semantic VAD is more advanced and uses a turn detection model (in conjunction with VAD) to semantically estimate whether the user has finished speaking, then dynamically sets a timeout based on this probability. For example, if user audio trails off with "uhhm", the model will score a low probability of turn end and wait longer for the user to continue speaking. This can be useful for more natural conversations, but may have a higher latency.
        
        For gpt-realtime-whisper transcription sessions, turn detection must be set to null; VAD is not supported.
        
        RealtimeTranscriptionSessionAudioInputTurnDetectionServerVad
        
        Type ServerVad
        
        Type of turn detection, server_vad to turn on simple Server VAD.
        
        const ServerVadServerVad ServerVad = "server_vad"
        
        CreateResponse bool
        
        Whether or not to automatically generate a response when a VAD stop event occurs. If interrupt_response is set to false this may fail to create a response if the model is already responding.
        
        If both create_response and interrupt_response are set to false, the model will never respond automatically but VAD events will still be emitted.
        
        IdleTimeoutMs int64
        
        Optional timeout after which a model response will be triggered automatically. This is useful for situations in which a long pause from the user is unexpected, such as a phone call. The model will effectively prompt the user to continue the conversation based on the current context.
        
        The timeout value will be applied after the last model response's audio has finished playing, i.e. it's set to the response.done time plus audio playback duration.
        
        An input_audio_buffer.timeout_triggered event (plus events associated with the Response) will be emitted when the timeout is reached. Idle timeout is currently only supported for server_vad mode.
        
        InterruptResponse bool
        
        Whether or not to automatically interrupt (cancel) any ongoing response with output to the default conversation (i.e. conversation of auto) when a VAD start event occurs. If true then the response will be cancelled, otherwise it will continue until complete.
        
        If both create_response and interrupt_response are set to false, the model will never respond automatically but VAD events will still be emitted.
        
        PrefixPaddingMs int64
        
        Used only for server_vad mode. Amount of audio to include before the VAD detected speech (in milliseconds). Defaults to 300ms.
        
        SilenceDurationMs int64
        
        Used only for server_vad mode. Duration of silence to detect speech stop (in milliseconds). Defaults to 500ms. With shorter values the model will respond more quickly, but may jump in on short pauses from the user.
        
        Threshold float64
        
        Used only for server_vad mode. Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A higher threshold will require louder audio to activate the model, and thus might perform better in noisy environments.
        
        RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVad
        
        Type SemanticVad
        
        Type of turn detection, semantic_vad to turn on Semantic VAD.
        
        const SemanticVadSemanticVad SemanticVad = "semantic_vad"
        
        CreateResponse bool
        
        Whether or not to automatically generate a response when a VAD stop event occurs.
        
        Eagerness string
        
        Used only for semantic_vad mode. The eagerness of the model to respond. low will wait longer for the user to continue speaking, high will respond more quickly. auto is the default and is equivalent to medium. low, medium, and high have max timeouts of 8s, 4s, and 2s respectively.
        
        const RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagernessLow RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagerness = "low"
        
        const RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagernessMedium RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagerness = "medium"
        
        const RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagernessHigh RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagerness = "high"
        
        const RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagernessAuto RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagerness = "auto"
        
        InterruptResponse bool
        
        Whether or not to automatically interrupt any ongoing response with output to the default conversation (i.e. conversation of auto) when a VAD start event occurs.
    - Include []string
      
      Additional fields to include in server outputs.
      
      item.input_audio_transcription.logprobs: Include logprobs for input audio transcription.
      - const RealtimeTranscriptionSessionCreateRequestIncludeItemInputAudioTranscriptionLogprobs RealtimeTranscriptionSessionCreateRequestInclude = "item.input_audio_transcription.logprobs"

Returns

type ClientSecretNewResponse struct{…}

Response from creating a session and client secret for the Realtime API.
- ExpiresAt int64
  
  Expiration timestamp for the client secret, in seconds since epoch.
- Session ClientSecretNewResponseSessionUnion
  
  The session configuration for either a realtime or transcription session.
  - type RealtimeSessionCreateResponse struct{…}
    
    A Realtime session configuration object.
    - ID string
      
      Unique identifier for the session that looks like sess_1234567890abcdef.
    - Object RealtimeSession
      
      The object type. Always realtime.session.
      - const RealtimeSessionRealtimeSession RealtimeSession = "realtime.session"
    - Type Realtime
      
      The type of session to create. Always realtime for the Realtime API.
      - const RealtimeRealtime Realtime = "realtime"
    - Audio RealtimeSessionCreateResponseAudio
      
      Configuration for input and output audio.
      - Input RealtimeSessionCreateResponseAudioInput
        
        Format RealtimeAudioFormatsUnion
        
        The format of the input audio.
        
        type RealtimeAudioFormatsAudioPCM struct{…}
        
        The PCM audio format. Only a 24kHz sample rate is supported.
        
        Rate int64
        
        The sample rate of the audio. Always 24000.
        
        const RealtimeAudioFormatsAudioPCMRate24000 RealtimeAudioFormatsAudioPCMRate = 24000
        
        Type string
        
        The audio format. Always audio/pcm.
        
        const RealtimeAudioFormatsAudioPCMTypeAudioPCM RealtimeAudioFormatsAudioPCMType = "audio/pcm"
        
        type RealtimeAudioFormatsAudioPCMU struct{…}
        
        The G.711 μ-law format.
        
        Type string
        
        The audio format. Always audio/pcmu.
        
        const RealtimeAudioFormatsAudioPCMUTypeAudioPCMU RealtimeAudioFormatsAudioPCMUType = "audio/pcmu"
        
        type RealtimeAudioFormatsAudioPCMA struct{…}
        
        The G.711 A-law format.
        
        Type string
        
        The audio format. Always audio/pcma.
        
        const RealtimeAudioFormatsAudioPCMATypeAudioPCMA RealtimeAudioFormatsAudioPCMAType = "audio/pcma"
        
        NoiseReduction RealtimeSessionCreateResponseAudioInputNoiseReduction
        
        Configuration for input audio noise reduction. This can be set to null to turn off. Noise reduction filters audio added to the input audio buffer before it is sent to VAD and the model. Filtering the audio can improve VAD and turn detection accuracy (reducing false positives) and model performance by improving perception of the input audio.
        
        Type NoiseReductionType
        
        Type of noise reduction. near_field is for close-talking microphones such as headphones, far_field is for far-field microphones such as laptop or conference room microphones.
        
        const NoiseReductionTypeNearField NoiseReductionType = "near_field"
        
        const NoiseReductionTypeFarField NoiseReductionType = "far_field"
        
        Transcription AudioTranscription
        
        Delay AudioTranscriptionDelay
        
        Controls how long the model waits before emitting transcription text. Higher values can improve transcription accuracy at the cost of latency. Only supported with gpt-realtime-whisper in GA Realtime sessions.
        
        const AudioTranscriptionDelayMinimal AudioTranscriptionDelay = "minimal"
        
        const AudioTranscriptionDelayLow AudioTranscriptionDelay = "low"
        
        const AudioTranscriptionDelayMedium AudioTranscriptionDelay = "medium"
        
        const AudioTranscriptionDelayHigh AudioTranscriptionDelay = "high"
        
        const AudioTranscriptionDelayXhigh AudioTranscriptionDelay = "xhigh"
        
        Language string
        
        The language of the input audio. Supplying the input language in ISO-639-1 (e.g. en) format will improve accuracy and latency.
        
        Model AudioTranscriptionModel
        
        The model to use for transcription. Current options are whisper-1, gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, and gpt-realtime-whisper. Use gpt-4o-transcribe-diarize when you need diarization with speaker labels.
        
        string
        
        type AudioTranscriptionModel string
        
        The model to use for transcription. Current options are whisper-1, gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, and gpt-realtime-whisper. Use gpt-4o-transcribe-diarize when you need diarization with speaker labels.
        
        const AudioTranscriptionModelWhisper1 AudioTranscriptionModel = "whisper-1"
        
        const AudioTranscriptionModelGPT4oMiniTranscribe AudioTranscriptionModel = "gpt-4o-mini-transcribe"
        
        const AudioTranscriptionModelGPT4oMiniTranscribe2025_12_15 AudioTranscriptionModel = "gpt-4o-mini-transcribe-2025-12-15"
        
        const AudioTranscriptionModelGPT4oTranscribe AudioTranscriptionModel = "gpt-4o-transcribe"
        
        const AudioTranscriptionModelGPT4oTranscribeDiarize AudioTranscriptionModel = "gpt-4o-transcribe-diarize"
        
        const AudioTranscriptionModelGPTRealtimeWhisper AudioTranscriptionModel = "gpt-realtime-whisper"
        
        Prompt string
        
        An optional text to guide the model's style or continue a previous audio segment. For whisper-1, the prompt is a list of keywords. For gpt-4o-transcribe models (excluding gpt-4o-transcribe-diarize), the prompt is a free text string, for example "expect words related to technology". Prompt is not supported with gpt-realtime-whisper in GA Realtime sessions.
        
        TurnDetection RealtimeSessionCreateResponseAudioInputTurnDetectionUnion
        
        Configuration for turn detection, ether Server VAD or Semantic VAD. This can be set to null to turn off, in which case the client must manually trigger model response.
        
        Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech.
        
        Semantic VAD is more advanced and uses a turn detection model (in conjunction with VAD) to semantically estimate whether the user has finished speaking, then dynamically sets a timeout based on this probability. For example, if user audio trails off with "uhhm", the model will score a low probability of turn end and wait longer for the user to continue speaking. This can be useful for more natural conversations, but may have a higher latency.
        
        For gpt-realtime-whisper transcription sessions, turn detection must be set to null; VAD is not supported.
        
        type RealtimeSessionCreateResponseAudioInputTurnDetectionServerVad struct{…}
        
        Server-side voice activity detection (VAD) which flips on when user speech is detected and off after a period of silence.
        
        Type ServerVad
        
        Type of turn detection, server_vad to turn on simple Server VAD.
        
        const ServerVadServerVad ServerVad = "server_vad"
        
        CreateResponse bool
        
        Whether or not to automatically generate a response when a VAD stop event occurs. If interrupt_response is set to false this may fail to create a response if the model is already responding.
        
        If both create_response and interrupt_response are set to false, the model will never respond automatically but VAD events will still be emitted.
        
        IdleTimeoutMs int64
        
        Optional timeout after which a model response will be triggered automatically. This is useful for situations in which a long pause from the user is unexpected, such as a phone call. The model will effectively prompt the user to continue the conversation based on the current context.
        
        The timeout value will be applied after the last model response's audio has finished playing, i.e. it's set to the response.done time plus audio playback duration.
        
        An input_audio_buffer.timeout_triggered event (plus events associated with the Response) will be emitted when the timeout is reached. Idle timeout is currently only supported for server_vad mode.
        
        InterruptResponse bool
        
        Whether or not to automatically interrupt (cancel) any ongoing response with output to the default conversation (i.e. conversation of auto) when a VAD start event occurs. If true then the response will be cancelled, otherwise it will continue until complete.
        
        If both create_response and interrupt_response are set to false, the model will never respond automatically but VAD events will still be emitted.
        
        PrefixPaddingMs int64
        
        Used only for server_vad mode. Amount of audio to include before the VAD detected speech (in milliseconds). Defaults to 300ms.
        
        SilenceDurationMs int64
        
        Used only for server_vad mode. Duration of silence to detect speech stop (in milliseconds). Defaults to 500ms. With shorter values the model will respond more quickly, but may jump in on short pauses from the user.
        
        Threshold float64
        
        Used only for server_vad mode. Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A higher threshold will require louder audio to activate the model, and thus might perform better in noisy environments.
        
        type RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVad struct{…}
        
        Server-side semantic turn detection which uses a model to determine when the user has finished speaking.
        
        Type SemanticVad
        
        Type of turn detection, semantic_vad to turn on Semantic VAD.
        
        const SemanticVadSemanticVad SemanticVad = "semantic_vad"
        
        CreateResponse bool
        
        Whether or not to automatically generate a response when a VAD stop event occurs.
        
        Eagerness string
        
        Used only for semantic_vad mode. The eagerness of the model to respond. low will wait longer for the user to continue speaking, high will respond more quickly. auto is the default and is equivalent to medium. low, medium, and high have max timeouts of 8s, 4s, and 2s respectively.
        
        const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessLow RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "low"
        
        const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessMedium RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "medium"
        
        const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessHigh RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "high"
        
        const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessAuto RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "auto"
        
        InterruptResponse bool
        
        Whether or not to automatically interrupt any ongoing response with output to the default conversation (i.e. conversation of auto) when a VAD start event occurs.
      - Output RealtimeSessionCreateResponseAudioOutput
        
        Format RealtimeAudioFormatsUnion
        
        The format of the output audio.
        
        Speed float64
        
        The speed of the model's spoken response as a multiple of the original speed. 1.0 is the default speed. 0.25 is the minimum speed. 1.5 is the maximum speed. This value can only be changed in between model turns, not while a response is in progress.
        
        This parameter is a post-processing adjustment to the audio after it is generated, it's also possible to prompt the model to speak faster or slower.
        
        Voice string
        
        The voice the model uses to respond. Voice cannot be changed during the session once the model has responded with audio at least once. Current voice options are alloy, ash, ballad, coral, echo, sage, shimmer, verse, marin, and cedar. We recommend marin and cedar for best quality.
        
        string
        
        string
        
        const RealtimeSessionCreateResponseAudioOutputVoiceAlloy RealtimeSessionCreateResponseAudioOutputVoice = "alloy"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceAsh RealtimeSessionCreateResponseAudioOutputVoice = "ash"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceBallad RealtimeSessionCreateResponseAudioOutputVoice = "ballad"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceCoral RealtimeSessionCreateResponseAudioOutputVoice = "coral"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceEcho RealtimeSessionCreateResponseAudioOutputVoice = "echo"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceSage RealtimeSessionCreateResponseAudioOutputVoice = "sage"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceShimmer RealtimeSessionCreateResponseAudioOutputVoice = "shimmer"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceVerse RealtimeSessionCreateResponseAudioOutputVoice = "verse"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceMarin RealtimeSessionCreateResponseAudioOutputVoice = "marin"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceCedar RealtimeSessionCreateResponseAudioOutputVoice = "cedar"
    - ExpiresAt int64
      
      Expiration timestamp for the session, in seconds since epoch.
    - Include []string
      
      Additional fields to include in server outputs.
      
      item.input_audio_transcription.logprobs: Include logprobs for input audio transcription.
      - const RealtimeSessionCreateResponseIncludeItemInputAudioTranscriptionLogprobs RealtimeSessionCreateResponseInclude = "item.input_audio_transcription.logprobs"
    - Instructions string
      
      The default system instructions (i.e. system message) prepended to model calls. This field allows the client to guide the model on desired responses. The model can be instructed on response content and format, (e.g. "be extremely succinct", "act friendly", "here are examples of good responses") and on audio behavior (e.g. "talk quickly", "inject emotion into your voice", "laugh frequently"). The instructions are not guaranteed to be followed by the model, but they provide guidance to the model on the desired behavior.
      
      Note that the server sets default instructions which will be used if this field is not set and are visible in the session.created event at the start of the session.
    - MaxOutputTokens RealtimeSessionCreateResponseMaxOutputTokensUnion
      
      Maximum number of output tokens for a single assistant response, inclusive of tool calls. Provide an integer between 1 and 4096 to limit output tokens, or inf for the maximum available tokens for a given model. Defaults to inf.
      - int64
      - type Inf string
        
        const InfInf Inf = "inf"
    - Model RealtimeSessionCreateResponseModel
      
      The Realtime model used for this session.
      - string
      - type RealtimeSessionCreateResponseModel string
        
        The Realtime model used for this session.
        
        const RealtimeSessionCreateResponseModelGPTRealtime RealtimeSessionCreateResponseModel = "gpt-realtime"
        
        const RealtimeSessionCreateResponseModelGPTRealtime1_5 RealtimeSessionCreateResponseModel = "gpt-realtime-1.5"
        
        const RealtimeSessionCreateResponseModelGPTRealtime2 RealtimeSessionCreateResponseModel = "gpt-realtime-2"
        
        const RealtimeSessionCreateResponseModelGPTRealtime2025_08_28 RealtimeSessionCreateResponseModel = "gpt-realtime-2025-08-28"
        
        const RealtimeSessionCreateResponseModelGPT4oRealtimePreview RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview"
        
        const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2024_10_01 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2024-10-01"
        
        const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2024_12_17 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2024-12-17"
        
        const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2025_06_03 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2025-06-03"
        
        const RealtimeSessionCreateResponseModelGPT4oMiniRealtimePreview RealtimeSessionCreateResponseModel = "gpt-4o-mini-realtime-preview"
        
        const RealtimeSessionCreateResponseModelGPT4oMiniRealtimePreview2024_12_17 RealtimeSessionCreateResponseModel = "gpt-4o-mini-realtime-preview-2024-12-17"
        
        const RealtimeSessionCreateResponseModelGPTRealtimeMini RealtimeSessionCreateResponseModel = "gpt-realtime-mini"
        
        const RealtimeSessionCreateResponseModelGPTRealtimeMini2025_10_06 RealtimeSessionCreateResponseModel = "gpt-realtime-mini-2025-10-06"
        
        const RealtimeSessionCreateResponseModelGPTRealtimeMini2025_12_15 RealtimeSessionCreateResponseModel = "gpt-realtime-mini-2025-12-15"
        
        const RealtimeSessionCreateResponseModelGPTAudio1_5 RealtimeSessionCreateResponseModel = "gpt-audio-1.5"
        
        const RealtimeSessionCreateResponseModelGPTAudioMini RealtimeSessionCreateResponseModel = "gpt-audio-mini"
        
        const RealtimeSessionCreateResponseModelGPTAudioMini2025_10_06 RealtimeSessionCreateResponseModel = "gpt-audio-mini-2025-10-06"
        
        const RealtimeSessionCreateResponseModelGPTAudioMini2025_12_15 RealtimeSessionCreateResponseModel = "gpt-audio-mini-2025-12-15"
    - OutputModalities []string
      
      The set of modalities the model can respond with. It defaults to ["audio"], indicating that the model will respond with audio plus a transcript. ["text"] can be used to make the model respond with text only. It is not possible to request both text and audio at the same time.
      - const RealtimeSessionCreateResponseOutputModalityText RealtimeSessionCreateResponseOutputModality = "text"
      - const RealtimeSessionCreateResponseOutputModalityAudio RealtimeSessionCreateResponseOutputModality = "audio"
    - Prompt ResponsePrompt
      
      Reference to a prompt template and its variables. Learn more.
      - ID string
        
        The unique identifier of the prompt template to use.
      - Variables map[string, ResponsePromptVariableUnion]
        
        Optional map of values to substitute in for variables in your prompt. The substitution values can either be strings, or other Response input types like images or files.
        
        string
        
        type ResponseInputText struct{…}
        
        A text input to the model.
        
        Text string
        
        The text input to the model.
        
        Type InputText
        
        The type of the input item. Always input_text.
        
        const InputTextInputText InputText = "input_text"
        
        type ResponseInputImage struct{…}
        
        An image input to the model. Learn about image inputs.
        
        Detail ResponseInputImageDetail
        
        The detail level of the image to be sent to the model. One of high, low, auto, or original. Defaults to auto.
        
        const ResponseInputImageDetailLow ResponseInputImageDetail = "low"
        
        const ResponseInputImageDetailHigh ResponseInputImageDetail = "high"
        
        const ResponseInputImageDetailAuto ResponseInputImageDetail = "auto"
        
        const ResponseInputImageDetailOriginal ResponseInputImageDetail = "original"
        
        Type InputImage
        
        The type of the input item. Always input_image.
        
        const InputImageInputImage InputImage = "input_image"
        
        FileID string
        
        The ID of the file to be sent to the model.
        
        ImageURL string
        
        The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL.
        
        type ResponseInputFile struct{…}
        
        A file input to the model.
        
        Type InputFile
        
        The type of the input item. Always input_file.
        
        const InputFileInputFile InputFile = "input_file"
        
        Detail ResponseInputFileDetail
        
        The detail level of the file to be sent to the model. Use low for the default rendering behavior, or high to render the file at higher quality. Defaults to low.
        
        const ResponseInputFileDetailLow ResponseInputFileDetail = "low"
        
        const ResponseInputFileDetailHigh ResponseInputFileDetail = "high"
        
        FileData string
        
        The content of the file to be sent to the model.
        
        FileID string
        
        The ID of the file to be sent to the model.
        
        FileURL string
        
        The URL of the file to be sent to the model.
        
        Filename string
        
        The name of the file to be sent to the model.
      - Version string
        
        Optional version of the prompt template.
    - Reasoning RealtimeReasoning
      
      Configuration for reasoning-capable Realtime models such as gpt-realtime-2.
      - Effort RealtimeReasoningEffort
        
        Constrains effort on reasoning for reasoning-capable Realtime models such as gpt-realtime-2.
        
        const RealtimeReasoningEffortMinimal RealtimeReasoningEffort = "minimal"
        
        const RealtimeReasoningEffortLow RealtimeReasoningEffort = "low"
        
        const RealtimeReasoningEffortMedium RealtimeReasoningEffort = "medium"
        
        const RealtimeReasoningEffortHigh RealtimeReasoningEffort = "high"
        
        const RealtimeReasoningEffortXhigh RealtimeReasoningEffort = "xhigh"
    - ToolChoice RealtimeSessionCreateResponseToolChoiceUnion
      
      How the model chooses tools. Provide one of the string modes or force a specific function/MCP tool.
      - type ToolChoiceOptions string
        
        Controls which (if any) tool is called by the model.
        
        none means the model will not call any tool and instead generates a message.
        
        auto means the model can pick between generating a message or calling one or more tools.
        
        required means the model must call one or more tools.
        
        const ToolChoiceOptionsNone ToolChoiceOptions = "none"
        
        const ToolChoiceOptionsAuto ToolChoiceOptions = "auto"
        
        const ToolChoiceOptionsRequired ToolChoiceOptions = "required"
      - type ToolChoiceFunction struct{…}
        
        Use this option to force the model to call a specific function.
        
        Name string
        
        The name of the function to call.
        
        Type Function
        
        For function calling, the type is always function.
        
        const FunctionFunction Function = "function"
      - type ToolChoiceMcp struct{…}
        
        Use this option to force the model to call a specific tool on a remote MCP server.
        
        ServerLabel string
        
        The label of the MCP server to use.
        
        Type Mcp
        
        For MCP tools, the type is always mcp.
        
        const McpMcp Mcp = "mcp"
        
        Name string
        
        The name of the tool to call on the server.
    - Tools []RealtimeSessionCreateResponseToolUnion
      
      Tools available to the model.
      - type RealtimeFunctionTool struct{…}
        
        Description string
        
        The description of the function, including guidance on when and how to call it, and guidance about what to tell the user when calling (if anything).
        
        Name string
        
        The name of the function.
        
        Parameters any
        
        Parameters of the function in JSON Schema.
        
        Type RealtimeFunctionToolType
        
        The type of the tool, i.e. function.
        
        const RealtimeFunctionToolTypeFunction RealtimeFunctionToolType = "function"
      - type RealtimeSessionCreateResponseToolMcpTool struct{…}
        
        Give the model access to additional tools via remote Model Context Protocol (MCP) servers. Learn more about MCP.
        
        ServerLabel string
        
        A label for this MCP server, used to identify it in tool calls.
        
        Type Mcp
        
        The type of the MCP tool. Always mcp.
        
        const McpMcp Mcp = "mcp"
        
        AllowedTools RealtimeSessionCreateResponseToolMcpToolAllowedToolsUnion
        
        List of allowed tool names or a filter object.
        
        type RealtimeSessionCreateResponseToolMcpToolAllowedToolsMcpAllowedTools []string
        
        A string array of allowed tool names
        
        type RealtimeSessionCreateResponseToolMcpToolAllowedToolsMcpToolFilter struct{…}
        
        A filter object to specify which tools are allowed.
        
        ReadOnly bool
        
        Indicates whether or not a tool modifies data or is read-only. If an MCP server is annotated with readOnlyHint, it will match this filter.
        
        ToolNames []string
        
        List of allowed tool names.
        
        Authorization string
        
        An OAuth access token that can be used with a remote MCP server, either with a custom MCP server URL or a service connector. Your application must handle the OAuth authorization flow and provide the token here.
        
        ConnectorID string
        
        Identifier for service connectors, like those available in ChatGPT. One of server_url or connector_id must be provided. Learn more about service connectors here.
        
        Currently supported connector_id values are:
        
        Dropbox: connector_dropbox
        
        Gmail: connector_gmail
        
        Google Calendar: connector_googlecalendar
        
        Google Drive: connector_googledrive
        
        Microsoft Teams: connector_microsoftteams
        
        Outlook Calendar: connector_outlookcalendar
        
        Outlook Email: connector_outlookemail
        
        SharePoint: connector_sharepoint
        
        const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorDropbox RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_dropbox"
        
        const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGmail RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_gmail"
        
        const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGooglecalendar RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_googlecalendar"
        
        const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGoogledrive RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_googledrive"
        
        const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorMicrosoftteams RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_microsoftteams"
        
        const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorOutlookcalendar RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_outlookcalendar"
        
        const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorOutlookemail RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_outlookemail"
        
        const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorSharepoint RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_sharepoint"
        
        DeferLoading bool
        
        Whether this MCP tool is deferred and discovered via tool search.
        
        Headers map[string, string]
        
        Optional HTTP headers to send to the MCP server. Use for authentication or other purposes.
        
        RequireApproval RealtimeSessionCreateResponseToolMcpToolRequireApprovalUnion
        
        Specify which of the MCP server's tools require approval.
        
        type RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilter struct{…}
        
        Specify which of the MCP server's tools require approval. Can be always, never, or a filter object associated with tools that require approval.
        
        Always RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilterAlways
        
        A filter object to specify which tools are allowed.
        
        ReadOnly bool
        
        Indicates whether or not a tool modifies data or is read-only. If an MCP server is annotated with readOnlyHint, it will match this filter.
        
        ToolNames []string
        
        List of allowed tool names.
        
        Never RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilterNever
        
        A filter object to specify which tools are allowed.
        
        ReadOnly bool
        
        Indicates whether or not a tool modifies data or is read-only. If an MCP server is annotated with readOnlyHint, it will match this filter.
        
        ToolNames []string
        
        List of allowed tool names.
        
        type RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting string
        
        Specify a single approval policy for all tools. One of always or never. When set to always, all tools will require approval. When set to never, all tools will not require approval.
        
        const RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSettingAlways RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting = "always"
        
        const RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSettingNever RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting = "never"
        
        ServerDescription string
        
        Optional description of the MCP server, used to provide more context.
        
        ServerURL string
        
        The URL for the MCP server. One of server_url or connector_id must be provided.
    - Tracing RealtimeSessionCreateResponseTracingUnion
      
      Realtime API can write session traces to the Traces Dashboard. Set to null to disable tracing. Once tracing is enabled for a session, the configuration cannot be modified.
      
      auto will create a trace for the session with default values for the workflow name, group id, and metadata.
      - type Auto string
        
        Enables tracing and sets default values for tracing configuration options. Always auto.
        
        const AutoAuto Auto = "auto"
      - type RealtimeSessionCreateResponseTracingTracingConfiguration struct{…}
        
        Granular configuration for tracing.
        
        GroupID string
        
        The group id to attach to this trace to enable filtering and grouping in the Traces Dashboard.
        
        Metadata any
        
        The arbitrary metadata to attach to this trace to enable filtering in the Traces Dashboard.
        
        WorkflowName string
        
        The name of the workflow to attach to this trace. This is used to name the trace in the Traces Dashboard.
    - Truncation RealtimeTruncationUnion
      
      When the number of tokens in a conversation exceeds the model's input token limit, the conversation be truncated, meaning messages (starting from the oldest) will not be included in the model's context. A 32k context model with 4,096 max output tokens can only include 28,224 tokens in the context before truncation occurs.
      
      Clients can configure truncation behavior to truncate with a lower max token limit, which is an effective way to control token usage and cost.
      
      Truncation will reduce the number of cached tokens on the next turn (busting the cache), since messages are dropped from the beginning of the context. However, clients can also configure truncation to retain messages up to a fraction of the maximum context size, which will reduce the need for future truncations and thus improve the cache rate.
      
      Truncation can be disabled entirely, which means the server will never truncate but would instead return an error if the conversation exceeds the model's input token limit.
      - type RealtimeTruncationRealtimeTruncationStrategy string
        
        The truncation strategy to use for the session. auto is the default truncation strategy. disabled will disable truncation and emit errors when the conversation exceeds the input token limit.
        
        const RealtimeTruncationRealtimeTruncationStrategyAuto RealtimeTruncationRealtimeTruncationStrategy = "auto"
        
        const RealtimeTruncationRealtimeTruncationStrategyDisabled RealtimeTruncationRealtimeTruncationStrategy = "disabled"
      - type RealtimeTruncationRetentionRatio struct{…}
        
        Retain a fraction of the conversation tokens when the conversation exceeds the input token limit. This allows you to amortize truncations across multiple turns, which can help improve cached token usage.
        
        RetentionRatio float64
        
        Fraction of post-instruction conversation tokens to retain (0.0 - 1.0) when the conversation exceeds the input token limit. Setting this to 0.8 means that messages will be dropped until 80% of the maximum allowed tokens are used. This helps reduce the frequency of truncations and improve cache rates.
        
        Type RetentionRatio
        
        Use retention ratio truncation.
        
        const RetentionRatioRetentionRatio RetentionRatio = "retention_ratio"
        
        TokenLimits RealtimeTruncationRetentionRatioTokenLimits
        
        Optional custom token limits for this truncation strategy. If not provided, the model's default token limits will be used.
        
        PostInstructions int64
        
        Maximum tokens allowed in the conversation after instructions (which including tool definitions). For example, setting this to 5,000 would mean that truncation would occur when the conversation exceeds 5,000 tokens after instructions. This cannot be higher than the model's context window size minus the maximum output tokens.
  - type RealtimeTranscriptionSessionCreateResponse struct{…}
    
    A Realtime transcription session configuration object.
    - ID string
      
      Unique identifier for the session that looks like sess_1234567890abcdef.
    - Object string
      
      The object type. Always realtime.transcription_session.
    - Type Transcription
      
      The type of session. Always transcription for transcription sessions.
      - const TranscriptionTranscription Transcription = "transcription"
    - Audio RealtimeTranscriptionSessionCreateResponseAudio
      
      Configuration for input audio for the session.
      - Input RealtimeTranscriptionSessionCreateResponseAudioInput
        
        Format RealtimeAudioFormatsUnion
        
        The PCM audio format. Only a 24kHz sample rate is supported.
        
        NoiseReduction RealtimeTranscriptionSessionCreateResponseAudioInputNoiseReduction
        
        Configuration for input audio noise reduction.
        
        Type NoiseReductionType
        
        Type of noise reduction. near_field is for close-talking microphones such as headphones, far_field is for far-field microphones such as laptop or conference room microphones.
        
        Transcription AudioTranscription
        
        TurnDetection RealtimeTranscriptionSessionTurnDetection
        
        Configuration for turn detection. Can be set to null to turn off. Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech. For gpt-realtime-whisper, this must be null; VAD is not supported.
        
        PrefixPaddingMs int64
        
        Amount of audio to include before the VAD detected speech (in milliseconds). Defaults to 300ms.
        
        SilenceDurationMs int64
        
        Duration of silence to detect speech stop (in milliseconds). Defaults to 500ms. With shorter values the model will respond more quickly, but may jump in on short pauses from the user.
        
        Threshold float64
        
        Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A higher threshold will require louder audio to activate the model, and thus might perform better in noisy environments.
        
        Type string
        
        Type of turn detection, only server_vad is currently supported.
    - ExpiresAt int64
      
      Expiration timestamp for the session, in seconds since epoch.
    - Include []string
      
      Additional fields to include in server outputs.
      - item.input_audio_transcription.logprobs: Include logprobs for input audio transcription.
      - const RealtimeTranscriptionSessionCreateResponseIncludeItemInputAudioTranscriptionLogprobs RealtimeTranscriptionSessionCreateResponseInclude = "item.input_audio_transcription.logprobs"
- Value string
  
  The generated client secret value.

Example

package main

import (
  "context"
  "fmt"

  "github.com/openai/openai-go"
  "github.com/openai/openai-go/option"
  "github.com/openai/openai-go/realtime"
)

func main() {
  client := openai.NewClient(
    option.WithAPIKey("My API Key"),
  )
  clientSecret, err := client.Realtime.ClientSecrets.New(context.TODO(), realtime.ClientSecretNewParams{

  })
  if err != nil {
    panic(err.Error())
  }
  fmt.Printf("%+v\n", clientSecret.ExpiresAt)
}

Response

{
  "expires_at": 0,
  "session": {
    "id": "id",
    "object": "realtime.session",
    "type": "realtime",
    "audio": {
      "input": {
        "format": {
          "rate": 24000,
          "type": "audio/pcm"
        },
        "noise_reduction": {
          "type": "near_field"
        },
        "transcription": {
          "delay": "minimal",
          "language": "language",
          "model": "whisper-1",
          "prompt": "prompt"
        },
        "turn_detection": {
          "type": "server_vad",
          "create_response": true,
          "idle_timeout_ms": 5000,
          "interrupt_response": true,
          "prefix_padding_ms": 0,
          "silence_duration_ms": 0,
          "threshold": 0
        }
      },
      "output": {
        "format": {
          "rate": 24000,
          "type": "audio/pcm"
        },
        "speed": 0.25,
        "voice": "ash"
      }
    },
    "expires_at": 0,
    "include": [
      "item.input_audio_transcription.logprobs"
    ],
    "instructions": "instructions",
    "max_output_tokens": "inf",
    "model": "gpt-realtime",
    "output_modalities": [
      "text"
    ],
    "prompt": {
      "id": "id",
      "variables": {
        "foo": "string"
      },
      "version": "version"
    },
    "reasoning": {
      "effort": "minimal"
    },
    "tool_choice": "none",
    "tools": [
      {
        "description": "description",
        "name": "name",
        "parameters": {},
        "type": "function"
      }
    ],
    "tracing": "auto",
    "truncation": "auto"
  },
  "value": "value"
}

Domain Types

Realtime Session Create Response

type RealtimeSessionCreateResponse struct{…}

A Realtime session configuration object.
- ID string
  
  Unique identifier for the session that looks like sess_1234567890abcdef.
- Object RealtimeSession
  
  The object type. Always realtime.session.
  - const RealtimeSessionRealtimeSession RealtimeSession = "realtime.session"
- Type Realtime
  
  The type of session to create. Always realtime for the Realtime API.
  - const RealtimeRealtime Realtime = "realtime"
- Audio RealtimeSessionCreateResponseAudio
  
  Configuration for input and output audio.
  - Input RealtimeSessionCreateResponseAudioInput
    - Format RealtimeAudioFormatsUnion
      
      The format of the input audio.
      - type RealtimeAudioFormatsAudioPCM struct{…}
        
        The PCM audio format. Only a 24kHz sample rate is supported.
        
        Rate int64
        
        The sample rate of the audio. Always 24000.
        
        const RealtimeAudioFormatsAudioPCMRate24000 RealtimeAudioFormatsAudioPCMRate = 24000
        
        Type string
        
        The audio format. Always audio/pcm.
        
        const RealtimeAudioFormatsAudioPCMTypeAudioPCM RealtimeAudioFormatsAudioPCMType = "audio/pcm"
      - type RealtimeAudioFormatsAudioPCMU struct{…}
        
        The G.711 μ-law format.
        
        Type string
        
        The audio format. Always audio/pcmu.
        
        const RealtimeAudioFormatsAudioPCMUTypeAudioPCMU RealtimeAudioFormatsAudioPCMUType = "audio/pcmu"
      - type RealtimeAudioFormatsAudioPCMA struct{…}
        
        The G.711 A-law format.
        
        Type string
        
        The audio format. Always audio/pcma.
        
        const RealtimeAudioFormatsAudioPCMATypeAudioPCMA RealtimeAudioFormatsAudioPCMAType = "audio/pcma"
    - NoiseReduction RealtimeSessionCreateResponseAudioInputNoiseReduction
      
      Configuration for input audio noise reduction. This can be set to null to turn off. Noise reduction filters audio added to the input audio buffer before it is sent to VAD and the model. Filtering the audio can improve VAD and turn detection accuracy (reducing false positives) and model performance by improving perception of the input audio.
      - Type NoiseReductionType
        
        Type of noise reduction. near_field is for close-talking microphones such as headphones, far_field is for far-field microphones such as laptop or conference room microphones.
        
        const NoiseReductionTypeNearField NoiseReductionType = "near_field"
        
        const NoiseReductionTypeFarField NoiseReductionType = "far_field"
    - Transcription AudioTranscription
      - Delay AudioTranscriptionDelay
        
        Controls how long the model waits before emitting transcription text. Higher values can improve transcription accuracy at the cost of latency. Only supported with gpt-realtime-whisper in GA Realtime sessions.
        
        const AudioTranscriptionDelayMinimal AudioTranscriptionDelay = "minimal"
        
        const AudioTranscriptionDelayLow AudioTranscriptionDelay = "low"
        
        const AudioTranscriptionDelayMedium AudioTranscriptionDelay = "medium"
        
        const AudioTranscriptionDelayHigh AudioTranscriptionDelay = "high"
        
        const AudioTranscriptionDelayXhigh AudioTranscriptionDelay = "xhigh"
      - Language string
        
        The language of the input audio. Supplying the input language in ISO-639-1 (e.g. en) format will improve accuracy and latency.
      - Model AudioTranscriptionModel
        
        The model to use for transcription. Current options are whisper-1, gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, and gpt-realtime-whisper. Use gpt-4o-transcribe-diarize when you need diarization with speaker labels.
        
        string
        
        type AudioTranscriptionModel string
        
        The model to use for transcription. Current options are whisper-1, gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, and gpt-realtime-whisper. Use gpt-4o-transcribe-diarize when you need diarization with speaker labels.
        
        const AudioTranscriptionModelWhisper1 AudioTranscriptionModel = "whisper-1"
        
        const AudioTranscriptionModelGPT4oMiniTranscribe AudioTranscriptionModel = "gpt-4o-mini-transcribe"
        
        const AudioTranscriptionModelGPT4oMiniTranscribe2025_12_15 AudioTranscriptionModel = "gpt-4o-mini-transcribe-2025-12-15"
        
        const AudioTranscriptionModelGPT4oTranscribe AudioTranscriptionModel = "gpt-4o-transcribe"
        
        const AudioTranscriptionModelGPT4oTranscribeDiarize AudioTranscriptionModel = "gpt-4o-transcribe-diarize"
        
        const AudioTranscriptionModelGPTRealtimeWhisper AudioTranscriptionModel = "gpt-realtime-whisper"
      - Prompt string
        
        An optional text to guide the model's style or continue a previous audio segment. For whisper-1, the prompt is a list of keywords. For gpt-4o-transcribe models (excluding gpt-4o-transcribe-diarize), the prompt is a free text string, for example "expect words related to technology". Prompt is not supported with gpt-realtime-whisper in GA Realtime sessions.
    - TurnDetection RealtimeSessionCreateResponseAudioInputTurnDetectionUnion
      
      Configuration for turn detection, ether Server VAD or Semantic VAD. This can be set to null to turn off, in which case the client must manually trigger model response.
      
      Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech.
      
      Semantic VAD is more advanced and uses a turn detection model (in conjunction with VAD) to semantically estimate whether the user has finished speaking, then dynamically sets a timeout based on this probability. For example, if user audio trails off with "uhhm", the model will score a low probability of turn end and wait longer for the user to continue speaking. This can be useful for more natural conversations, but may have a higher latency.
      
      For gpt-realtime-whisper transcription sessions, turn detection must be set to null; VAD is not supported.
      - type RealtimeSessionCreateResponseAudioInputTurnDetectionServerVad struct{…}
        
        Server-side voice activity detection (VAD) which flips on when user speech is detected and off after a period of silence.
        
        Type ServerVad
        
        Type of turn detection, server_vad to turn on simple Server VAD.
        
        const ServerVadServerVad ServerVad = "server_vad"
        
        CreateResponse bool
        
        Whether or not to automatically generate a response when a VAD stop event occurs. If interrupt_response is set to false this may fail to create a response if the model is already responding.
        
        If both create_response and interrupt_response are set to false, the model will never respond automatically but VAD events will still be emitted.
        
        IdleTimeoutMs int64
        
        Optional timeout after which a model response will be triggered automatically. This is useful for situations in which a long pause from the user is unexpected, such as a phone call. The model will effectively prompt the user to continue the conversation based on the current context.
        
        The timeout value will be applied after the last model response's audio has finished playing, i.e. it's set to the response.done time plus audio playback duration.
        
        An input_audio_buffer.timeout_triggered event (plus events associated with the Response) will be emitted when the timeout is reached. Idle timeout is currently only supported for server_vad mode.
        
        InterruptResponse bool
        
        Whether or not to automatically interrupt (cancel) any ongoing response with output to the default conversation (i.e. conversation of auto) when a VAD start event occurs. If true then the response will be cancelled, otherwise it will continue until complete.
        
        If both create_response and interrupt_response are set to false, the model will never respond automatically but VAD events will still be emitted.
        
        PrefixPaddingMs int64
        
        Used only for server_vad mode. Amount of audio to include before the VAD detected speech (in milliseconds). Defaults to 300ms.
        
        SilenceDurationMs int64
        
        Used only for server_vad mode. Duration of silence to detect speech stop (in milliseconds). Defaults to 500ms. With shorter values the model will respond more quickly, but may jump in on short pauses from the user.
        
        Threshold float64
        
        Used only for server_vad mode. Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A higher threshold will require louder audio to activate the model, and thus might perform better in noisy environments.
      - type RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVad struct{…}
        
        Server-side semantic turn detection which uses a model to determine when the user has finished speaking.
        
        Type SemanticVad
        
        Type of turn detection, semantic_vad to turn on Semantic VAD.
        
        const SemanticVadSemanticVad SemanticVad = "semantic_vad"
        
        CreateResponse bool
        
        Whether or not to automatically generate a response when a VAD stop event occurs.
        
        Eagerness string
        
        Used only for semantic_vad mode. The eagerness of the model to respond. low will wait longer for the user to continue speaking, high will respond more quickly. auto is the default and is equivalent to medium. low, medium, and high have max timeouts of 8s, 4s, and 2s respectively.
        
        const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessLow RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "low"
        
        const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessMedium RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "medium"
        
        const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessHigh RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "high"
        
        const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessAuto RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "auto"
        
        InterruptResponse bool
        
        Whether or not to automatically interrupt any ongoing response with output to the default conversation (i.e. conversation of auto) when a VAD start event occurs.
  - Output RealtimeSessionCreateResponseAudioOutput
    - Format RealtimeAudioFormatsUnion
      
      The format of the output audio.
    - Speed float64
      
      The speed of the model's spoken response as a multiple of the original speed. 1.0 is the default speed. 0.25 is the minimum speed. 1.5 is the maximum speed. This value can only be changed in between model turns, not while a response is in progress.
      
      This parameter is a post-processing adjustment to the audio after it is generated, it's also possible to prompt the model to speak faster or slower.
    - Voice string
      
      The voice the model uses to respond. Voice cannot be changed during the session once the model has responded with audio at least once. Current voice options are alloy, ash, ballad, coral, echo, sage, shimmer, verse, marin, and cedar. We recommend marin and cedar for best quality.
      - string
      - string
        
        const RealtimeSessionCreateResponseAudioOutputVoiceAlloy RealtimeSessionCreateResponseAudioOutputVoice = "alloy"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceAsh RealtimeSessionCreateResponseAudioOutputVoice = "ash"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceBallad RealtimeSessionCreateResponseAudioOutputVoice = "ballad"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceCoral RealtimeSessionCreateResponseAudioOutputVoice = "coral"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceEcho RealtimeSessionCreateResponseAudioOutputVoice = "echo"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceSage RealtimeSessionCreateResponseAudioOutputVoice = "sage"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceShimmer RealtimeSessionCreateResponseAudioOutputVoice = "shimmer"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceVerse RealtimeSessionCreateResponseAudioOutputVoice = "verse"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceMarin RealtimeSessionCreateResponseAudioOutputVoice = "marin"
        
        const RealtimeSessionCreateResponseAudioOutputVoiceCedar RealtimeSessionCreateResponseAudioOutputVoice = "cedar"
- ExpiresAt int64
  
  Expiration timestamp for the session, in seconds since epoch.
- Include []string
  
  Additional fields to include in server outputs.
  
  item.input_audio_transcription.logprobs: Include logprobs for input audio transcription.
  - const RealtimeSessionCreateResponseIncludeItemInputAudioTranscriptionLogprobs RealtimeSessionCreateResponseInclude = "item.input_audio_transcription.logprobs"
- Instructions string
  
  The default system instructions (i.e. system message) prepended to model calls. This field allows the client to guide the model on desired responses. The model can be instructed on response content and format, (e.g. "be extremely succinct", "act friendly", "here are examples of good responses") and on audio behavior (e.g. "talk quickly", "inject emotion into your voice", "laugh frequently"). The instructions are not guaranteed to be followed by the model, but they provide guidance to the model on the desired behavior.
  
  Note that the server sets default instructions which will be used if this field is not set and are visible in the session.created event at the start of the session.
- MaxOutputTokens RealtimeSessionCreateResponseMaxOutputTokensUnion
  
  Maximum number of output tokens for a single assistant response, inclusive of tool calls. Provide an integer between 1 and 4096 to limit output tokens, or inf for the maximum available tokens for a given model. Defaults to inf.
  - int64
  - type Inf string
    - const InfInf Inf = "inf"
- Model RealtimeSessionCreateResponseModel
  
  The Realtime model used for this session.
  - string
  - type RealtimeSessionCreateResponseModel string
    
    The Realtime model used for this session.
    - const RealtimeSessionCreateResponseModelGPTRealtime RealtimeSessionCreateResponseModel = "gpt-realtime"
    - const RealtimeSessionCreateResponseModelGPTRealtime1_5 RealtimeSessionCreateResponseModel = "gpt-realtime-1.5"
    - const RealtimeSessionCreateResponseModelGPTRealtime2 RealtimeSessionCreateResponseModel = "gpt-realtime-2"
    - const RealtimeSessionCreateResponseModelGPTRealtime2025_08_28 RealtimeSessionCreateResponseModel = "gpt-realtime-2025-08-28"
    - const RealtimeSessionCreateResponseModelGPT4oRealtimePreview RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview"
    - const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2024_10_01 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2024-10-01"
    - const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2024_12_17 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2024-12-17"
    - const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2025_06_03 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2025-06-03"
    - const RealtimeSessionCreateResponseModelGPT4oMiniRealtimePreview RealtimeSessionCreateResponseModel = "gpt-4o-mini-realtime-preview"
    - const RealtimeSessionCreateResponseModelGPT4oMiniRealtimePreview2024_12_17 RealtimeSessionCreateResponseModel = "gpt-4o-mini-realtime-preview-2024-12-17"
    - const RealtimeSessionCreateResponseModelGPTRealtimeMini RealtimeSessionCreateResponseModel = "gpt-realtime-mini"
    - const RealtimeSessionCreateResponseModelGPTRealtimeMini2025_10_06 RealtimeSessionCreateResponseModel = "gpt-realtime-mini-2025-10-06"
    - const RealtimeSessionCreateResponseModelGPTRealtimeMini2025_12_15 RealtimeSessionCreateResponseModel = "gpt-realtime-mini-2025-12-15"
    - const RealtimeSessionCreateResponseModelGPTAudio1_5 RealtimeSessionCreateResponseModel = "gpt-audio-1.5"
    - const RealtimeSessionCreateResponseModelGPTAudioMini RealtimeSessionCreateResponseModel = "gpt-audio-mini"
    - const RealtimeSessionCreateResponseModelGPTAudioMini2025_10_06 RealtimeSessionCreateResponseModel = "gpt-audio-mini-2025-10-06"
    - const RealtimeSessionCreateResponseModelGPTAudioMini2025_12_15 RealtimeSessionCreateResponseModel = "gpt-audio-mini-2025-12-15"
- OutputModalities []string
  
  The set of modalities the model can respond with. It defaults to ["audio"], indicating that the model will respond with audio plus a transcript. ["text"] can be used to make the model respond with text only. It is not possible to request both text and audio at the same time.
  - const RealtimeSessionCreateResponseOutputModalityText RealtimeSessionCreateResponseOutputModality = "text"
  - const RealtimeSessionCreateResponseOutputModalityAudio RealtimeSessionCreateResponseOutputModality = "audio"
- Prompt ResponsePrompt
  
  Reference to a prompt template and its variables. Learn more.
  - ID string
    
    The unique identifier of the prompt template to use.
  - Variables map[string, ResponsePromptVariableUnion]
    
    Optional map of values to substitute in for variables in your prompt. The substitution values can either be strings, or other Response input types like images or files.
    - string
    - type ResponseInputText struct{…}
      
      A text input to the model.
      - Text string
        
        The text input to the model.
      - Type InputText
        
        The type of the input item. Always input_text.
        
        const InputTextInputText InputText = "input_text"
    - type ResponseInputImage struct{…}
      
      An image input to the model. Learn about image inputs.
      - Detail ResponseInputImageDetail
        
        The detail level of the image to be sent to the model. One of high, low, auto, or original. Defaults to auto.
        
        const ResponseInputImageDetailLow ResponseInputImageDetail = "low"
        
        const ResponseInputImageDetailHigh ResponseInputImageDetail = "high"
        
        const ResponseInputImageDetailAuto ResponseInputImageDetail = "auto"
        
        const ResponseInputImageDetailOriginal ResponseInputImageDetail = "original"
      - Type InputImage
        
        The type of the input item. Always input_image.
        
        const InputImageInputImage InputImage = "input_image"
      - FileID string
        
        The ID of the file to be sent to the model.
      - ImageURL string
        
        The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL.
    - type ResponseInputFile struct{…}
      
      A file input to the model.
      - Type InputFile
        
        The type of the input item. Always input_file.
        
        const InputFileInputFile InputFile = "input_file"
      - Detail ResponseInputFileDetail
        
        The detail level of the file to be sent to the model. Use low for the default rendering behavior, or high to render the file at higher quality. Defaults to low.
        
        const ResponseInputFileDetailLow ResponseInputFileDetail = "low"
        
        const ResponseInputFileDetailHigh ResponseInputFileDetail = "high"
      - FileData string
        
        The content of the file to be sent to the model.
      - FileID string
        
        The ID of the file to be sent to the model.
      - FileURL string
        
        The URL of the file to be sent to the model.
      - Filename string
        
        The name of the file to be sent to the model.
  - Version string
    
    Optional version of the prompt template.
- Reasoning RealtimeReasoning
  
  Configuration for reasoning-capable Realtime models such as gpt-realtime-2.
  - Effort RealtimeReasoningEffort
    
    Constrains effort on reasoning for reasoning-capable Realtime models such as gpt-realtime-2.
    - const RealtimeReasoningEffortMinimal RealtimeReasoningEffort = "minimal"
    - const RealtimeReasoningEffortLow RealtimeReasoningEffort = "low"
    - const RealtimeReasoningEffortMedium RealtimeReasoningEffort = "medium"
    - const RealtimeReasoningEffortHigh RealtimeReasoningEffort = "high"
    - const RealtimeReasoningEffortXhigh RealtimeReasoningEffort = "xhigh"
- ToolChoice RealtimeSessionCreateResponseToolChoiceUnion
  
  How the model chooses tools. Provide one of the string modes or force a specific function/MCP tool.
  - type ToolChoiceOptions string
    
    Controls which (if any) tool is called by the model.
    
    none means the model will not call any tool and instead generates a message.
    
    auto means the model can pick between generating a message or calling one or more tools.
    
    required means the model must call one or more tools.
    - const ToolChoiceOptionsNone ToolChoiceOptions = "none"
    - const ToolChoiceOptionsAuto ToolChoiceOptions = "auto"
    - const ToolChoiceOptionsRequired ToolChoiceOptions = "required"
  - type ToolChoiceFunction struct{…}
    
    Use this option to force the model to call a specific function.
    - Name string
      
      The name of the function to call.
    - Type Function
      
      For function calling, the type is always function.
      - const FunctionFunction Function = "function"
  - type ToolChoiceMcp struct{…}
    
    Use this option to force the model to call a specific tool on a remote MCP server.
    - ServerLabel string
      
      The label of the MCP server to use.
    - Type Mcp
      
      For MCP tools, the type is always mcp.
      - const McpMcp Mcp = "mcp"
    - Name string
      
      The name of the tool to call on the server.
- Tools []RealtimeSessionCreateResponseToolUnion
  
  Tools available to the model.
  - type RealtimeFunctionTool struct{…}
    - Description string
      
      The description of the function, including guidance on when and how to call it, and guidance about what to tell the user when calling (if anything).
    - Name string
      
      The name of the function.
    - Parameters any
      
      Parameters of the function in JSON Schema.
    - Type RealtimeFunctionToolType
      
      The type of the tool, i.e. function.
      - const RealtimeFunctionToolTypeFunction RealtimeFunctionToolType = "function"
  - type RealtimeSessionCreateResponseToolMcpTool struct{…}
    
    Give the model access to additional tools via remote Model Context Protocol (MCP) servers. Learn more about MCP.
    - ServerLabel string
      
      A label for this MCP server, used to identify it in tool calls.
    - Type Mcp
      
      The type of the MCP tool. Always mcp.
      - const McpMcp Mcp = "mcp"
    - AllowedTools RealtimeSessionCreateResponseToolMcpToolAllowedToolsUnion
      
      List of allowed tool names or a filter object.
      - type RealtimeSessionCreateResponseToolMcpToolAllowedToolsMcpAllowedTools []string
        
        A string array of allowed tool names
      - type RealtimeSessionCreateResponseToolMcpToolAllowedToolsMcpToolFilter struct{…}
        
        A filter object to specify which tools are allowed.
        
        ReadOnly bool
        
        Indicates whether or not a tool modifies data or is read-only. If an MCP server is annotated with readOnlyHint, it will match this filter.
        
        ToolNames []string
        
        List of allowed tool names.
    - Authorization string
      
      An OAuth access token that can be used with a remote MCP server, either with a custom MCP server URL or a service connector. Your application must handle the OAuth authorization flow and provide the token here.
    - ConnectorID string
      
      Identifier for service connectors, like those available in ChatGPT. One of server_url or connector_id must be provided. Learn more about service connectors here.
      
      Currently supported connector_id values are:
      - Dropbox: connector_dropbox
      - Gmail: connector_gmail
      - Google Calendar: connector_googlecalendar
      - Google Drive: connector_googledrive
      - Microsoft Teams: connector_microsoftteams
      - Outlook Calendar: connector_outlookcalendar
      - Outlook Email: connector_outlookemail
      - SharePoint: connector_sharepoint
      - const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorDropbox RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_dropbox"
      - const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGmail RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_gmail"
      - const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGooglecalendar RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_googlecalendar"
      - const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGoogledrive RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_googledrive"
      - const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorMicrosoftteams RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_microsoftteams"
      - const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorOutlookcalendar RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_outlookcalendar"
      - const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorOutlookemail RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_outlookemail"
      - const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorSharepoint RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_sharepoint"
    - DeferLoading bool
      
      Whether this MCP tool is deferred and discovered via tool search.
    - Headers map[string, string]
      
      Optional HTTP headers to send to the MCP server. Use for authentication or other purposes.
    - RequireApproval RealtimeSessionCreateResponseToolMcpToolRequireApprovalUnion
      
      Specify which of the MCP server's tools require approval.
      - type RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilter struct{…}
        
        Specify which of the MCP server's tools require approval. Can be always, never, or a filter object associated with tools that require approval.
        
        Always RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilterAlways
        
        A filter object to specify which tools are allowed.
        
        ReadOnly bool
        
        Indicates whether or not a tool modifies data or is read-only. If an MCP server is annotated with readOnlyHint, it will match this filter.
        
        ToolNames []string
        
        List of allowed tool names.
        
        Never RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilterNever
        
        A filter object to specify which tools are allowed.
        
        ReadOnly bool
        
        Indicates whether or not a tool modifies data or is read-only. If an MCP server is annotated with readOnlyHint, it will match this filter.
        
        ToolNames []string
        
        List of allowed tool names.
      - type RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting string
        
        Specify a single approval policy for all tools. One of always or never. When set to always, all tools will require approval. When set to never, all tools will not require approval.
        
        const RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSettingAlways RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting = "always"
        
        const RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSettingNever RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting = "never"
    - ServerDescription string
      
      Optional description of the MCP server, used to provide more context.
    - ServerURL string
      
      The URL for the MCP server. One of server_url or connector_id must be provided.
- Tracing RealtimeSessionCreateResponseTracingUnion
  
  Realtime API can write session traces to the Traces Dashboard. Set to null to disable tracing. Once tracing is enabled for a session, the configuration cannot be modified.
  
  auto will create a trace for the session with default values for the workflow name, group id, and metadata.
  - type Auto string
    
    Enables tracing and sets default values for tracing configuration options. Always auto.
    - const AutoAuto Auto = "auto"
  - type RealtimeSessionCreateResponseTracingTracingConfiguration struct{…}
    
    Granular configuration for tracing.
    - GroupID string
      
      The group id to attach to this trace to enable filtering and grouping in the Traces Dashboard.
    - Metadata any
      
      The arbitrary metadata to attach to this trace to enable filtering in the Traces Dashboard.
    - WorkflowName string
      
      The name of the workflow to attach to this trace. This is used to name the trace in the Traces Dashboard.
- Truncation RealtimeTruncationUnion
  
  When the number of tokens in a conversation exceeds the model's input token limit, the conversation be truncated, meaning messages (starting from the oldest) will not be included in the model's context. A 32k context model with 4,096 max output tokens can only include 28,224 tokens in the context before truncation occurs.
  
  Clients can configure truncation behavior to truncate with a lower max token limit, which is an effective way to control token usage and cost.
  
  Truncation will reduce the number of cached tokens on the next turn (busting the cache), since messages are dropped from the beginning of the context. However, clients can also configure truncation to retain messages up to a fraction of the maximum context size, which will reduce the need for future truncations and thus improve the cache rate.
  
  Truncation can be disabled entirely, which means the server will never truncate but would instead return an error if the conversation exceeds the model's input token limit.
  - type RealtimeTruncationRealtimeTruncationStrategy string
    
    The truncation strategy to use for the session. auto is the default truncation strategy. disabled will disable truncation and emit errors when the conversation exceeds the input token limit.
    - const RealtimeTruncationRealtimeTruncationStrategyAuto RealtimeTruncationRealtimeTruncationStrategy = "auto"
    - const RealtimeTruncationRealtimeTruncationStrategyDisabled RealtimeTruncationRealtimeTruncationStrategy = "disabled"
  - type RealtimeTruncationRetentionRatio struct{…}
    
    Retain a fraction of the conversation tokens when the conversation exceeds the input token limit. This allows you to amortize truncations across multiple turns, which can help improve cached token usage.
    - RetentionRatio float64
      
      Fraction of post-instruction conversation tokens to retain (0.0 - 1.0) when the conversation exceeds the input token limit. Setting this to 0.8 means that messages will be dropped until 80% of the maximum allowed tokens are used. This helps reduce the frequency of truncations and improve cache rates.
    - Type RetentionRatio
      
      Use retention ratio truncation.
      - const RetentionRatioRetentionRatio RetentionRatio = "retention_ratio"
    - TokenLimits RealtimeTruncationRetentionRatioTokenLimits
      
      Optional custom token limits for this truncation strategy. If not provided, the model's default token limits will be used.
      - PostInstructions int64
        
        Maximum tokens allowed in the conversation after instructions (which including tool definitions). For example, setting this to 5,000 would mean that truncation would occur when the conversation exceeds 5,000 tokens after instructions. This cannot be higher than the model's context window size minus the maximum output tokens.

Realtime Transcription Session Create Response

type RealtimeTranscriptionSessionCreateResponse struct{…}

A Realtime transcription session configuration object.
- ID string
  
  Unique identifier for the session that looks like sess_1234567890abcdef.
- Object string
  
  The object type. Always realtime.transcription_session.
- Type Transcription
  
  The type of session. Always transcription for transcription sessions.
  - const TranscriptionTranscription Transcription = "transcription"
- Audio RealtimeTranscriptionSessionCreateResponseAudio
  
  Configuration for input audio for the session.
  - Input RealtimeTranscriptionSessionCreateResponseAudioInput
    - Format RealtimeAudioFormatsUnion
      
      The PCM audio format. Only a 24kHz sample rate is supported.
      - type RealtimeAudioFormatsAudioPCM struct{…}
        
        The PCM audio format. Only a 24kHz sample rate is supported.
        
        Rate int64
        
        The sample rate of the audio. Always 24000.
        
        const RealtimeAudioFormatsAudioPCMRate24000 RealtimeAudioFormatsAudioPCMRate = 24000
        
        Type string
        
        The audio format. Always audio/pcm.
        
        const RealtimeAudioFormatsAudioPCMTypeAudioPCM RealtimeAudioFormatsAudioPCMType = "audio/pcm"
      - type RealtimeAudioFormatsAudioPCMU struct{…}
        
        The G.711 μ-law format.
        
        Type string
        
        The audio format. Always audio/pcmu.
        
        const RealtimeAudioFormatsAudioPCMUTypeAudioPCMU RealtimeAudioFormatsAudioPCMUType = "audio/pcmu"
      - type RealtimeAudioFormatsAudioPCMA struct{…}
        
        The G.711 A-law format.
        
        Type string
        
        The audio format. Always audio/pcma.
        
        const RealtimeAudioFormatsAudioPCMATypeAudioPCMA RealtimeAudioFormatsAudioPCMAType = "audio/pcma"
    - NoiseReduction RealtimeTranscriptionSessionCreateResponseAudioInputNoiseReduction
      
      Configuration for input audio noise reduction.
      - Type NoiseReductionType
        
        Type of noise reduction. near_field is for close-talking microphones such as headphones, far_field is for far-field microphones such as laptop or conference room microphones.
        
        const NoiseReductionTypeNearField NoiseReductionType = "near_field"
        
        const NoiseReductionTypeFarField NoiseReductionType = "far_field"
    - Transcription AudioTranscription
      - Delay AudioTranscriptionDelay
        
        Controls how long the model waits before emitting transcription text. Higher values can improve transcription accuracy at the cost of latency. Only supported with gpt-realtime-whisper in GA Realtime sessions.
        
        const AudioTranscriptionDelayMinimal AudioTranscriptionDelay = "minimal"
        
        const AudioTranscriptionDelayLow AudioTranscriptionDelay = "low"
        
        const AudioTranscriptionDelayMedium AudioTranscriptionDelay = "medium"
        
        const AudioTranscriptionDelayHigh AudioTranscriptionDelay = "high"
        
        const AudioTranscriptionDelayXhigh AudioTranscriptionDelay = "xhigh"
      - Language string
        
        The language of the input audio. Supplying the input language in ISO-639-1 (e.g. en) format will improve accuracy and latency.
      - Model AudioTranscriptionModel
        
        The model to use for transcription. Current options are whisper-1, gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, and gpt-realtime-whisper. Use gpt-4o-transcribe-diarize when you need diarization with speaker labels.
        
        string
        
        type AudioTranscriptionModel string
        
        The model to use for transcription. Current options are whisper-1, gpt-4o-mini-transcribe, gpt-4o-mini-transcribe-2025-12-15, gpt-4o-transcribe, gpt-4o-transcribe-diarize, and gpt-realtime-whisper. Use gpt-4o-transcribe-diarize when you need diarization with speaker labels.
        
        const AudioTranscriptionModelWhisper1 AudioTranscriptionModel = "whisper-1"
        
        const AudioTranscriptionModelGPT4oMiniTranscribe AudioTranscriptionModel = "gpt-4o-mini-transcribe"
        
        const AudioTranscriptionModelGPT4oMiniTranscribe2025_12_15 AudioTranscriptionModel = "gpt-4o-mini-transcribe-2025-12-15"
        
        const AudioTranscriptionModelGPT4oTranscribe AudioTranscriptionModel = "gpt-4o-transcribe"
        
        const AudioTranscriptionModelGPT4oTranscribeDiarize AudioTranscriptionModel = "gpt-4o-transcribe-diarize"
        
        const AudioTranscriptionModelGPTRealtimeWhisper AudioTranscriptionModel = "gpt-realtime-whisper"
      - Prompt string
        
        An optional text to guide the model's style or continue a previous audio segment. For whisper-1, the prompt is a list of keywords. For gpt-4o-transcribe models (excluding gpt-4o-transcribe-diarize), the prompt is a free text string, for example "expect words related to technology". Prompt is not supported with gpt-realtime-whisper in GA Realtime sessions.
    - TurnDetection RealtimeTranscriptionSessionTurnDetection
      
      Configuration for turn detection. Can be set to null to turn off. Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech. For gpt-realtime-whisper, this must be null; VAD is not supported.
      - PrefixPaddingMs int64
        
        Amount of audio to include before the VAD detected speech (in milliseconds). Defaults to 300ms.
      - SilenceDurationMs int64
        
        Duration of silence to detect speech stop (in milliseconds). Defaults to 500ms. With shorter values the model will respond more quickly, but may jump in on short pauses from the user.
      - Threshold float64
        
        Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A higher threshold will require louder audio to activate the model, and thus might perform better in noisy environments.
      - Type string
        
        Type of turn detection, only server_vad is currently supported.
- ExpiresAt int64
  
  Expiration timestamp for the session, in seconds since epoch.
- Include []string
  
  Additional fields to include in server outputs.
  - item.input_audio_transcription.logprobs: Include logprobs for input audio transcription.
  - const RealtimeTranscriptionSessionCreateResponseIncludeItemInputAudioTranscriptionLogprobs RealtimeTranscriptionSessionCreateResponseInclude = "item.input_audio_transcription.logprobs"

Realtime Transcription Session Turn Detection

type RealtimeTranscriptionSessionTurnDetection struct{…}

Configuration for turn detection. Can be set to null to turn off. Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech. For gpt-realtime-whisper, this must be null; VAD is not supported.
- PrefixPaddingMs int64
  
  Amount of audio to include before the VAD detected speech (in milliseconds). Defaults to 300ms.
- SilenceDurationMs int64
  
  Duration of silence to detect speech stop (in milliseconds). Defaults to 500ms. With shorter values the model will respond more quickly, but may jump in on short pauses from the user.
- Threshold float64
  
  Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A higher threshold will require louder audio to activate the model, and thus might perform better in noisy environments.
- Type string
  
  Type of turn detection, only server_vad is currently supported.

go/resources/realtime/subresources/client_secrets/index.md +2884 −0 created

1# Client Secrets

3## Create client secret

5`client.Realtime.ClientSecrets.New(ctx, body) (*ClientSecretNewResponse, error)`

7**post** `/realtime/client_secrets`

9Create a Realtime client secret with an associated session configuration.

11Client secrets are short-lived tokens that can be passed to a client app,

12such as a web frontend or mobile client, which grants access to the Realtime API without

13leaking your main API key. You can configure a custom TTL for each client secret.

15You can also attach session configuration options to the client secret, which will be

16applied to any sessions created using that client secret, but these can also be overridden

17by the client connection.

19[Learn more about authentication with client secrets over WebRTC](https://platform.openai.com/docs/guides/realtime-webrtc).

21Returns the created client secret and the effective session object. The client secret is a string that looks like `ek_1234`.

23### Parameters

25- `body ClientSecretNewParams`

27 - `ExpiresAfter param.Field[ClientSecretNewParamsExpiresAfter]`

29 Configuration for the client secret expiration. Expiration refers to the time after which

30 a client secret will no longer be valid for creating sessions. The session itself may

31 continue after that time once started. A secret can be used to create multiple sessions

32 until it expires.

34 - `Anchor string`

36 The anchor point for the client secret expiration, meaning that `seconds` will be added to the `created_at` time of the client secret to produce an expiration timestamp. Only `created_at` is currently supported.

38 - `const ClientSecretNewParamsExpiresAfterAnchorCreatedAt ClientSecretNewParamsExpiresAfterAnchor = "created_at"`

40 - `Seconds int64`

42 The number of seconds from the anchor point to the expiration. Select a value between `10` and `7200` (2 hours). This default to 600 seconds (10 minutes) if not specified.

44 - `Session param.Field[ClientSecretNewParamsSessionUnion]`

46 Session configuration to use for the client secret. Choose either a realtime

47 session or a transcription session.

49 - `type RealtimeSessionCreateRequest struct{…}`

51 Realtime session object configuration.

53 - `Type Realtime`

55 The type of session to create. Always `realtime` for the Realtime API.

57 - `const RealtimeRealtime Realtime = "realtime"`

59 - `Audio RealtimeAudioConfig`

61 Configuration for input and output audio.

63 - `Input RealtimeAudioConfigInput`

65 - `Format RealtimeAudioFormatsUnion`

67 The format of the input audio.

69 - `type RealtimeAudioFormatsAudioPCM struct{…}`

71 The PCM audio format. Only a 24kHz sample rate is supported.

73 - `Rate int64`

75 The sample rate of the audio. Always `24000`.

77 - `const RealtimeAudioFormatsAudioPCMRate24000 RealtimeAudioFormatsAudioPCMRate = 24000`

79 - `Type string`

81 The audio format. Always `audio/pcm`.

83 - `const RealtimeAudioFormatsAudioPCMTypeAudioPCM RealtimeAudioFormatsAudioPCMType = "audio/pcm"`

85 - `type RealtimeAudioFormatsAudioPCMU struct{…}`

87 The G.711 μ-law format.

89 - `Type string`

91 The audio format. Always `audio/pcmu`.

93 - `const RealtimeAudioFormatsAudioPCMUTypeAudioPCMU RealtimeAudioFormatsAudioPCMUType = "audio/pcmu"`

95 - `type RealtimeAudioFormatsAudioPCMA struct{…}`

97 The G.711 A-law format.

99 - `Type string`

100

101 The audio format. Always `audio/pcma`.

102

103 - `const RealtimeAudioFormatsAudioPCMATypeAudioPCMA RealtimeAudioFormatsAudioPCMAType = "audio/pcma"`

104

105 - `NoiseReduction RealtimeAudioConfigInputNoiseReduction`

106

107 Configuration for input audio noise reduction. This can be set to `null` to turn off.

108 Noise reduction filters audio added to the input audio buffer before it is sent to VAD and the model.

109 Filtering the audio can improve VAD and turn detection accuracy (reducing false positives) and model performance by improving perception of the input audio.

110

111 - `Type NoiseReductionType`

112

113 Type of noise reduction. `near_field` is for close-talking microphones such as headphones, `far_field` is for far-field microphones such as laptop or conference room microphones.

114

115 - `const NoiseReductionTypeNearField NoiseReductionType = "near_field"`

116

117 - `const NoiseReductionTypeFarField NoiseReductionType = "far_field"`

118

119 - `Transcription AudioTranscription`

120

121 Configuration for input audio transcription, defaults to off and can be set to `null` to turn off once on. Input audio transcription is not native to the model, since the model consumes audio directly. Transcription runs asynchronously through [the /audio/transcriptions endpoint](https://platform.openai.com/docs/api-reference/audio/createTranscription) and should be treated as guidance of input audio content rather than precisely what the model heard. The client can optionally set the language and prompt for transcription, these offer additional guidance to the transcription service.

122

123 - `Delay AudioTranscriptionDelay`

124

125 Controls how long the model waits before emitting transcription text.

126 Higher values can improve transcription accuracy at the cost of latency.

127 Only supported with `gpt-realtime-whisper` in GA Realtime sessions.

128

129 - `const AudioTranscriptionDelayMinimal AudioTranscriptionDelay = "minimal"`

130

131 - `const AudioTranscriptionDelayLow AudioTranscriptionDelay = "low"`

132

133 - `const AudioTranscriptionDelayMedium AudioTranscriptionDelay = "medium"`

134

135 - `const AudioTranscriptionDelayHigh AudioTranscriptionDelay = "high"`

136

137 - `const AudioTranscriptionDelayXhigh AudioTranscriptionDelay = "xhigh"`

138

139 - `Language string`

140

141 The language of the input audio. Supplying the input language in

142 [ISO-639-1](https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) (e.g. `en`) format

143 will improve accuracy and latency.

144

145 - `Model AudioTranscriptionModel`

146

147 The model to use for transcription. Current options are `whisper-1`, `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`, `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`. Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.

148

149 - `string`

150

151 - `type AudioTranscriptionModel string`

152

153 The model to use for transcription. Current options are `whisper-1`, `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`, `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`. Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.

154

155 - `const AudioTranscriptionModelWhisper1 AudioTranscriptionModel = "whisper-1"`

156

157 - `const AudioTranscriptionModelGPT4oMiniTranscribe AudioTranscriptionModel = "gpt-4o-mini-transcribe"`

158

159 - `const AudioTranscriptionModelGPT4oMiniTranscribe2025_12_15 AudioTranscriptionModel = "gpt-4o-mini-transcribe-2025-12-15"`

160

161 - `const AudioTranscriptionModelGPT4oTranscribe AudioTranscriptionModel = "gpt-4o-transcribe"`

162

163 - `const AudioTranscriptionModelGPT4oTranscribeDiarize AudioTranscriptionModel = "gpt-4o-transcribe-diarize"`

164

165 - `const AudioTranscriptionModelGPTRealtimeWhisper AudioTranscriptionModel = "gpt-realtime-whisper"`

166

167 - `Prompt string`

168

169 An optional text to guide the model's style or continue a previous audio

170 segment.

171 For `whisper-1`, the [prompt is a list of keywords](https://platform.openai.com/docs/guides/speech-to-text#prompting).

172 For `gpt-4o-transcribe` models (excluding `gpt-4o-transcribe-diarize`), the prompt is a free text string, for example "expect words related to technology".

173 Prompt is not supported with `gpt-realtime-whisper` in GA Realtime sessions.

174

175 - `TurnDetection RealtimeAudioInputTurnDetectionUnion`

176

177 Configuration for turn detection, ether Server VAD or Semantic VAD. This can be set to `null` to turn off, in which case the client must manually trigger model response.

178

179 Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech.

180

181 Semantic VAD is more advanced and uses a turn detection model (in conjunction with VAD) to semantically estimate whether the user has finished speaking, then dynamically sets a timeout based on this probability. For example, if user audio trails off with "uhhm", the model will score a low probability of turn end and wait longer for the user to continue speaking. This can be useful for more natural conversations, but may have a higher latency.

182

183 For `gpt-realtime-whisper` transcription sessions, turn detection must be

184 set to `null`; VAD is not supported.

185

186 - `RealtimeAudioInputTurnDetectionServerVad`

187

188 - `Type ServerVad`

189

190 Type of turn detection, `server_vad` to turn on simple Server VAD.

191

192 - `const ServerVadServerVad ServerVad = "server_vad"`

193

194 - `CreateResponse bool`

195

196 Whether or not to automatically generate a response when a VAD stop event occurs. If `interrupt_response` is set to `false` this may fail to create a response if the model is already responding.

197

198 If both `create_response` and `interrupt_response` are set to `false`, the model will never respond automatically but VAD events will still be emitted.

199

200 - `IdleTimeoutMs int64`

201

202 Optional timeout after which a model response will be triggered automatically. This is

203 useful for situations in which a long pause from the user is unexpected, such as a phone

204 call. The model will effectively prompt the user to continue the conversation based

205 on the current context.

206

207 The timeout value will be applied after the last model response's audio has finished playing,

208 i.e. it's set to the `response.done` time plus audio playback duration.

209

210 An `input_audio_buffer.timeout_triggered` event (plus events

211 associated with the Response) will be emitted when the timeout is reached.

212 Idle timeout is currently only supported for `server_vad` mode.

213

214 - `InterruptResponse bool`

215

216 Whether or not to automatically interrupt (cancel) any ongoing response with output to the default

217 conversation (i.e. `conversation` of `auto`) when a VAD start event occurs. If `true` then the response will be cancelled, otherwise it will continue until complete.

218

219 If both `create_response` and `interrupt_response` are set to `false`, the model will never respond automatically but VAD events will still be emitted.

220

221 - `PrefixPaddingMs int64`

222

223 Used only for `server_vad` mode. Amount of audio to include before the VAD detected speech (in

224 milliseconds). Defaults to 300ms.

225

226 - `SilenceDurationMs int64`

227

228 Used only for `server_vad` mode. Duration of silence to detect speech stop (in milliseconds). Defaults

229 to 500ms. With shorter values the model will respond more quickly,

230 but may jump in on short pauses from the user.

231

232 - `Threshold float64`

233

234 Used only for `server_vad` mode. Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A

235 higher threshold will require louder audio to activate the model, and

236 thus might perform better in noisy environments.

237

238 - `RealtimeAudioInputTurnDetectionSemanticVad`

239

240 - `Type SemanticVad`

241

242 Type of turn detection, `semantic_vad` to turn on Semantic VAD.

243

244 - `const SemanticVadSemanticVad SemanticVad = "semantic_vad"`

245

246 - `CreateResponse bool`

247

248 Whether or not to automatically generate a response when a VAD stop event occurs.

249

250 - `Eagerness string`

251

252 Used only for `semantic_vad` mode. The eagerness of the model to respond. `low` will wait longer for the user to continue speaking, `high` will respond more quickly. `auto` is the default and is equivalent to `medium`. `low`, `medium`, and `high` have max timeouts of 8s, 4s, and 2s respectively.

253

254 - `const RealtimeAudioInputTurnDetectionSemanticVadEagernessLow RealtimeAudioInputTurnDetectionSemanticVadEagerness = "low"`

255

256 - `const RealtimeAudioInputTurnDetectionSemanticVadEagernessMedium RealtimeAudioInputTurnDetectionSemanticVadEagerness = "medium"`

257

258 - `const RealtimeAudioInputTurnDetectionSemanticVadEagernessHigh RealtimeAudioInputTurnDetectionSemanticVadEagerness = "high"`

259

260 - `const RealtimeAudioInputTurnDetectionSemanticVadEagernessAuto RealtimeAudioInputTurnDetectionSemanticVadEagerness = "auto"`

261

262 - `InterruptResponse bool`

263

264 Whether or not to automatically interrupt any ongoing response with output to the default

265 conversation (i.e. `conversation` of `auto`) when a VAD start event occurs.

266

267 - `Output RealtimeAudioConfigOutput`

268

269 - `Format RealtimeAudioFormatsUnion`

270

271 The format of the output audio.

272

273 - `Speed float64`

274

275 The speed of the model's spoken response as a multiple of the original speed.

276 1.0 is the default speed. 0.25 is the minimum speed. 1.5 is the maximum speed. This value can only be changed in between model turns, not while a response is in progress.

277

278 This parameter is a post-processing adjustment to the audio after it is generated, it's

279 also possible to prompt the model to speak faster or slower.

280

281 - `Voice RealtimeAudioConfigOutputVoiceUnion`

282

283 The voice the model uses to respond. Supported built-in voices are

284 `alloy`, `ash`, `ballad`, `coral`, `echo`, `sage`, `shimmer`, `verse`,

285 `marin`, and `cedar`. You may also provide a custom voice object with

286 an `id`, for example `{ "id": "voice_1234" }`. Voice cannot be changed

287 during the session once the model has responded with audio at least once.

288 We recommend `marin` and `cedar` for best quality.

289

290 - `string`

291

292 - `string`

293

294 - `const RealtimeAudioConfigOutputVoiceString2Alloy RealtimeAudioConfigOutputVoiceString2 = "alloy"`

295

296 - `const RealtimeAudioConfigOutputVoiceString2Ash RealtimeAudioConfigOutputVoiceString2 = "ash"`

297

298 - `const RealtimeAudioConfigOutputVoiceString2Ballad RealtimeAudioConfigOutputVoiceString2 = "ballad"`

299

300 - `const RealtimeAudioConfigOutputVoiceString2Coral RealtimeAudioConfigOutputVoiceString2 = "coral"`

301

302 - `const RealtimeAudioConfigOutputVoiceString2Echo RealtimeAudioConfigOutputVoiceString2 = "echo"`

303

304 - `const RealtimeAudioConfigOutputVoiceString2Sage RealtimeAudioConfigOutputVoiceString2 = "sage"`

305

306 - `const RealtimeAudioConfigOutputVoiceString2Shimmer RealtimeAudioConfigOutputVoiceString2 = "shimmer"`

307

308 - `const RealtimeAudioConfigOutputVoiceString2Verse RealtimeAudioConfigOutputVoiceString2 = "verse"`

309

310 - `const RealtimeAudioConfigOutputVoiceString2Marin RealtimeAudioConfigOutputVoiceString2 = "marin"`

311

312 - `const RealtimeAudioConfigOutputVoiceString2Cedar RealtimeAudioConfigOutputVoiceString2 = "cedar"`

313

314 - `RealtimeAudioConfigOutputVoiceID`

315

316 - `ID string`

317

318 The custom voice ID, e.g. `voice_1234`.

319

320 - `Include []string`

321

322 Additional fields to include in server outputs.

323

324 `item.input_audio_transcription.logprobs`: Include logprobs for input audio transcription.

325

326 - `const RealtimeSessionCreateRequestIncludeItemInputAudioTranscriptionLogprobs RealtimeSessionCreateRequestInclude = "item.input_audio_transcription.logprobs"`

327

328 - `Instructions string`

329

330 The default system instructions (i.e. system message) prepended to model calls. This field allows the client to guide the model on desired responses. The model can be instructed on response content and format, (e.g. "be extremely succinct", "act friendly", "here are examples of good responses") and on audio behavior (e.g. "talk quickly", "inject emotion into your voice", "laugh frequently"). The instructions are not guaranteed to be followed by the model, but they provide guidance to the model on the desired behavior.

331

332 Note that the server sets default instructions which will be used if this field is not set and are visible in the `session.created` event at the start of the session.

333

334 - `MaxOutputTokens RealtimeSessionCreateRequestMaxOutputTokensUnion`

335

336 Maximum number of output tokens for a single assistant response,

337 inclusive of tool calls. Provide an integer between 1 and 4096 to

338 limit output tokens, or `inf` for the maximum available tokens for a

339 given model. Defaults to `inf`.

340

341 - `int64`

342

343 - `Inf`

344

345 - `const InfInf Inf = "inf"`

346

347 - `Model RealtimeSessionCreateRequestModel`

348

349 The Realtime model used for this session.

350

351 - `string`

352

353 - `RealtimeSessionCreateRequestModel`

354

355 - `const RealtimeSessionCreateRequestModelGPTRealtime RealtimeSessionCreateRequestModel = "gpt-realtime"`

356

357 - `const RealtimeSessionCreateRequestModelGPTRealtime1_5 RealtimeSessionCreateRequestModel = "gpt-realtime-1.5"`

358

359 - `const RealtimeSessionCreateRequestModelGPTRealtime2 RealtimeSessionCreateRequestModel = "gpt-realtime-2"`

360

361 - `const RealtimeSessionCreateRequestModelGPTRealtime2025_08_28 RealtimeSessionCreateRequestModel = "gpt-realtime-2025-08-28"`

362

363 - `const RealtimeSessionCreateRequestModelGPT4oRealtimePreview RealtimeSessionCreateRequestModel = "gpt-4o-realtime-preview"`

364

365 - `const RealtimeSessionCreateRequestModelGPT4oRealtimePreview2024_10_01 RealtimeSessionCreateRequestModel = "gpt-4o-realtime-preview-2024-10-01"`

366

367 - `const RealtimeSessionCreateRequestModelGPT4oRealtimePreview2024_12_17 RealtimeSessionCreateRequestModel = "gpt-4o-realtime-preview-2024-12-17"`

368

369 - `const RealtimeSessionCreateRequestModelGPT4oRealtimePreview2025_06_03 RealtimeSessionCreateRequestModel = "gpt-4o-realtime-preview-2025-06-03"`

370

371 - `const RealtimeSessionCreateRequestModelGPT4oMiniRealtimePreview RealtimeSessionCreateRequestModel = "gpt-4o-mini-realtime-preview"`

372

373 - `const RealtimeSessionCreateRequestModelGPT4oMiniRealtimePreview2024_12_17 RealtimeSessionCreateRequestModel = "gpt-4o-mini-realtime-preview-2024-12-17"`

374

375 - `const RealtimeSessionCreateRequestModelGPTRealtimeMini RealtimeSessionCreateRequestModel = "gpt-realtime-mini"`

376

377 - `const RealtimeSessionCreateRequestModelGPTRealtimeMini2025_10_06 RealtimeSessionCreateRequestModel = "gpt-realtime-mini-2025-10-06"`

378

379 - `const RealtimeSessionCreateRequestModelGPTRealtimeMini2025_12_15 RealtimeSessionCreateRequestModel = "gpt-realtime-mini-2025-12-15"`

380

381 - `const RealtimeSessionCreateRequestModelGPTAudio1_5 RealtimeSessionCreateRequestModel = "gpt-audio-1.5"`

382

383 - `const RealtimeSessionCreateRequestModelGPTAudioMini RealtimeSessionCreateRequestModel = "gpt-audio-mini"`

384

385 - `const RealtimeSessionCreateRequestModelGPTAudioMini2025_10_06 RealtimeSessionCreateRequestModel = "gpt-audio-mini-2025-10-06"`

386

387 - `const RealtimeSessionCreateRequestModelGPTAudioMini2025_12_15 RealtimeSessionCreateRequestModel = "gpt-audio-mini-2025-12-15"`

388

389 - `OutputModalities []string`

390

391 The set of modalities the model can respond with. It defaults to `["audio"]`, indicating

392 that the model will respond with audio plus a transcript. `["text"]` can be used to make

393 the model respond with text only. It is not possible to request both `text` and `audio` at the same time.

394

395 - `const RealtimeSessionCreateRequestOutputModalityText RealtimeSessionCreateRequestOutputModality = "text"`

396

397 - `const RealtimeSessionCreateRequestOutputModalityAudio RealtimeSessionCreateRequestOutputModality = "audio"`

398

399 - `ParallelToolCalls bool`

400

401 Whether the model may call multiple tools in parallel. Only supported by

402 reasoning Realtime models such as `gpt-realtime-2`.

403

404 - `Prompt ResponsePrompt`

405

406 Reference to a prompt template and its variables.

407 [Learn more](https://platform.openai.com/docs/guides/text?api-mode=responses#reusable-prompts).

408

409 - `ID string`

410

411 The unique identifier of the prompt template to use.

412

413 - `Variables map[string, ResponsePromptVariableUnion]`

414

415 Optional map of values to substitute in for variables in your

416 prompt. The substitution values can either be strings, or other

417 Response input types like images or files.

418

419 - `string`

420

421 - `type ResponseInputText struct{…}`

422

423 A text input to the model.

424

425 - `Text string`

426

427 The text input to the model.

428

429 - `Type InputText`

430

431 The type of the input item. Always `input_text`.

432

433 - `const InputTextInputText InputText = "input_text"`

434

435 - `type ResponseInputImage struct{…}`

436

437 An image input to the model. Learn about [image inputs](https://platform.openai.com/docs/guides/vision).

438

439 - `Detail ResponseInputImageDetail`

440

441 The detail level of the image to be sent to the model. One of `high`, `low`, `auto`, or `original`. Defaults to `auto`.

442

443 - `const ResponseInputImageDetailLow ResponseInputImageDetail = "low"`

444

445 - `const ResponseInputImageDetailHigh ResponseInputImageDetail = "high"`

446

447 - `const ResponseInputImageDetailAuto ResponseInputImageDetail = "auto"`

448

449 - `const ResponseInputImageDetailOriginal ResponseInputImageDetail = "original"`

450

451 - `Type InputImage`

452

453 The type of the input item. Always `input_image`.

454

455 - `const InputImageInputImage InputImage = "input_image"`

456

457 - `FileID string`

458

459 The ID of the file to be sent to the model.

460

461 - `ImageURL string`

462

463 The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL.

464

465 - `type ResponseInputFile struct{…}`

466

467 A file input to the model.

468

469 - `Type InputFile`

470

471 The type of the input item. Always `input_file`.

472

473 - `const InputFileInputFile InputFile = "input_file"`

474

475 - `Detail ResponseInputFileDetail`

476

477 The detail level of the file to be sent to the model. Use `low` for the default rendering behavior, or `high` to render the file at higher quality. Defaults to `low`.

478

479 - `const ResponseInputFileDetailLow ResponseInputFileDetail = "low"`

480

481 - `const ResponseInputFileDetailHigh ResponseInputFileDetail = "high"`

482

483 - `FileData string`

484

485 The content of the file to be sent to the model.

486

487 - `FileID string`

488

489 The ID of the file to be sent to the model.

490

491 - `FileURL string`

492

493 The URL of the file to be sent to the model.

494

495 - `Filename string`

496

497 The name of the file to be sent to the model.

498

499 - `Version string`

500

501 Optional version of the prompt template.

502

503 - `Reasoning RealtimeReasoning`

504

505 Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.

506

507 - `Effort RealtimeReasoningEffort`

508

509 Constrains effort on reasoning for reasoning-capable Realtime models such as

510 `gpt-realtime-2`.

511

512 - `const RealtimeReasoningEffortMinimal RealtimeReasoningEffort = "minimal"`

513

514 - `const RealtimeReasoningEffortLow RealtimeReasoningEffort = "low"`

515

516 - `const RealtimeReasoningEffortMedium RealtimeReasoningEffort = "medium"`

517

518 - `const RealtimeReasoningEffortHigh RealtimeReasoningEffort = "high"`

519

520 - `const RealtimeReasoningEffortXhigh RealtimeReasoningEffort = "xhigh"`

521

522 - `ToolChoice RealtimeToolChoiceConfigUnion`

523

524 How the model chooses tools. Provide one of the string modes or force a specific

525 function/MCP tool.

526

527 - `type ToolChoiceOptions string`

528

529 Controls which (if any) tool is called by the model.

530

531 `none` means the model will not call any tool and instead generates a message.

532

533 `auto` means the model can pick between generating a message or calling one or

534 more tools.

535

536 `required` means the model must call one or more tools.

537

538 - `const ToolChoiceOptionsNone ToolChoiceOptions = "none"`

539

540 - `const ToolChoiceOptionsAuto ToolChoiceOptions = "auto"`

541

542 - `const ToolChoiceOptionsRequired ToolChoiceOptions = "required"`

543

544 - `type ToolChoiceFunction struct{…}`

545

546 Use this option to force the model to call a specific function.

547

548 - `Name string`

549

550 The name of the function to call.

551

552 - `Type Function`

553

554 For function calling, the type is always `function`.

555

556 - `const FunctionFunction Function = "function"`

557

558 - `type ToolChoiceMcp struct{…}`

559

560 Use this option to force the model to call a specific tool on a remote MCP server.

561

562 - `ServerLabel string`

563

564 The label of the MCP server to use.

565

566 - `Type Mcp`

567

568 For MCP tools, the type is always `mcp`.

569

570 - `const McpMcp Mcp = "mcp"`

571

572 - `Name string`

573

574 The name of the tool to call on the server.

575

576 - `Tools RealtimeToolsConfig`

577

578 Tools available to the model.

579

580 - `type RealtimeFunctionTool struct{…}`

581

582 - `Description string`

583

584 The description of the function, including guidance on when and how

585 to call it, and guidance about what to tell the user when calling

586 (if anything).

587

588 - `Name string`

589

590 The name of the function.

591

592 - `Parameters any`

593

594 Parameters of the function in JSON Schema.

595

596 - `Type RealtimeFunctionToolType`

597

598 The type of the tool, i.e. `function`.

599

600 - `const RealtimeFunctionToolTypeFunction RealtimeFunctionToolType = "function"`

601

602 - `RealtimeToolsConfigUnionMcp`

603

604 - `ServerLabel string`

605

606 A label for this MCP server, used to identify it in tool calls.

607

608 - `Type Mcp`

609

610 The type of the MCP tool. Always `mcp`.

611

612 - `const McpMcp Mcp = "mcp"`

613

614 - `AllowedTools RealtimeToolsConfigUnionMcpAllowedTools`

615

616 List of allowed tool names or a filter object.

617

618 - `[]string`

619

620 - `RealtimeToolsConfigUnionMcpAllowedToolsMcpToolFilter`

621

622 - `ReadOnly bool`

623

624 Indicates whether or not a tool modifies data or is read-only. If an

625 MCP server is [annotated with `readOnlyHint`](https://modelcontextprotocol.io/specification/2025-06-18/schema#toolannotations-readonlyhint),

626 it will match this filter.

627

628 - `ToolNames []string`

629

630 List of allowed tool names.

631

632 - `Authorization string`

633

634 An OAuth access token that can be used with a remote MCP server, either

635 with a custom MCP server URL or a service connector. Your application

636 must handle the OAuth authorization flow and provide the token here.

637

638 - `ConnectorID string`

639

640 Identifier for service connectors, like those available in ChatGPT. One of

641 `server_url` or `connector_id` must be provided. Learn more about service

642 connectors [here](https://platform.openai.com/docs/guides/tools-remote-mcp#connectors).

643

644 Currently supported `connector_id` values are:

645

646 - Dropbox: `connector_dropbox`

647 - Gmail: `connector_gmail`

648 - Google Calendar: `connector_googlecalendar`

649 - Google Drive: `connector_googledrive`

650 - Microsoft Teams: `connector_microsoftteams`

651 - Outlook Calendar: `connector_outlookcalendar`

652 - Outlook Email: `connector_outlookemail`

653 - SharePoint: `connector_sharepoint`

654

655 - `const RealtimeToolsConfigUnionMcpConnectorIDConnectorDropbox RealtimeToolsConfigUnionMcpConnectorID = "connector_dropbox"`

656

657 - `const RealtimeToolsConfigUnionMcpConnectorIDConnectorGmail RealtimeToolsConfigUnionMcpConnectorID = "connector_gmail"`

658

659 - `const RealtimeToolsConfigUnionMcpConnectorIDConnectorGooglecalendar RealtimeToolsConfigUnionMcpConnectorID = "connector_googlecalendar"`

660

661 - `const RealtimeToolsConfigUnionMcpConnectorIDConnectorGoogledrive RealtimeToolsConfigUnionMcpConnectorID = "connector_googledrive"`

662

663 - `const RealtimeToolsConfigUnionMcpConnectorIDConnectorMicrosoftteams RealtimeToolsConfigUnionMcpConnectorID = "connector_microsoftteams"`

664

665 - `const RealtimeToolsConfigUnionMcpConnectorIDConnectorOutlookcalendar RealtimeToolsConfigUnionMcpConnectorID = "connector_outlookcalendar"`

666

667 - `const RealtimeToolsConfigUnionMcpConnectorIDConnectorOutlookemail RealtimeToolsConfigUnionMcpConnectorID = "connector_outlookemail"`

668

669 - `const RealtimeToolsConfigUnionMcpConnectorIDConnectorSharepoint RealtimeToolsConfigUnionMcpConnectorID = "connector_sharepoint"`

670

671 - `DeferLoading bool`

672

673 Whether this MCP tool is deferred and discovered via tool search.

674

675 - `Headers map[string, string]`

676

677 Optional HTTP headers to send to the MCP server. Use for authentication

678 or other purposes.

679

680 - `RequireApproval RealtimeToolsConfigUnionMcpRequireApproval`

681

682 Specify which of the MCP server's tools require approval.

683

684 - `RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalFilter`

685

686 - `Always RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalFilterAlways`

687

688 A filter object to specify which tools are allowed.

689

690 - `ReadOnly bool`

691

692 Indicates whether or not a tool modifies data or is read-only. If an

693 MCP server is [annotated with `readOnlyHint`](https://modelcontextprotocol.io/specification/2025-06-18/schema#toolannotations-readonlyhint),

694 it will match this filter.

695

696 - `ToolNames []string`

697

698 List of allowed tool names.

699

700 - `Never RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalFilterNever`

701

702 A filter object to specify which tools are allowed.

703

704 - `ReadOnly bool`

705

706 Indicates whether or not a tool modifies data or is read-only. If an

707 MCP server is [annotated with `readOnlyHint`](https://modelcontextprotocol.io/specification/2025-06-18/schema#toolannotations-readonlyhint),

708 it will match this filter.

709

710 - `ToolNames []string`

711

712 List of allowed tool names.

713

714 - `string`

715

716 - `const RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalSettingAlways RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalSetting = "always"`

717

718 - `const RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalSettingNever RealtimeToolsConfigUnionMcpRequireApprovalMcpToolApprovalSetting = "never"`

719

720 - `ServerDescription string`

721

722 Optional description of the MCP server, used to provide more context.

723

724 - `ServerURL string`

725

726 The URL for the MCP server. One of `server_url` or `connector_id` must be

727 provided.

728

729 - `Tracing RealtimeTracingConfigUnion`

730

731 Realtime API can write session traces to the [Traces Dashboard](https://platform.openai.com/logs?api=traces). Set to null to disable tracing. Once

732 tracing is enabled for a session, the configuration cannot be modified.

733

734 `auto` will create a trace for the session with default values for the

735 workflow name, group id, and metadata.

736

737 - `Auto`

738

739 - `const AutoAuto Auto = "auto"`

740

741 - `RealtimeTracingConfigTracingConfiguration`

742

743 - `GroupID string`

744

745 The group id to attach to this trace to enable filtering and

746 grouping in the Traces Dashboard.

747

748 - `Metadata any`

749

750 The arbitrary metadata to attach to this trace to enable

751 filtering in the Traces Dashboard.

752

753 - `WorkflowName string`

754

755 The name of the workflow to attach to this trace. This is used to

756 name the trace in the Traces Dashboard.

757

758 - `Truncation RealtimeTruncationUnion`

759

760 When the number of tokens in a conversation exceeds the model's input token limit, the conversation be truncated, meaning messages (starting from the oldest) will not be included in the model's context. A 32k context model with 4,096 max output tokens can only include 28,224 tokens in the context before truncation occurs.

761

762 Clients can configure truncation behavior to truncate with a lower max token limit, which is an effective way to control token usage and cost.

763

764 Truncation will reduce the number of cached tokens on the next turn (busting the cache), since messages are dropped from the beginning of the context. However, clients can also configure truncation to retain messages up to a fraction of the maximum context size, which will reduce the need for future truncations and thus improve the cache rate.

765

766 Truncation can be disabled entirely, which means the server will never truncate but would instead return an error if the conversation exceeds the model's input token limit.

767

768 - `type RealtimeTruncationRealtimeTruncationStrategy string`

769

770 The truncation strategy to use for the session. `auto` is the default truncation strategy. `disabled` will disable truncation and emit errors when the conversation exceeds the input token limit.

771

772 - `const RealtimeTruncationRealtimeTruncationStrategyAuto RealtimeTruncationRealtimeTruncationStrategy = "auto"`

773

774 - `const RealtimeTruncationRealtimeTruncationStrategyDisabled RealtimeTruncationRealtimeTruncationStrategy = "disabled"`

775

776 - `type RealtimeTruncationRetentionRatio struct{…}`

777

778 Retain a fraction of the conversation tokens when the conversation exceeds the input token limit. This allows you to amortize truncations across multiple turns, which can help improve cached token usage.

779

780 - `RetentionRatio float64`

781

782 Fraction of post-instruction conversation tokens to retain (`0.0` - `1.0`) when the conversation exceeds the input token limit. Setting this to `0.8` means that messages will be dropped until 80% of the maximum allowed tokens are used. This helps reduce the frequency of truncations and improve cache rates.

783

784 - `Type RetentionRatio`

785

786 Use retention ratio truncation.

787

788 - `const RetentionRatioRetentionRatio RetentionRatio = "retention_ratio"`

789

790 - `TokenLimits RealtimeTruncationRetentionRatioTokenLimits`

791

792 Optional custom token limits for this truncation strategy. If not provided, the model's default token limits will be used.

793

794 - `PostInstructions int64`

795

796 Maximum tokens allowed in the conversation after instructions (which including tool definitions). For example, setting this to 5,000 would mean that truncation would occur when the conversation exceeds 5,000 tokens after instructions. This cannot be higher than the model's context window size minus the maximum output tokens.

797

798 - `type RealtimeTranscriptionSessionCreateRequest struct{…}`

799

800 Realtime transcription session object configuration.

801

802 - `Type Transcription`

803

804 The type of session to create. Always `transcription` for transcription sessions.

805

806 - `const TranscriptionTranscription Transcription = "transcription"`

807

808 - `Audio RealtimeTranscriptionSessionAudio`

809

810 Configuration for input and output audio.

811

812 - `Input RealtimeTranscriptionSessionAudioInput`

813

814 - `Format RealtimeAudioFormatsUnion`

815

816 The PCM audio format. Only a 24kHz sample rate is supported.

817

818 - `NoiseReduction RealtimeTranscriptionSessionAudioInputNoiseReduction`

819

820 Configuration for input audio noise reduction. This can be set to `null` to turn off.

821 Noise reduction filters audio added to the input audio buffer before it is sent to VAD and the model.

822 Filtering the audio can improve VAD and turn detection accuracy (reducing false positives) and model performance by improving perception of the input audio.

823

824 - `Type NoiseReductionType`

825

826 Type of noise reduction. `near_field` is for close-talking microphones such as headphones, `far_field` is for far-field microphones such as laptop or conference room microphones.

827

828 - `Transcription AudioTranscription`

829

830 Configuration for input audio transcription, defaults to off and can be set to `null` to turn off once on. Input audio transcription is not native to the model, since the model consumes audio directly. Transcription runs asynchronously through [the /audio/transcriptions endpoint](https://platform.openai.com/docs/api-reference/audio/createTranscription) and should be treated as guidance of input audio content rather than precisely what the model heard. The client can optionally set the language and prompt for transcription, these offer additional guidance to the transcription service.

831

832 - `TurnDetection RealtimeTranscriptionSessionAudioInputTurnDetectionUnion`

833

834 Configuration for turn detection, ether Server VAD or Semantic VAD. This can be set to `null` to turn off, in which case the client must manually trigger model response.

835

836 Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech.

837

838 Semantic VAD is more advanced and uses a turn detection model (in conjunction with VAD) to semantically estimate whether the user has finished speaking, then dynamically sets a timeout based on this probability. For example, if user audio trails off with "uhhm", the model will score a low probability of turn end and wait longer for the user to continue speaking. This can be useful for more natural conversations, but may have a higher latency.

839

840 For `gpt-realtime-whisper` transcription sessions, turn detection must be

841 set to `null`; VAD is not supported.

842

843 - `RealtimeTranscriptionSessionAudioInputTurnDetectionServerVad`

844

845 - `Type ServerVad`

846

847 Type of turn detection, `server_vad` to turn on simple Server VAD.

848

849 - `const ServerVadServerVad ServerVad = "server_vad"`

850

851 - `CreateResponse bool`

852

853 Whether or not to automatically generate a response when a VAD stop event occurs. If `interrupt_response` is set to `false` this may fail to create a response if the model is already responding.

854

855 If both `create_response` and `interrupt_response` are set to `false`, the model will never respond automatically but VAD events will still be emitted.

856

857 - `IdleTimeoutMs int64`

858

859 Optional timeout after which a model response will be triggered automatically. This is

860 useful for situations in which a long pause from the user is unexpected, such as a phone

861 call. The model will effectively prompt the user to continue the conversation based

862 on the current context.

863

864 The timeout value will be applied after the last model response's audio has finished playing,

865 i.e. it's set to the `response.done` time plus audio playback duration.

866

867 An `input_audio_buffer.timeout_triggered` event (plus events

868 associated with the Response) will be emitted when the timeout is reached.

869 Idle timeout is currently only supported for `server_vad` mode.

870

871 - `InterruptResponse bool`

872

873 Whether or not to automatically interrupt (cancel) any ongoing response with output to the default

874 conversation (i.e. `conversation` of `auto`) when a VAD start event occurs. If `true` then the response will be cancelled, otherwise it will continue until complete.

875

876 If both `create_response` and `interrupt_response` are set to `false`, the model will never respond automatically but VAD events will still be emitted.

877

878 - `PrefixPaddingMs int64`

879

880 Used only for `server_vad` mode. Amount of audio to include before the VAD detected speech (in

881 milliseconds). Defaults to 300ms.

882

883 - `SilenceDurationMs int64`

884

885 Used only for `server_vad` mode. Duration of silence to detect speech stop (in milliseconds). Defaults

886 to 500ms. With shorter values the model will respond more quickly,

887 but may jump in on short pauses from the user.

888

889 - `Threshold float64`

890

891 Used only for `server_vad` mode. Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A

892 higher threshold will require louder audio to activate the model, and

893 thus might perform better in noisy environments.

894

895 - `RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVad`

896

897 - `Type SemanticVad`

898

899 Type of turn detection, `semantic_vad` to turn on Semantic VAD.

900

901 - `const SemanticVadSemanticVad SemanticVad = "semantic_vad"`

902

903 - `CreateResponse bool`

904

905 Whether or not to automatically generate a response when a VAD stop event occurs.

906

907 - `Eagerness string`

908

909 Used only for `semantic_vad` mode. The eagerness of the model to respond. `low` will wait longer for the user to continue speaking, `high` will respond more quickly. `auto` is the default and is equivalent to `medium`. `low`, `medium`, and `high` have max timeouts of 8s, 4s, and 2s respectively.

910

911 - `const RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagernessLow RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagerness = "low"`

912

913 - `const RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagernessMedium RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagerness = "medium"`

914

915 - `const RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagernessHigh RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagerness = "high"`

916

917 - `const RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagernessAuto RealtimeTranscriptionSessionAudioInputTurnDetectionSemanticVadEagerness = "auto"`

918

919 - `InterruptResponse bool`

920

921 Whether or not to automatically interrupt any ongoing response with output to the default

922 conversation (i.e. `conversation` of `auto`) when a VAD start event occurs.

923

924 - `Include []string`

925

926 Additional fields to include in server outputs.

927

928 `item.input_audio_transcription.logprobs`: Include logprobs for input audio transcription.

929

930 - `const RealtimeTranscriptionSessionCreateRequestIncludeItemInputAudioTranscriptionLogprobs RealtimeTranscriptionSessionCreateRequestInclude = "item.input_audio_transcription.logprobs"`

931

932### Returns

933

934- `type ClientSecretNewResponse struct{…}`

935

936 Response from creating a session and client secret for the Realtime API.

937

938 - `ExpiresAt int64`

939

940 Expiration timestamp for the client secret, in seconds since epoch.

941

942 - `Session ClientSecretNewResponseSessionUnion`

943

944 The session configuration for either a realtime or transcription session.

945

946 - `type RealtimeSessionCreateResponse struct{…}`

947

948 A Realtime session configuration object.

949

950 - `ID string`

951

952 Unique identifier for the session that looks like `sess_1234567890abcdef`.

953

954 - `Object RealtimeSession`

955

956 The object type. Always `realtime.session`.

957

958 - `const RealtimeSessionRealtimeSession RealtimeSession = "realtime.session"`

959

960 - `Type Realtime`

961

962 The type of session to create. Always `realtime` for the Realtime API.

963

964 - `const RealtimeRealtime Realtime = "realtime"`

965

966 - `Audio RealtimeSessionCreateResponseAudio`

967

968 Configuration for input and output audio.

969

970 - `Input RealtimeSessionCreateResponseAudioInput`

971

972 - `Format RealtimeAudioFormatsUnion`

973

974 The format of the input audio.

975

976 - `type RealtimeAudioFormatsAudioPCM struct{…}`

977

978 The PCM audio format. Only a 24kHz sample rate is supported.

979

980 - `Rate int64`

981

982 The sample rate of the audio. Always `24000`.

983

984 - `const RealtimeAudioFormatsAudioPCMRate24000 RealtimeAudioFormatsAudioPCMRate = 24000`

985

986 - `Type string`

987

988 The audio format. Always `audio/pcm`.

989

990 - `const RealtimeAudioFormatsAudioPCMTypeAudioPCM RealtimeAudioFormatsAudioPCMType = "audio/pcm"`

991

992 - `type RealtimeAudioFormatsAudioPCMU struct{…}`

993

994 The G.711 μ-law format.

995

996 - `Type string`

997

998 The audio format. Always `audio/pcmu`.

999

1000 - `const RealtimeAudioFormatsAudioPCMUTypeAudioPCMU RealtimeAudioFormatsAudioPCMUType = "audio/pcmu"`

1001

1002 - `type RealtimeAudioFormatsAudioPCMA struct{…}`

1003

1004 The G.711 A-law format.

1005

1006 - `Type string`

1007

1008 The audio format. Always `audio/pcma`.

1009

1010 - `const RealtimeAudioFormatsAudioPCMATypeAudioPCMA RealtimeAudioFormatsAudioPCMAType = "audio/pcma"`

1011

1012 - `NoiseReduction RealtimeSessionCreateResponseAudioInputNoiseReduction`

1013

1014 Configuration for input audio noise reduction. This can be set to `null` to turn off.

1015 Noise reduction filters audio added to the input audio buffer before it is sent to VAD and the model.

1016 Filtering the audio can improve VAD and turn detection accuracy (reducing false positives) and model performance by improving perception of the input audio.

1017

1018 - `Type NoiseReductionType`

1019

1020 Type of noise reduction. `near_field` is for close-talking microphones such as headphones, `far_field` is for far-field microphones such as laptop or conference room microphones.

1021

1022 - `const NoiseReductionTypeNearField NoiseReductionType = "near_field"`

1023

1024 - `const NoiseReductionTypeFarField NoiseReductionType = "far_field"`

1025

1026 - `Transcription AudioTranscription`

1027

1028 - `Delay AudioTranscriptionDelay`

1029

1030 Controls how long the model waits before emitting transcription text.

1031 Higher values can improve transcription accuracy at the cost of latency.

1032 Only supported with `gpt-realtime-whisper` in GA Realtime sessions.

1033

1034 - `const AudioTranscriptionDelayMinimal AudioTranscriptionDelay = "minimal"`

1035

1036 - `const AudioTranscriptionDelayLow AudioTranscriptionDelay = "low"`

1037

1038 - `const AudioTranscriptionDelayMedium AudioTranscriptionDelay = "medium"`

1039

1040 - `const AudioTranscriptionDelayHigh AudioTranscriptionDelay = "high"`

1041

1042 - `const AudioTranscriptionDelayXhigh AudioTranscriptionDelay = "xhigh"`

1043

1044 - `Language string`

1045

1046 The language of the input audio. Supplying the input language in

1047 [ISO-639-1](https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) (e.g. `en`) format

1048 will improve accuracy and latency.

1049

1050 - `Model AudioTranscriptionModel`

1051

1052 The model to use for transcription. Current options are `whisper-1`, `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`, `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`. Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.

1053

1054 - `string`

1055

1056 - `type AudioTranscriptionModel string`

1057

1058 The model to use for transcription. Current options are `whisper-1`, `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`, `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`. Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.

1059

1060 - `const AudioTranscriptionModelWhisper1 AudioTranscriptionModel = "whisper-1"`

1061

1062 - `const AudioTranscriptionModelGPT4oMiniTranscribe AudioTranscriptionModel = "gpt-4o-mini-transcribe"`

1063

1064 - `const AudioTranscriptionModelGPT4oMiniTranscribe2025_12_15 AudioTranscriptionModel = "gpt-4o-mini-transcribe-2025-12-15"`

1065

1066 - `const AudioTranscriptionModelGPT4oTranscribe AudioTranscriptionModel = "gpt-4o-transcribe"`

1067

1068 - `const AudioTranscriptionModelGPT4oTranscribeDiarize AudioTranscriptionModel = "gpt-4o-transcribe-diarize"`

1069

1070 - `const AudioTranscriptionModelGPTRealtimeWhisper AudioTranscriptionModel = "gpt-realtime-whisper"`

1071

1072 - `Prompt string`

1073

1074 An optional text to guide the model's style or continue a previous audio

1075 segment.

1076 For `whisper-1`, the [prompt is a list of keywords](https://platform.openai.com/docs/guides/speech-to-text#prompting).

1077 For `gpt-4o-transcribe` models (excluding `gpt-4o-transcribe-diarize`), the prompt is a free text string, for example "expect words related to technology".

1078 Prompt is not supported with `gpt-realtime-whisper` in GA Realtime sessions.

1079

1080 - `TurnDetection RealtimeSessionCreateResponseAudioInputTurnDetectionUnion`

1081

1082 Configuration for turn detection, ether Server VAD or Semantic VAD. This can be set to `null` to turn off, in which case the client must manually trigger model response.

1083

1084 Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech.

1085

1086 Semantic VAD is more advanced and uses a turn detection model (in conjunction with VAD) to semantically estimate whether the user has finished speaking, then dynamically sets a timeout based on this probability. For example, if user audio trails off with "uhhm", the model will score a low probability of turn end and wait longer for the user to continue speaking. This can be useful for more natural conversations, but may have a higher latency.

1087

1088 For `gpt-realtime-whisper` transcription sessions, turn detection must be

1089 set to `null`; VAD is not supported.

1090

1091 - `type RealtimeSessionCreateResponseAudioInputTurnDetectionServerVad struct{…}`

1092

1093 Server-side voice activity detection (VAD) which flips on when user speech is detected and off after a period of silence.

1094

1095 - `Type ServerVad`

1096

1097 Type of turn detection, `server_vad` to turn on simple Server VAD.

1098

1099 - `const ServerVadServerVad ServerVad = "server_vad"`

1100

1101 - `CreateResponse bool`

1102

1103 Whether or not to automatically generate a response when a VAD stop event occurs. If `interrupt_response` is set to `false` this may fail to create a response if the model is already responding.

1104

1105 If both `create_response` and `interrupt_response` are set to `false`, the model will never respond automatically but VAD events will still be emitted.

1106

1107 - `IdleTimeoutMs int64`

1108

1109 Optional timeout after which a model response will be triggered automatically. This is

1110 useful for situations in which a long pause from the user is unexpected, such as a phone

1111 call. The model will effectively prompt the user to continue the conversation based

1112 on the current context.

1113

1114 The timeout value will be applied after the last model response's audio has finished playing,

1115 i.e. it's set to the `response.done` time plus audio playback duration.

1116

1117 An `input_audio_buffer.timeout_triggered` event (plus events

1118 associated with the Response) will be emitted when the timeout is reached.

1119 Idle timeout is currently only supported for `server_vad` mode.

1120

1121 - `InterruptResponse bool`

1122

1123 Whether or not to automatically interrupt (cancel) any ongoing response with output to the default

1124 conversation (i.e. `conversation` of `auto`) when a VAD start event occurs. If `true` then the response will be cancelled, otherwise it will continue until complete.

1125

1126 If both `create_response` and `interrupt_response` are set to `false`, the model will never respond automatically but VAD events will still be emitted.

1127

1128 - `PrefixPaddingMs int64`

1129

1130 Used only for `server_vad` mode. Amount of audio to include before the VAD detected speech (in

1131 milliseconds). Defaults to 300ms.

1132

1133 - `SilenceDurationMs int64`

1134

1135 Used only for `server_vad` mode. Duration of silence to detect speech stop (in milliseconds). Defaults

1136 to 500ms. With shorter values the model will respond more quickly,

1137 but may jump in on short pauses from the user.

1138

1139 - `Threshold float64`

1140

1141 Used only for `server_vad` mode. Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A

1142 higher threshold will require louder audio to activate the model, and

1143 thus might perform better in noisy environments.

1144

1145 - `type RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVad struct{…}`

1146

1147 Server-side semantic turn detection which uses a model to determine when the user has finished speaking.

1148

1149 - `Type SemanticVad`

1150

1151 Type of turn detection, `semantic_vad` to turn on Semantic VAD.

1152

1153 - `const SemanticVadSemanticVad SemanticVad = "semantic_vad"`

1154

1155 - `CreateResponse bool`

1156

1157 Whether or not to automatically generate a response when a VAD stop event occurs.

1158

1159 - `Eagerness string`

1160

1161 Used only for `semantic_vad` mode. The eagerness of the model to respond. `low` will wait longer for the user to continue speaking, `high` will respond more quickly. `auto` is the default and is equivalent to `medium`. `low`, `medium`, and `high` have max timeouts of 8s, 4s, and 2s respectively.

1162

1163 - `const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessLow RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "low"`

1164

1165 - `const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessMedium RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "medium"`

1166

1167 - `const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessHigh RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "high"`

1168

1169 - `const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessAuto RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "auto"`

1170

1171 - `InterruptResponse bool`

1172

1173 Whether or not to automatically interrupt any ongoing response with output to the default

1174 conversation (i.e. `conversation` of `auto`) when a VAD start event occurs.

1175

1176 - `Output RealtimeSessionCreateResponseAudioOutput`

1177

1178 - `Format RealtimeAudioFormatsUnion`

1179

1180 The format of the output audio.

1181

1182 - `Speed float64`

1183

1184 The speed of the model's spoken response as a multiple of the original speed.

1185 1.0 is the default speed. 0.25 is the minimum speed. 1.5 is the maximum speed. This value can only be changed in between model turns, not while a response is in progress.

1186

1187 This parameter is a post-processing adjustment to the audio after it is generated, it's

1188 also possible to prompt the model to speak faster or slower.

1189

1190 - `Voice string`

1191

1192 The voice the model uses to respond. Voice cannot be changed during the

1193 session once the model has responded with audio at least once. Current

1194 voice options are `alloy`, `ash`, `ballad`, `coral`, `echo`, `sage`,

1195 `shimmer`, `verse`, `marin`, and `cedar`. We recommend `marin` and `cedar` for

1196 best quality.

1197

1198 - `string`

1199

1200 - `string`

1201

1202 - `const RealtimeSessionCreateResponseAudioOutputVoiceAlloy RealtimeSessionCreateResponseAudioOutputVoice = "alloy"`

1203

1204 - `const RealtimeSessionCreateResponseAudioOutputVoiceAsh RealtimeSessionCreateResponseAudioOutputVoice = "ash"`

1205

1206 - `const RealtimeSessionCreateResponseAudioOutputVoiceBallad RealtimeSessionCreateResponseAudioOutputVoice = "ballad"`

1207

1208 - `const RealtimeSessionCreateResponseAudioOutputVoiceCoral RealtimeSessionCreateResponseAudioOutputVoice = "coral"`

1209

1210 - `const RealtimeSessionCreateResponseAudioOutputVoiceEcho RealtimeSessionCreateResponseAudioOutputVoice = "echo"`

1211

1212 - `const RealtimeSessionCreateResponseAudioOutputVoiceSage RealtimeSessionCreateResponseAudioOutputVoice = "sage"`

1213

1214 - `const RealtimeSessionCreateResponseAudioOutputVoiceShimmer RealtimeSessionCreateResponseAudioOutputVoice = "shimmer"`

1215

1216 - `const RealtimeSessionCreateResponseAudioOutputVoiceVerse RealtimeSessionCreateResponseAudioOutputVoice = "verse"`

1217

1218 - `const RealtimeSessionCreateResponseAudioOutputVoiceMarin RealtimeSessionCreateResponseAudioOutputVoice = "marin"`

1219

1220 - `const RealtimeSessionCreateResponseAudioOutputVoiceCedar RealtimeSessionCreateResponseAudioOutputVoice = "cedar"`

1221

1222 - `ExpiresAt int64`

1223

1224 Expiration timestamp for the session, in seconds since epoch.

1225

1226 - `Include []string`

1227

1228 Additional fields to include in server outputs.

1229

1230 `item.input_audio_transcription.logprobs`: Include logprobs for input audio transcription.

1231

1232 - `const RealtimeSessionCreateResponseIncludeItemInputAudioTranscriptionLogprobs RealtimeSessionCreateResponseInclude = "item.input_audio_transcription.logprobs"`

1233

1234 - `Instructions string`

1235

1236 The default system instructions (i.e. system message) prepended to model calls. This field allows the client to guide the model on desired responses. The model can be instructed on response content and format, (e.g. "be extremely succinct", "act friendly", "here are examples of good responses") and on audio behavior (e.g. "talk quickly", "inject emotion into your voice", "laugh frequently"). The instructions are not guaranteed to be followed by the model, but they provide guidance to the model on the desired behavior.

1237

1238 Note that the server sets default instructions which will be used if this field is not set and are visible in the `session.created` event at the start of the session.

1239

1240 - `MaxOutputTokens RealtimeSessionCreateResponseMaxOutputTokensUnion`

1241

1242 Maximum number of output tokens for a single assistant response,

1243 inclusive of tool calls. Provide an integer between 1 and 4096 to

1244 limit output tokens, or `inf` for the maximum available tokens for a

1245 given model. Defaults to `inf`.

1246

1247 - `int64`

1248

1249 - `type Inf string`

1250

1251 - `const InfInf Inf = "inf"`

1252

1253 - `Model RealtimeSessionCreateResponseModel`

1254

1255 The Realtime model used for this session.

1256

1257 - `string`

1258

1259 - `type RealtimeSessionCreateResponseModel string`

1260

1261 The Realtime model used for this session.

1262

1263 - `const RealtimeSessionCreateResponseModelGPTRealtime RealtimeSessionCreateResponseModel = "gpt-realtime"`

1264

1265 - `const RealtimeSessionCreateResponseModelGPTRealtime1_5 RealtimeSessionCreateResponseModel = "gpt-realtime-1.5"`

1266

1267 - `const RealtimeSessionCreateResponseModelGPTRealtime2 RealtimeSessionCreateResponseModel = "gpt-realtime-2"`

1268

1269 - `const RealtimeSessionCreateResponseModelGPTRealtime2025_08_28 RealtimeSessionCreateResponseModel = "gpt-realtime-2025-08-28"`

1270

1271 - `const RealtimeSessionCreateResponseModelGPT4oRealtimePreview RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview"`

1272

1273 - `const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2024_10_01 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2024-10-01"`

1274

1275 - `const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2024_12_17 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2024-12-17"`

1276

1277 - `const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2025_06_03 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2025-06-03"`

1278

1279 - `const RealtimeSessionCreateResponseModelGPT4oMiniRealtimePreview RealtimeSessionCreateResponseModel = "gpt-4o-mini-realtime-preview"`

1280

1281 - `const RealtimeSessionCreateResponseModelGPT4oMiniRealtimePreview2024_12_17 RealtimeSessionCreateResponseModel = "gpt-4o-mini-realtime-preview-2024-12-17"`

1282

1283 - `const RealtimeSessionCreateResponseModelGPTRealtimeMini RealtimeSessionCreateResponseModel = "gpt-realtime-mini"`

1284

1285 - `const RealtimeSessionCreateResponseModelGPTRealtimeMini2025_10_06 RealtimeSessionCreateResponseModel = "gpt-realtime-mini-2025-10-06"`

1286

1287 - `const RealtimeSessionCreateResponseModelGPTRealtimeMini2025_12_15 RealtimeSessionCreateResponseModel = "gpt-realtime-mini-2025-12-15"`

1288

1289 - `const RealtimeSessionCreateResponseModelGPTAudio1_5 RealtimeSessionCreateResponseModel = "gpt-audio-1.5"`

1290

1291 - `const RealtimeSessionCreateResponseModelGPTAudioMini RealtimeSessionCreateResponseModel = "gpt-audio-mini"`

1292

1293 - `const RealtimeSessionCreateResponseModelGPTAudioMini2025_10_06 RealtimeSessionCreateResponseModel = "gpt-audio-mini-2025-10-06"`

1294

1295 - `const RealtimeSessionCreateResponseModelGPTAudioMini2025_12_15 RealtimeSessionCreateResponseModel = "gpt-audio-mini-2025-12-15"`

1296

1297 - `OutputModalities []string`

1298

1299 The set of modalities the model can respond with. It defaults to `["audio"]`, indicating

1300 that the model will respond with audio plus a transcript. `["text"]` can be used to make

1301 the model respond with text only. It is not possible to request both `text` and `audio` at the same time.

1302

1303 - `const RealtimeSessionCreateResponseOutputModalityText RealtimeSessionCreateResponseOutputModality = "text"`

1304

1305 - `const RealtimeSessionCreateResponseOutputModalityAudio RealtimeSessionCreateResponseOutputModality = "audio"`

1306

1307 - `Prompt ResponsePrompt`

1308

1309 Reference to a prompt template and its variables.

1310 [Learn more](https://platform.openai.com/docs/guides/text?api-mode=responses#reusable-prompts).

1311

1312 - `ID string`

1313

1314 The unique identifier of the prompt template to use.

1315

1316 - `Variables map[string, ResponsePromptVariableUnion]`

1317

1318 Optional map of values to substitute in for variables in your

1319 prompt. The substitution values can either be strings, or other

1320 Response input types like images or files.

1321

1322 - `string`

1323

1324 - `type ResponseInputText struct{…}`

1325

1326 A text input to the model.

1327

1328 - `Text string`

1329

1330 The text input to the model.

1331

1332 - `Type InputText`

1333

1334 The type of the input item. Always `input_text`.

1335

1336 - `const InputTextInputText InputText = "input_text"`

1337

1338 - `type ResponseInputImage struct{…}`

1339

1340 An image input to the model. Learn about [image inputs](https://platform.openai.com/docs/guides/vision).

1341

1342 - `Detail ResponseInputImageDetail`

1343

1344 The detail level of the image to be sent to the model. One of `high`, `low`, `auto`, or `original`. Defaults to `auto`.

1345

1346 - `const ResponseInputImageDetailLow ResponseInputImageDetail = "low"`

1347

1348 - `const ResponseInputImageDetailHigh ResponseInputImageDetail = "high"`

1349

1350 - `const ResponseInputImageDetailAuto ResponseInputImageDetail = "auto"`

1351

1352 - `const ResponseInputImageDetailOriginal ResponseInputImageDetail = "original"`

1353

1354 - `Type InputImage`

1355

1356 The type of the input item. Always `input_image`.

1357

1358 - `const InputImageInputImage InputImage = "input_image"`

1359

1360 - `FileID string`

1361

1362 The ID of the file to be sent to the model.

1363

1364 - `ImageURL string`

1365

1366 The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL.

1367

1368 - `type ResponseInputFile struct{…}`

1369

1370 A file input to the model.

1371

1372 - `Type InputFile`

1373

1374 The type of the input item. Always `input_file`.

1375

1376 - `const InputFileInputFile InputFile = "input_file"`

1377

1378 - `Detail ResponseInputFileDetail`

1379

1380 The detail level of the file to be sent to the model. Use `low` for the default rendering behavior, or `high` to render the file at higher quality. Defaults to `low`.

1381

1382 - `const ResponseInputFileDetailLow ResponseInputFileDetail = "low"`

1383

1384 - `const ResponseInputFileDetailHigh ResponseInputFileDetail = "high"`

1385

1386 - `FileData string`

1387

1388 The content of the file to be sent to the model.

1389

1390 - `FileID string`

1391

1392 The ID of the file to be sent to the model.

1393

1394 - `FileURL string`

1395

1396 The URL of the file to be sent to the model.

1397

1398 - `Filename string`

1399

1400 The name of the file to be sent to the model.

1401

1402 - `Version string`

1403

1404 Optional version of the prompt template.

1405

1406 - `Reasoning RealtimeReasoning`

1407

1408 Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.

1409

1410 - `Effort RealtimeReasoningEffort`

1411

1412 Constrains effort on reasoning for reasoning-capable Realtime models such as

1413 `gpt-realtime-2`.

1414

1415 - `const RealtimeReasoningEffortMinimal RealtimeReasoningEffort = "minimal"`

1416

1417 - `const RealtimeReasoningEffortLow RealtimeReasoningEffort = "low"`

1418

1419 - `const RealtimeReasoningEffortMedium RealtimeReasoningEffort = "medium"`

1420

1421 - `const RealtimeReasoningEffortHigh RealtimeReasoningEffort = "high"`

1422

1423 - `const RealtimeReasoningEffortXhigh RealtimeReasoningEffort = "xhigh"`

1424

1425 - `ToolChoice RealtimeSessionCreateResponseToolChoiceUnion`

1426

1427 How the model chooses tools. Provide one of the string modes or force a specific

1428 function/MCP tool.

1429

1430 - `type ToolChoiceOptions string`

1431

1432 Controls which (if any) tool is called by the model.

1433

1434 `none` means the model will not call any tool and instead generates a message.

1435

1436 `auto` means the model can pick between generating a message or calling one or

1437 more tools.

1438

1439 `required` means the model must call one or more tools.

1440

1441 - `const ToolChoiceOptionsNone ToolChoiceOptions = "none"`

1442

1443 - `const ToolChoiceOptionsAuto ToolChoiceOptions = "auto"`

1444

1445 - `const ToolChoiceOptionsRequired ToolChoiceOptions = "required"`

1446

1447 - `type ToolChoiceFunction struct{…}`

1448

1449 Use this option to force the model to call a specific function.

1450

1451 - `Name string`

1452

1453 The name of the function to call.

1454

1455 - `Type Function`

1456

1457 For function calling, the type is always `function`.

1458

1459 - `const FunctionFunction Function = "function"`

1460

1461 - `type ToolChoiceMcp struct{…}`

1462

1463 Use this option to force the model to call a specific tool on a remote MCP server.

1464

1465 - `ServerLabel string`

1466

1467 The label of the MCP server to use.

1468

1469 - `Type Mcp`

1470

1471 For MCP tools, the type is always `mcp`.

1472

1473 - `const McpMcp Mcp = "mcp"`

1474

1475 - `Name string`

1476

1477 The name of the tool to call on the server.

1478

1479 - `Tools []RealtimeSessionCreateResponseToolUnion`

1480

1481 Tools available to the model.

1482

1483 - `type RealtimeFunctionTool struct{…}`

1484

1485 - `Description string`

1486

1487 The description of the function, including guidance on when and how

1488 to call it, and guidance about what to tell the user when calling

1489 (if anything).

1490

1491 - `Name string`

1492

1493 The name of the function.

1494

1495 - `Parameters any`

1496

1497 Parameters of the function in JSON Schema.

1498

1499 - `Type RealtimeFunctionToolType`

1500

1501 The type of the tool, i.e. `function`.

1502

1503 - `const RealtimeFunctionToolTypeFunction RealtimeFunctionToolType = "function"`

1504

1505 - `type RealtimeSessionCreateResponseToolMcpTool struct{…}`

1506

1507 Give the model access to additional tools via remote Model Context Protocol

1508 (MCP) servers. [Learn more about MCP](https://platform.openai.com/docs/guides/tools-remote-mcp).

1509

1510 - `ServerLabel string`

1511

1512 A label for this MCP server, used to identify it in tool calls.

1513

1514 - `Type Mcp`

1515

1516 The type of the MCP tool. Always `mcp`.

1517

1518 - `const McpMcp Mcp = "mcp"`

1519

1520 - `AllowedTools RealtimeSessionCreateResponseToolMcpToolAllowedToolsUnion`

1521

1522 List of allowed tool names or a filter object.

1523

1524 - `type RealtimeSessionCreateResponseToolMcpToolAllowedToolsMcpAllowedTools []string`

1525

1526 A string array of allowed tool names

1527

1528 - `type RealtimeSessionCreateResponseToolMcpToolAllowedToolsMcpToolFilter struct{…}`

1529

1530 A filter object to specify which tools are allowed.

1531

1532 - `ReadOnly bool`

1533

1534 Indicates whether or not a tool modifies data or is read-only. If an

1535 MCP server is [annotated with `readOnlyHint`](https://modelcontextprotocol.io/specification/2025-06-18/schema#toolannotations-readonlyhint),

1536 it will match this filter.

1537

1538 - `ToolNames []string`

1539

1540 List of allowed tool names.

1541

1542 - `Authorization string`

1543

1544 An OAuth access token that can be used with a remote MCP server, either

1545 with a custom MCP server URL or a service connector. Your application

1546 must handle the OAuth authorization flow and provide the token here.

1547

1548 - `ConnectorID string`

1549

1550 Identifier for service connectors, like those available in ChatGPT. One of

1551 `server_url` or `connector_id` must be provided. Learn more about service

1552 connectors [here](https://platform.openai.com/docs/guides/tools-remote-mcp#connectors).

1553

1554 Currently supported `connector_id` values are:

1555

1556 - Dropbox: `connector_dropbox`

1557 - Gmail: `connector_gmail`

1558 - Google Calendar: `connector_googlecalendar`

1559 - Google Drive: `connector_googledrive`

1560 - Microsoft Teams: `connector_microsoftteams`

1561 - Outlook Calendar: `connector_outlookcalendar`

1562 - Outlook Email: `connector_outlookemail`

1563 - SharePoint: `connector_sharepoint`

1564

1565 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorDropbox RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_dropbox"`

1566

1567 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGmail RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_gmail"`

1568

1569 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGooglecalendar RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_googlecalendar"`

1570

1571 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGoogledrive RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_googledrive"`

1572

1573 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorMicrosoftteams RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_microsoftteams"`

1574

1575 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorOutlookcalendar RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_outlookcalendar"`

1576

1577 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorOutlookemail RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_outlookemail"`

1578

1579 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorSharepoint RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_sharepoint"`

1580

1581 - `DeferLoading bool`

1582

1583 Whether this MCP tool is deferred and discovered via tool search.

1584

1585 - `Headers map[string, string]`

1586

1587 Optional HTTP headers to send to the MCP server. Use for authentication

1588 or other purposes.

1589

1590 - `RequireApproval RealtimeSessionCreateResponseToolMcpToolRequireApprovalUnion`

1591

1592 Specify which of the MCP server's tools require approval.

1593

1594 - `type RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilter struct{…}`

1595

1596 Specify which of the MCP server's tools require approval. Can be

1597 `always`, `never`, or a filter object associated with tools

1598 that require approval.

1599

1600 - `Always RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilterAlways`

1601

1602 A filter object to specify which tools are allowed.

1603

1604 - `ReadOnly bool`

1605

1606 Indicates whether or not a tool modifies data or is read-only. If an

1607 MCP server is [annotated with `readOnlyHint`](https://modelcontextprotocol.io/specification/2025-06-18/schema#toolannotations-readonlyhint),

1608 it will match this filter.

1609

1610 - `ToolNames []string`

1611

1612 List of allowed tool names.

1613

1614 - `Never RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilterNever`

1615

1616 A filter object to specify which tools are allowed.

1617

1618 - `ReadOnly bool`

1619

1620 Indicates whether or not a tool modifies data or is read-only. If an

1621 MCP server is [annotated with `readOnlyHint`](https://modelcontextprotocol.io/specification/2025-06-18/schema#toolannotations-readonlyhint),

1622 it will match this filter.

1623

1624 - `ToolNames []string`

1625

1626 List of allowed tool names.

1627

1628 - `type RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting string`

1629

1630 Specify a single approval policy for all tools. One of `always` or

1631 `never`. When set to `always`, all tools will require approval. When

1632 set to `never`, all tools will not require approval.

1633

1634 - `const RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSettingAlways RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting = "always"`

1635

1636 - `const RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSettingNever RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting = "never"`

1637

1638 - `ServerDescription string`

1639

1640 Optional description of the MCP server, used to provide more context.

1641

1642 - `ServerURL string`

1643

1644 The URL for the MCP server. One of `server_url` or `connector_id` must be

1645 provided.

1646

1647 - `Tracing RealtimeSessionCreateResponseTracingUnion`

1648

1649 Realtime API can write session traces to the [Traces Dashboard](https://platform.openai.com/logs?api=traces). Set to null to disable tracing. Once

1650 tracing is enabled for a session, the configuration cannot be modified.

1651

1652 `auto` will create a trace for the session with default values for the

1653 workflow name, group id, and metadata.

1654

1655 - `type Auto string`

1656

1657 Enables tracing and sets default values for tracing configuration options. Always `auto`.

1658

1659 - `const AutoAuto Auto = "auto"`

1660

1661 - `type RealtimeSessionCreateResponseTracingTracingConfiguration struct{…}`

1662

1663 Granular configuration for tracing.

1664

1665 - `GroupID string`

1666

1667 The group id to attach to this trace to enable filtering and

1668 grouping in the Traces Dashboard.

1669

1670 - `Metadata any`

1671

1672 The arbitrary metadata to attach to this trace to enable

1673 filtering in the Traces Dashboard.

1674

1675 - `WorkflowName string`

1676

1677 The name of the workflow to attach to this trace. This is used to

1678 name the trace in the Traces Dashboard.

1679

1680 - `Truncation RealtimeTruncationUnion`

1681

1682 When the number of tokens in a conversation exceeds the model's input token limit, the conversation be truncated, meaning messages (starting from the oldest) will not be included in the model's context. A 32k context model with 4,096 max output tokens can only include 28,224 tokens in the context before truncation occurs.

1683

1684 Clients can configure truncation behavior to truncate with a lower max token limit, which is an effective way to control token usage and cost.

1685

1686 Truncation will reduce the number of cached tokens on the next turn (busting the cache), since messages are dropped from the beginning of the context. However, clients can also configure truncation to retain messages up to a fraction of the maximum context size, which will reduce the need for future truncations and thus improve the cache rate.

1687

1688 Truncation can be disabled entirely, which means the server will never truncate but would instead return an error if the conversation exceeds the model's input token limit.

1689

1690 - `type RealtimeTruncationRealtimeTruncationStrategy string`

1691

1692 The truncation strategy to use for the session. `auto` is the default truncation strategy. `disabled` will disable truncation and emit errors when the conversation exceeds the input token limit.

1693

1694 - `const RealtimeTruncationRealtimeTruncationStrategyAuto RealtimeTruncationRealtimeTruncationStrategy = "auto"`

1695

1696 - `const RealtimeTruncationRealtimeTruncationStrategyDisabled RealtimeTruncationRealtimeTruncationStrategy = "disabled"`

1697

1698 - `type RealtimeTruncationRetentionRatio struct{…}`

1699

1700 Retain a fraction of the conversation tokens when the conversation exceeds the input token limit. This allows you to amortize truncations across multiple turns, which can help improve cached token usage.

1701

1702 - `RetentionRatio float64`

1703

1704 Fraction of post-instruction conversation tokens to retain (`0.0` - `1.0`) when the conversation exceeds the input token limit. Setting this to `0.8` means that messages will be dropped until 80% of the maximum allowed tokens are used. This helps reduce the frequency of truncations and improve cache rates.

1705

1706 - `Type RetentionRatio`

1707

1708 Use retention ratio truncation.

1709

1710 - `const RetentionRatioRetentionRatio RetentionRatio = "retention_ratio"`

1711

1712 - `TokenLimits RealtimeTruncationRetentionRatioTokenLimits`

1713

1714 Optional custom token limits for this truncation strategy. If not provided, the model's default token limits will be used.

1715

1716 - `PostInstructions int64`

1717

1718 Maximum tokens allowed in the conversation after instructions (which including tool definitions). For example, setting this to 5,000 would mean that truncation would occur when the conversation exceeds 5,000 tokens after instructions. This cannot be higher than the model's context window size minus the maximum output tokens.

1719

1720 - `type RealtimeTranscriptionSessionCreateResponse struct{…}`

1721

1722 A Realtime transcription session configuration object.

1723

1724 - `ID string`

1725

1726 Unique identifier for the session that looks like `sess_1234567890abcdef`.

1727

1728 - `Object string`

1729

1730 The object type. Always `realtime.transcription_session`.

1731

1732 - `Type Transcription`

1733

1734 The type of session. Always `transcription` for transcription sessions.

1735

1736 - `const TranscriptionTranscription Transcription = "transcription"`

1737

1738 - `Audio RealtimeTranscriptionSessionCreateResponseAudio`

1739

1740 Configuration for input audio for the session.

1741

1742 - `Input RealtimeTranscriptionSessionCreateResponseAudioInput`

1743

1744 - `Format RealtimeAudioFormatsUnion`

1745

1746 The PCM audio format. Only a 24kHz sample rate is supported.

1747

1748 - `NoiseReduction RealtimeTranscriptionSessionCreateResponseAudioInputNoiseReduction`

1749

1750 Configuration for input audio noise reduction.

1751

1752 - `Type NoiseReductionType`

1753

1754 Type of noise reduction. `near_field` is for close-talking microphones such as headphones, `far_field` is for far-field microphones such as laptop or conference room microphones.

1755

1756 - `Transcription AudioTranscription`

1757

1758 - `TurnDetection RealtimeTranscriptionSessionTurnDetection`

1759

1760 Configuration for turn detection. Can be set to `null` to turn off. Server

1761 VAD means that the model will detect the start and end of speech based on

1762 audio volume and respond at the end of user speech. For `gpt-realtime-whisper`, this must be `null`; VAD is not supported.

1763

1764 - `PrefixPaddingMs int64`

1765

1766 Amount of audio to include before the VAD detected speech (in

1767 milliseconds). Defaults to 300ms.

1768

1769 - `SilenceDurationMs int64`

1770

1771 Duration of silence to detect speech stop (in milliseconds). Defaults

1772 to 500ms. With shorter values the model will respond more quickly,

1773 but may jump in on short pauses from the user.

1774

1775 - `Threshold float64`

1776

1777 Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A

1778 higher threshold will require louder audio to activate the model, and

1779 thus might perform better in noisy environments.

1780

1781 - `Type string`

1782

1783 Type of turn detection, only `server_vad` is currently supported.

1784

1785 - `ExpiresAt int64`

1786

1787 Expiration timestamp for the session, in seconds since epoch.

1788

1789 - `Include []string`

1790

1791 Additional fields to include in server outputs.

1792

1793 - `item.input_audio_transcription.logprobs`: Include logprobs for input audio transcription.

1794

1795 - `const RealtimeTranscriptionSessionCreateResponseIncludeItemInputAudioTranscriptionLogprobs RealtimeTranscriptionSessionCreateResponseInclude = "item.input_audio_transcription.logprobs"`

1796

1797 - `Value string`

1798

1799 The generated client secret value.

1800

1801### Example

1802

1803```go

1804package main

1805

1806import (

1807 "context"

1808 "fmt"

1809

1810 "github.com/openai/openai-go"

1811 "github.com/openai/openai-go/option"

1812 "github.com/openai/openai-go/realtime"

1813)

1814

1815func main() {

1816 client := openai.NewClient(

1817 option.WithAPIKey("My API Key"),

1818 )

1819 clientSecret, err := client.Realtime.ClientSecrets.New(context.TODO(), realtime.ClientSecretNewParams{

1820

1821 })

1822 if err != nil {

1823 panic(err.Error())

1824 }

1825 fmt.Printf("%+v\n", clientSecret.ExpiresAt)

1826}

1827```

1828

1829#### Response

1830

1831```json

1832{

1833 "expires_at": 0,

1834 "session": {

1835 "id": "id",

1836 "object": "realtime.session",

1837 "type": "realtime",

1838 "audio": {

1839 "input": {

1840 "format": {

1841 "rate": 24000,

1842 "type": "audio/pcm"

1843 },

1844 "noise_reduction": {

1845 "type": "near_field"

1846 },

1847 "transcription": {

1848 "delay": "minimal",

1849 "language": "language",

1850 "model": "whisper-1",

1851 "prompt": "prompt"

1852 },

1853 "turn_detection": {

1854 "type": "server_vad",

1855 "create_response": true,

1856 "idle_timeout_ms": 5000,

1857 "interrupt_response": true,

1858 "prefix_padding_ms": 0,

1859 "silence_duration_ms": 0,

1860 "threshold": 0

1861 }

1862 },

1863 "output": {

1864 "format": {

1865 "rate": 24000,

1866 "type": "audio/pcm"

1867 },

1868 "speed": 0.25,

1869 "voice": "ash"

1870 }

1871 },

1872 "expires_at": 0,

1873 "include": [

1874 "item.input_audio_transcription.logprobs"

1875 ],

1876 "instructions": "instructions",

1877 "max_output_tokens": "inf",

1878 "model": "gpt-realtime",

1879 "output_modalities": [

1880 "text"

1881 ],

1882 "prompt": {

1883 "id": "id",

1884 "variables": {

1885 "foo": "string"

1886 },

1887 "version": "version"

1888 },

1889 "reasoning": {

1890 "effort": "minimal"

1891 },

1892 "tool_choice": "none",

1893 "tools": [

1894 {

1895 "description": "description",

1896 "name": "name",

1897 "parameters": {},

1898 "type": "function"

1899 }

1900 ],

1901 "tracing": "auto",

1902 "truncation": "auto"

1903 },

1904 "value": "value"

1905}

1906```

1907

1908## Domain Types

1909

1910### Realtime Session Create Response

1911

1912- `type RealtimeSessionCreateResponse struct{…}`

1913

1914 A Realtime session configuration object.

1915

1916 - `ID string`

1917

1918 Unique identifier for the session that looks like `sess_1234567890abcdef`.

1919

1920 - `Object RealtimeSession`

1921

1922 The object type. Always `realtime.session`.

1923

1924 - `const RealtimeSessionRealtimeSession RealtimeSession = "realtime.session"`

1925

1926 - `Type Realtime`

1927

1928 The type of session to create. Always `realtime` for the Realtime API.

1929

1930 - `const RealtimeRealtime Realtime = "realtime"`

1931

1932 - `Audio RealtimeSessionCreateResponseAudio`

1933

1934 Configuration for input and output audio.

1935

1936 - `Input RealtimeSessionCreateResponseAudioInput`

1937

1938 - `Format RealtimeAudioFormatsUnion`

1939

1940 The format of the input audio.

1941

1942 - `type RealtimeAudioFormatsAudioPCM struct{…}`

1943

1944 The PCM audio format. Only a 24kHz sample rate is supported.

1945

1946 - `Rate int64`

1947

1948 The sample rate of the audio. Always `24000`.

1949

1950 - `const RealtimeAudioFormatsAudioPCMRate24000 RealtimeAudioFormatsAudioPCMRate = 24000`

1951

1952 - `Type string`

1953

1954 The audio format. Always `audio/pcm`.

1955

1956 - `const RealtimeAudioFormatsAudioPCMTypeAudioPCM RealtimeAudioFormatsAudioPCMType = "audio/pcm"`

1957

1958 - `type RealtimeAudioFormatsAudioPCMU struct{…}`

1959

1960 The G.711 μ-law format.

1961

1962 - `Type string`

1963

1964 The audio format. Always `audio/pcmu`.

1965

1966 - `const RealtimeAudioFormatsAudioPCMUTypeAudioPCMU RealtimeAudioFormatsAudioPCMUType = "audio/pcmu"`

1967

1968 - `type RealtimeAudioFormatsAudioPCMA struct{…}`

1969

1970 The G.711 A-law format.

1971

1972 - `Type string`

1973

1974 The audio format. Always `audio/pcma`.

1975

1976 - `const RealtimeAudioFormatsAudioPCMATypeAudioPCMA RealtimeAudioFormatsAudioPCMAType = "audio/pcma"`

1977

1978 - `NoiseReduction RealtimeSessionCreateResponseAudioInputNoiseReduction`

1979

1980 Configuration for input audio noise reduction. This can be set to `null` to turn off.

1981 Noise reduction filters audio added to the input audio buffer before it is sent to VAD and the model.

1982 Filtering the audio can improve VAD and turn detection accuracy (reducing false positives) and model performance by improving perception of the input audio.

1983

1984 - `Type NoiseReductionType`

1985

1986 Type of noise reduction. `near_field` is for close-talking microphones such as headphones, `far_field` is for far-field microphones such as laptop or conference room microphones.

1987

1988 - `const NoiseReductionTypeNearField NoiseReductionType = "near_field"`

1989

1990 - `const NoiseReductionTypeFarField NoiseReductionType = "far_field"`

1991

1992 - `Transcription AudioTranscription`

1993

1994 - `Delay AudioTranscriptionDelay`

1995

1996 Controls how long the model waits before emitting transcription text.

1997 Higher values can improve transcription accuracy at the cost of latency.

1998 Only supported with `gpt-realtime-whisper` in GA Realtime sessions.

1999

2000 - `const AudioTranscriptionDelayMinimal AudioTranscriptionDelay = "minimal"`

2001

2002 - `const AudioTranscriptionDelayLow AudioTranscriptionDelay = "low"`

2003

2004 - `const AudioTranscriptionDelayMedium AudioTranscriptionDelay = "medium"`

2005

2006 - `const AudioTranscriptionDelayHigh AudioTranscriptionDelay = "high"`

2007

2008 - `const AudioTranscriptionDelayXhigh AudioTranscriptionDelay = "xhigh"`

2009

2010 - `Language string`

2011

2012 The language of the input audio. Supplying the input language in

2013 [ISO-639-1](https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) (e.g. `en`) format

2014 will improve accuracy and latency.

2015

2016 - `Model AudioTranscriptionModel`

2017

2018 The model to use for transcription. Current options are `whisper-1`, `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`, `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`. Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.

2019

2020 - `string`

2021

2022 - `type AudioTranscriptionModel string`

2023

2024 The model to use for transcription. Current options are `whisper-1`, `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`, `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`. Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.

2025

2026 - `const AudioTranscriptionModelWhisper1 AudioTranscriptionModel = "whisper-1"`

2027

2028 - `const AudioTranscriptionModelGPT4oMiniTranscribe AudioTranscriptionModel = "gpt-4o-mini-transcribe"`

2029

2030 - `const AudioTranscriptionModelGPT4oMiniTranscribe2025_12_15 AudioTranscriptionModel = "gpt-4o-mini-transcribe-2025-12-15"`

2031

2032 - `const AudioTranscriptionModelGPT4oTranscribe AudioTranscriptionModel = "gpt-4o-transcribe"`

2033

2034 - `const AudioTranscriptionModelGPT4oTranscribeDiarize AudioTranscriptionModel = "gpt-4o-transcribe-diarize"`

2035

2036 - `const AudioTranscriptionModelGPTRealtimeWhisper AudioTranscriptionModel = "gpt-realtime-whisper"`

2037

2038 - `Prompt string`

2039

2040 An optional text to guide the model's style or continue a previous audio

2041 segment.

2042 For `whisper-1`, the [prompt is a list of keywords](https://platform.openai.com/docs/guides/speech-to-text#prompting).

2043 For `gpt-4o-transcribe` models (excluding `gpt-4o-transcribe-diarize`), the prompt is a free text string, for example "expect words related to technology".

2044 Prompt is not supported with `gpt-realtime-whisper` in GA Realtime sessions.

2045

2046 - `TurnDetection RealtimeSessionCreateResponseAudioInputTurnDetectionUnion`

2047

2048 Configuration for turn detection, ether Server VAD or Semantic VAD. This can be set to `null` to turn off, in which case the client must manually trigger model response.

2049

2050 Server VAD means that the model will detect the start and end of speech based on audio volume and respond at the end of user speech.

2051

2052 Semantic VAD is more advanced and uses a turn detection model (in conjunction with VAD) to semantically estimate whether the user has finished speaking, then dynamically sets a timeout based on this probability. For example, if user audio trails off with "uhhm", the model will score a low probability of turn end and wait longer for the user to continue speaking. This can be useful for more natural conversations, but may have a higher latency.

2053

2054 For `gpt-realtime-whisper` transcription sessions, turn detection must be

2055 set to `null`; VAD is not supported.

2056

2057 - `type RealtimeSessionCreateResponseAudioInputTurnDetectionServerVad struct{…}`

2058

2059 Server-side voice activity detection (VAD) which flips on when user speech is detected and off after a period of silence.

2060

2061 - `Type ServerVad`

2062

2063 Type of turn detection, `server_vad` to turn on simple Server VAD.

2064

2065 - `const ServerVadServerVad ServerVad = "server_vad"`

2066

2067 - `CreateResponse bool`

2068

2069 Whether or not to automatically generate a response when a VAD stop event occurs. If `interrupt_response` is set to `false` this may fail to create a response if the model is already responding.

2070

2071 If both `create_response` and `interrupt_response` are set to `false`, the model will never respond automatically but VAD events will still be emitted.

2072

2073 - `IdleTimeoutMs int64`

2074

2075 Optional timeout after which a model response will be triggered automatically. This is

2076 useful for situations in which a long pause from the user is unexpected, such as a phone

2077 call. The model will effectively prompt the user to continue the conversation based

2078 on the current context.

2079

2080 The timeout value will be applied after the last model response's audio has finished playing,

2081 i.e. it's set to the `response.done` time plus audio playback duration.

2082

2083 An `input_audio_buffer.timeout_triggered` event (plus events

2084 associated with the Response) will be emitted when the timeout is reached.

2085 Idle timeout is currently only supported for `server_vad` mode.

2086

2087 - `InterruptResponse bool`

2088

2089 Whether or not to automatically interrupt (cancel) any ongoing response with output to the default

2090 conversation (i.e. `conversation` of `auto`) when a VAD start event occurs. If `true` then the response will be cancelled, otherwise it will continue until complete.

2091

2092 If both `create_response` and `interrupt_response` are set to `false`, the model will never respond automatically but VAD events will still be emitted.

2093

2094 - `PrefixPaddingMs int64`

2095

2096 Used only for `server_vad` mode. Amount of audio to include before the VAD detected speech (in

2097 milliseconds). Defaults to 300ms.

2098

2099 - `SilenceDurationMs int64`

2100

2101 Used only for `server_vad` mode. Duration of silence to detect speech stop (in milliseconds). Defaults

2102 to 500ms. With shorter values the model will respond more quickly,

2103 but may jump in on short pauses from the user.

2104

2105 - `Threshold float64`

2106

2107 Used only for `server_vad` mode. Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A

2108 higher threshold will require louder audio to activate the model, and

2109 thus might perform better in noisy environments.

2110

2111 - `type RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVad struct{…}`

2112

2113 Server-side semantic turn detection which uses a model to determine when the user has finished speaking.

2114

2115 - `Type SemanticVad`

2116

2117 Type of turn detection, `semantic_vad` to turn on Semantic VAD.

2118

2119 - `const SemanticVadSemanticVad SemanticVad = "semantic_vad"`

2120

2121 - `CreateResponse bool`

2122

2123 Whether or not to automatically generate a response when a VAD stop event occurs.

2124

2125 - `Eagerness string`

2126

2127 Used only for `semantic_vad` mode. The eagerness of the model to respond. `low` will wait longer for the user to continue speaking, `high` will respond more quickly. `auto` is the default and is equivalent to `medium`. `low`, `medium`, and `high` have max timeouts of 8s, 4s, and 2s respectively.

2128

2129 - `const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessLow RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "low"`

2130

2131 - `const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessMedium RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "medium"`

2132

2133 - `const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessHigh RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "high"`

2134

2135 - `const RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagernessAuto RealtimeSessionCreateResponseAudioInputTurnDetectionSemanticVadEagerness = "auto"`

2136

2137 - `InterruptResponse bool`

2138

2139 Whether or not to automatically interrupt any ongoing response with output to the default

2140 conversation (i.e. `conversation` of `auto`) when a VAD start event occurs.

2141

2142 - `Output RealtimeSessionCreateResponseAudioOutput`

2143

2144 - `Format RealtimeAudioFormatsUnion`

2145

2146 The format of the output audio.

2147

2148 - `Speed float64`

2149

2150 The speed of the model's spoken response as a multiple of the original speed.

2151 1.0 is the default speed. 0.25 is the minimum speed. 1.5 is the maximum speed. This value can only be changed in between model turns, not while a response is in progress.

2152

2153 This parameter is a post-processing adjustment to the audio after it is generated, it's

2154 also possible to prompt the model to speak faster or slower.

2155

2156 - `Voice string`

2157

2158 The voice the model uses to respond. Voice cannot be changed during the

2159 session once the model has responded with audio at least once. Current

2160 voice options are `alloy`, `ash`, `ballad`, `coral`, `echo`, `sage`,

2161 `shimmer`, `verse`, `marin`, and `cedar`. We recommend `marin` and `cedar` for

2162 best quality.

2163

2164 - `string`

2165

2166 - `string`

2167

2168 - `const RealtimeSessionCreateResponseAudioOutputVoiceAlloy RealtimeSessionCreateResponseAudioOutputVoice = "alloy"`

2169

2170 - `const RealtimeSessionCreateResponseAudioOutputVoiceAsh RealtimeSessionCreateResponseAudioOutputVoice = "ash"`

2171

2172 - `const RealtimeSessionCreateResponseAudioOutputVoiceBallad RealtimeSessionCreateResponseAudioOutputVoice = "ballad"`

2173

2174 - `const RealtimeSessionCreateResponseAudioOutputVoiceCoral RealtimeSessionCreateResponseAudioOutputVoice = "coral"`

2175

2176 - `const RealtimeSessionCreateResponseAudioOutputVoiceEcho RealtimeSessionCreateResponseAudioOutputVoice = "echo"`

2177

2178 - `const RealtimeSessionCreateResponseAudioOutputVoiceSage RealtimeSessionCreateResponseAudioOutputVoice = "sage"`

2179

2180 - `const RealtimeSessionCreateResponseAudioOutputVoiceShimmer RealtimeSessionCreateResponseAudioOutputVoice = "shimmer"`

2181

2182 - `const RealtimeSessionCreateResponseAudioOutputVoiceVerse RealtimeSessionCreateResponseAudioOutputVoice = "verse"`

2183

2184 - `const RealtimeSessionCreateResponseAudioOutputVoiceMarin RealtimeSessionCreateResponseAudioOutputVoice = "marin"`

2185

2186 - `const RealtimeSessionCreateResponseAudioOutputVoiceCedar RealtimeSessionCreateResponseAudioOutputVoice = "cedar"`

2187

2188 - `ExpiresAt int64`

2189

2190 Expiration timestamp for the session, in seconds since epoch.

2191

2192 - `Include []string`

2193

2194 Additional fields to include in server outputs.

2195

2196 `item.input_audio_transcription.logprobs`: Include logprobs for input audio transcription.

2197

2198 - `const RealtimeSessionCreateResponseIncludeItemInputAudioTranscriptionLogprobs RealtimeSessionCreateResponseInclude = "item.input_audio_transcription.logprobs"`

2199

2200 - `Instructions string`

2201

2202 The default system instructions (i.e. system message) prepended to model calls. This field allows the client to guide the model on desired responses. The model can be instructed on response content and format, (e.g. "be extremely succinct", "act friendly", "here are examples of good responses") and on audio behavior (e.g. "talk quickly", "inject emotion into your voice", "laugh frequently"). The instructions are not guaranteed to be followed by the model, but they provide guidance to the model on the desired behavior.

2203

2204 Note that the server sets default instructions which will be used if this field is not set and are visible in the `session.created` event at the start of the session.

2205

2206 - `MaxOutputTokens RealtimeSessionCreateResponseMaxOutputTokensUnion`

2207

2208 Maximum number of output tokens for a single assistant response,

2209 inclusive of tool calls. Provide an integer between 1 and 4096 to

2210 limit output tokens, or `inf` for the maximum available tokens for a

2211 given model. Defaults to `inf`.

2212

2213 - `int64`

2214

2215 - `type Inf string`

2216

2217 - `const InfInf Inf = "inf"`

2218

2219 - `Model RealtimeSessionCreateResponseModel`

2220

2221 The Realtime model used for this session.

2222

2223 - `string`

2224

2225 - `type RealtimeSessionCreateResponseModel string`

2226

2227 The Realtime model used for this session.

2228

2229 - `const RealtimeSessionCreateResponseModelGPTRealtime RealtimeSessionCreateResponseModel = "gpt-realtime"`

2230

2231 - `const RealtimeSessionCreateResponseModelGPTRealtime1_5 RealtimeSessionCreateResponseModel = "gpt-realtime-1.5"`

2232

2233 - `const RealtimeSessionCreateResponseModelGPTRealtime2 RealtimeSessionCreateResponseModel = "gpt-realtime-2"`

2234

2235 - `const RealtimeSessionCreateResponseModelGPTRealtime2025_08_28 RealtimeSessionCreateResponseModel = "gpt-realtime-2025-08-28"`

2236

2237 - `const RealtimeSessionCreateResponseModelGPT4oRealtimePreview RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview"`

2238

2239 - `const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2024_10_01 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2024-10-01"`

2240

2241 - `const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2024_12_17 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2024-12-17"`

2242

2243 - `const RealtimeSessionCreateResponseModelGPT4oRealtimePreview2025_06_03 RealtimeSessionCreateResponseModel = "gpt-4o-realtime-preview-2025-06-03"`

2244

2245 - `const RealtimeSessionCreateResponseModelGPT4oMiniRealtimePreview RealtimeSessionCreateResponseModel = "gpt-4o-mini-realtime-preview"`

2246

2247 - `const RealtimeSessionCreateResponseModelGPT4oMiniRealtimePreview2024_12_17 RealtimeSessionCreateResponseModel = "gpt-4o-mini-realtime-preview-2024-12-17"`

2248

2249 - `const RealtimeSessionCreateResponseModelGPTRealtimeMini RealtimeSessionCreateResponseModel = "gpt-realtime-mini"`

2250

2251 - `const RealtimeSessionCreateResponseModelGPTRealtimeMini2025_10_06 RealtimeSessionCreateResponseModel = "gpt-realtime-mini-2025-10-06"`

2252

2253 - `const RealtimeSessionCreateResponseModelGPTRealtimeMini2025_12_15 RealtimeSessionCreateResponseModel = "gpt-realtime-mini-2025-12-15"`

2254

2255 - `const RealtimeSessionCreateResponseModelGPTAudio1_5 RealtimeSessionCreateResponseModel = "gpt-audio-1.5"`

2256

2257 - `const RealtimeSessionCreateResponseModelGPTAudioMini RealtimeSessionCreateResponseModel = "gpt-audio-mini"`

2258

2259 - `const RealtimeSessionCreateResponseModelGPTAudioMini2025_10_06 RealtimeSessionCreateResponseModel = "gpt-audio-mini-2025-10-06"`

2260

2261 - `const RealtimeSessionCreateResponseModelGPTAudioMini2025_12_15 RealtimeSessionCreateResponseModel = "gpt-audio-mini-2025-12-15"`

2262

2263 - `OutputModalities []string`

2264

2265 The set of modalities the model can respond with. It defaults to `["audio"]`, indicating

2266 that the model will respond with audio plus a transcript. `["text"]` can be used to make

2267 the model respond with text only. It is not possible to request both `text` and `audio` at the same time.

2268

2269 - `const RealtimeSessionCreateResponseOutputModalityText RealtimeSessionCreateResponseOutputModality = "text"`

2270

2271 - `const RealtimeSessionCreateResponseOutputModalityAudio RealtimeSessionCreateResponseOutputModality = "audio"`

2272

2273 - `Prompt ResponsePrompt`

2274

2275 Reference to a prompt template and its variables.

2276 [Learn more](https://platform.openai.com/docs/guides/text?api-mode=responses#reusable-prompts).

2277

2278 - `ID string`

2279

2280 The unique identifier of the prompt template to use.

2281

2282 - `Variables map[string, ResponsePromptVariableUnion]`

2283

2284 Optional map of values to substitute in for variables in your

2285 prompt. The substitution values can either be strings, or other

2286 Response input types like images or files.

2287

2288 - `string`

2289

2290 - `type ResponseInputText struct{…}`

2291

2292 A text input to the model.

2293

2294 - `Text string`

2295

2296 The text input to the model.

2297

2298 - `Type InputText`

2299

2300 The type of the input item. Always `input_text`.

2301

2302 - `const InputTextInputText InputText = "input_text"`

2303

2304 - `type ResponseInputImage struct{…}`

2305

2306 An image input to the model. Learn about [image inputs](https://platform.openai.com/docs/guides/vision).

2307

2308 - `Detail ResponseInputImageDetail`

2309

2310 The detail level of the image to be sent to the model. One of `high`, `low`, `auto`, or `original`. Defaults to `auto`.

2311

2312 - `const ResponseInputImageDetailLow ResponseInputImageDetail = "low"`

2313

2314 - `const ResponseInputImageDetailHigh ResponseInputImageDetail = "high"`

2315

2316 - `const ResponseInputImageDetailAuto ResponseInputImageDetail = "auto"`

2317

2318 - `const ResponseInputImageDetailOriginal ResponseInputImageDetail = "original"`

2319

2320 - `Type InputImage`

2321

2322 The type of the input item. Always `input_image`.

2323

2324 - `const InputImageInputImage InputImage = "input_image"`

2325

2326 - `FileID string`

2327

2328 The ID of the file to be sent to the model.

2329

2330 - `ImageURL string`

2331

2332 The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL.

2333

2334 - `type ResponseInputFile struct{…}`

2335

2336 A file input to the model.

2337

2338 - `Type InputFile`

2339

2340 The type of the input item. Always `input_file`.

2341

2342 - `const InputFileInputFile InputFile = "input_file"`

2343

2344 - `Detail ResponseInputFileDetail`

2345

2346 The detail level of the file to be sent to the model. Use `low` for the default rendering behavior, or `high` to render the file at higher quality. Defaults to `low`.

2347

2348 - `const ResponseInputFileDetailLow ResponseInputFileDetail = "low"`

2349

2350 - `const ResponseInputFileDetailHigh ResponseInputFileDetail = "high"`

2351

2352 - `FileData string`

2353

2354 The content of the file to be sent to the model.

2355

2356 - `FileID string`

2357

2358 The ID of the file to be sent to the model.

2359

2360 - `FileURL string`

2361

2362 The URL of the file to be sent to the model.

2363

2364 - `Filename string`

2365

2366 The name of the file to be sent to the model.

2367

2368 - `Version string`

2369

2370 Optional version of the prompt template.

2371

2372 - `Reasoning RealtimeReasoning`

2373

2374 Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.

2375

2376 - `Effort RealtimeReasoningEffort`

2377

2378 Constrains effort on reasoning for reasoning-capable Realtime models such as

2379 `gpt-realtime-2`.

2380

2381 - `const RealtimeReasoningEffortMinimal RealtimeReasoningEffort = "minimal"`

2382

2383 - `const RealtimeReasoningEffortLow RealtimeReasoningEffort = "low"`

2384

2385 - `const RealtimeReasoningEffortMedium RealtimeReasoningEffort = "medium"`

2386

2387 - `const RealtimeReasoningEffortHigh RealtimeReasoningEffort = "high"`

2388

2389 - `const RealtimeReasoningEffortXhigh RealtimeReasoningEffort = "xhigh"`

2390

2391 - `ToolChoice RealtimeSessionCreateResponseToolChoiceUnion`

2392

2393 How the model chooses tools. Provide one of the string modes or force a specific

2394 function/MCP tool.

2395

2396 - `type ToolChoiceOptions string`

2397

2398 Controls which (if any) tool is called by the model.

2399

2400 `none` means the model will not call any tool and instead generates a message.

2401

2402 `auto` means the model can pick between generating a message or calling one or

2403 more tools.

2404

2405 `required` means the model must call one or more tools.

2406

2407 - `const ToolChoiceOptionsNone ToolChoiceOptions = "none"`

2408

2409 - `const ToolChoiceOptionsAuto ToolChoiceOptions = "auto"`

2410

2411 - `const ToolChoiceOptionsRequired ToolChoiceOptions = "required"`

2412

2413 - `type ToolChoiceFunction struct{…}`

2414

2415 Use this option to force the model to call a specific function.

2416

2417 - `Name string`

2418

2419 The name of the function to call.

2420

2421 - `Type Function`

2422

2423 For function calling, the type is always `function`.

2424

2425 - `const FunctionFunction Function = "function"`

2426

2427 - `type ToolChoiceMcp struct{…}`

2428

2429 Use this option to force the model to call a specific tool on a remote MCP server.

2430

2431 - `ServerLabel string`

2432

2433 The label of the MCP server to use.

2434

2435 - `Type Mcp`

2436

2437 For MCP tools, the type is always `mcp`.

2438

2439 - `const McpMcp Mcp = "mcp"`

2440

2441 - `Name string`

2442

2443 The name of the tool to call on the server.

2444

2445 - `Tools []RealtimeSessionCreateResponseToolUnion`

2446

2447 Tools available to the model.

2448

2449 - `type RealtimeFunctionTool struct{…}`

2450

2451 - `Description string`

2452

2453 The description of the function, including guidance on when and how

2454 to call it, and guidance about what to tell the user when calling

2455 (if anything).

2456

2457 - `Name string`

2458

2459 The name of the function.

2460

2461 - `Parameters any`

2462

2463 Parameters of the function in JSON Schema.

2464

2465 - `Type RealtimeFunctionToolType`

2466

2467 The type of the tool, i.e. `function`.

2468

2469 - `const RealtimeFunctionToolTypeFunction RealtimeFunctionToolType = "function"`

2470

2471 - `type RealtimeSessionCreateResponseToolMcpTool struct{…}`

2472

2473 Give the model access to additional tools via remote Model Context Protocol

2474 (MCP) servers. [Learn more about MCP](https://platform.openai.com/docs/guides/tools-remote-mcp).

2475

2476 - `ServerLabel string`

2477

2478 A label for this MCP server, used to identify it in tool calls.

2479

2480 - `Type Mcp`

2481

2482 The type of the MCP tool. Always `mcp`.

2483

2484 - `const McpMcp Mcp = "mcp"`

2485

2486 - `AllowedTools RealtimeSessionCreateResponseToolMcpToolAllowedToolsUnion`

2487

2488 List of allowed tool names or a filter object.

2489

2490 - `type RealtimeSessionCreateResponseToolMcpToolAllowedToolsMcpAllowedTools []string`

2491

2492 A string array of allowed tool names

2493

2494 - `type RealtimeSessionCreateResponseToolMcpToolAllowedToolsMcpToolFilter struct{…}`

2495

2496 A filter object to specify which tools are allowed.

2497

2498 - `ReadOnly bool`

2499

2500 Indicates whether or not a tool modifies data or is read-only. If an

2501 MCP server is [annotated with `readOnlyHint`](https://modelcontextprotocol.io/specification/2025-06-18/schema#toolannotations-readonlyhint),

2502 it will match this filter.

2503

2504 - `ToolNames []string`

2505

2506 List of allowed tool names.

2507

2508 - `Authorization string`

2509

2510 An OAuth access token that can be used with a remote MCP server, either

2511 with a custom MCP server URL or a service connector. Your application

2512 must handle the OAuth authorization flow and provide the token here.

2513

2514 - `ConnectorID string`

2515

2516 Identifier for service connectors, like those available in ChatGPT. One of

2517 `server_url` or `connector_id` must be provided. Learn more about service

2518 connectors [here](https://platform.openai.com/docs/guides/tools-remote-mcp#connectors).

2519

2520 Currently supported `connector_id` values are:

2521

2522 - Dropbox: `connector_dropbox`

2523 - Gmail: `connector_gmail`

2524 - Google Calendar: `connector_googlecalendar`

2525 - Google Drive: `connector_googledrive`

2526 - Microsoft Teams: `connector_microsoftteams`

2527 - Outlook Calendar: `connector_outlookcalendar`

2528 - Outlook Email: `connector_outlookemail`

2529 - SharePoint: `connector_sharepoint`

2530

2531 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorDropbox RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_dropbox"`

2532

2533 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGmail RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_gmail"`

2534

2535 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGooglecalendar RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_googlecalendar"`

2536

2537 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorGoogledrive RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_googledrive"`

2538

2539 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorMicrosoftteams RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_microsoftteams"`

2540

2541 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorOutlookcalendar RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_outlookcalendar"`

2542

2543 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorOutlookemail RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_outlookemail"`

2544

2545 - `const RealtimeSessionCreateResponseToolMcpToolConnectorIDConnectorSharepoint RealtimeSessionCreateResponseToolMcpToolConnectorID = "connector_sharepoint"`

2546

2547 - `DeferLoading bool`

2548

2549 Whether this MCP tool is deferred and discovered via tool search.

2550

2551 - `Headers map[string, string]`

2552

2553 Optional HTTP headers to send to the MCP server. Use for authentication

2554 or other purposes.

2555

2556 - `RequireApproval RealtimeSessionCreateResponseToolMcpToolRequireApprovalUnion`

2557

2558 Specify which of the MCP server's tools require approval.

2559

2560 - `type RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilter struct{…}`

2561

2562 Specify which of the MCP server's tools require approval. Can be

2563 `always`, `never`, or a filter object associated with tools

2564 that require approval.

2565

2566 - `Always RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilterAlways`

2567

2568 A filter object to specify which tools are allowed.

2569

2570 - `ReadOnly bool`

2571

2572 Indicates whether or not a tool modifies data or is read-only. If an

2573 MCP server is [annotated with `readOnlyHint`](https://modelcontextprotocol.io/specification/2025-06-18/schema#toolannotations-readonlyhint),

2574 it will match this filter.

2575

2576 - `ToolNames []string`

2577

2578 List of allowed tool names.

2579

2580 - `Never RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalFilterNever`

2581

2582 A filter object to specify which tools are allowed.

2583

2584 - `ReadOnly bool`

2585

2586 Indicates whether or not a tool modifies data or is read-only. If an

2587 MCP server is [annotated with `readOnlyHint`](https://modelcontextprotocol.io/specification/2025-06-18/schema#toolannotations-readonlyhint),

2588 it will match this filter.

2589

2590 - `ToolNames []string`

2591

2592 List of allowed tool names.

2593

2594 - `type RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting string`

2595

2596 Specify a single approval policy for all tools. One of `always` or

2597 `never`. When set to `always`, all tools will require approval. When

2598 set to `never`, all tools will not require approval.

2599

2600 - `const RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSettingAlways RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting = "always"`

2601

2602 - `const RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSettingNever RealtimeSessionCreateResponseToolMcpToolRequireApprovalMcpToolApprovalSetting = "never"`

2603

2604 - `ServerDescription string`

2605

2606 Optional description of the MCP server, used to provide more context.

2607

2608 - `ServerURL string`

2609

2610 The URL for the MCP server. One of `server_url` or `connector_id` must be

2611 provided.

2612

2613 - `Tracing RealtimeSessionCreateResponseTracingUnion`

2614

2615 Realtime API can write session traces to the [Traces Dashboard](https://platform.openai.com/logs?api=traces). Set to null to disable tracing. Once

2616 tracing is enabled for a session, the configuration cannot be modified.

2617

2618 `auto` will create a trace for the session with default values for the

2619 workflow name, group id, and metadata.

2620

2621 - `type Auto string`

2622

2623 Enables tracing and sets default values for tracing configuration options. Always `auto`.

2624

2625 - `const AutoAuto Auto = "auto"`

2626

2627 - `type RealtimeSessionCreateResponseTracingTracingConfiguration struct{…}`

2628

2629 Granular configuration for tracing.

2630

2631 - `GroupID string`

2632

2633 The group id to attach to this trace to enable filtering and

2634 grouping in the Traces Dashboard.

2635

2636 - `Metadata any`

2637

2638 The arbitrary metadata to attach to this trace to enable

2639 filtering in the Traces Dashboard.

2640

2641 - `WorkflowName string`

2642

2643 The name of the workflow to attach to this trace. This is used to

2644 name the trace in the Traces Dashboard.

2645

2646 - `Truncation RealtimeTruncationUnion`

2647

2648 When the number of tokens in a conversation exceeds the model's input token limit, the conversation be truncated, meaning messages (starting from the oldest) will not be included in the model's context. A 32k context model with 4,096 max output tokens can only include 28,224 tokens in the context before truncation occurs.

2649

2650 Clients can configure truncation behavior to truncate with a lower max token limit, which is an effective way to control token usage and cost.

2651

2652 Truncation will reduce the number of cached tokens on the next turn (busting the cache), since messages are dropped from the beginning of the context. However, clients can also configure truncation to retain messages up to a fraction of the maximum context size, which will reduce the need for future truncations and thus improve the cache rate.

2653

2654 Truncation can be disabled entirely, which means the server will never truncate but would instead return an error if the conversation exceeds the model's input token limit.

2655

2656 - `type RealtimeTruncationRealtimeTruncationStrategy string`

2657

2658 The truncation strategy to use for the session. `auto` is the default truncation strategy. `disabled` will disable truncation and emit errors when the conversation exceeds the input token limit.

2659

2660 - `const RealtimeTruncationRealtimeTruncationStrategyAuto RealtimeTruncationRealtimeTruncationStrategy = "auto"`

2661

2662 - `const RealtimeTruncationRealtimeTruncationStrategyDisabled RealtimeTruncationRealtimeTruncationStrategy = "disabled"`

2663

2664 - `type RealtimeTruncationRetentionRatio struct{…}`

2665

2666 Retain a fraction of the conversation tokens when the conversation exceeds the input token limit. This allows you to amortize truncations across multiple turns, which can help improve cached token usage.

2667

2668 - `RetentionRatio float64`

2669

2670 Fraction of post-instruction conversation tokens to retain (`0.0` - `1.0`) when the conversation exceeds the input token limit. Setting this to `0.8` means that messages will be dropped until 80% of the maximum allowed tokens are used. This helps reduce the frequency of truncations and improve cache rates.

2671

2672 - `Type RetentionRatio`

2673

2674 Use retention ratio truncation.

2675

2676 - `const RetentionRatioRetentionRatio RetentionRatio = "retention_ratio"`

2677

2678 - `TokenLimits RealtimeTruncationRetentionRatioTokenLimits`

2679

2680 Optional custom token limits for this truncation strategy. If not provided, the model's default token limits will be used.

2681

2682 - `PostInstructions int64`

2683

2684 Maximum tokens allowed in the conversation after instructions (which including tool definitions). For example, setting this to 5,000 would mean that truncation would occur when the conversation exceeds 5,000 tokens after instructions. This cannot be higher than the model's context window size minus the maximum output tokens.

2685

2686### Realtime Transcription Session Create Response

2687

2688- `type RealtimeTranscriptionSessionCreateResponse struct{…}`

2689

2690 A Realtime transcription session configuration object.

2691

2692 - `ID string`

2693

2694 Unique identifier for the session that looks like `sess_1234567890abcdef`.

2695

2696 - `Object string`

2697

2698 The object type. Always `realtime.transcription_session`.

2699

2700 - `Type Transcription`

2701

2702 The type of session. Always `transcription` for transcription sessions.

2703

2704 - `const TranscriptionTranscription Transcription = "transcription"`

2705

2706 - `Audio RealtimeTranscriptionSessionCreateResponseAudio`

2707

2708 Configuration for input audio for the session.

2709

2710 - `Input RealtimeTranscriptionSessionCreateResponseAudioInput`

2711

2712 - `Format RealtimeAudioFormatsUnion`

2713

2714 The PCM audio format. Only a 24kHz sample rate is supported.

2715

2716 - `type RealtimeAudioFormatsAudioPCM struct{…}`

2717

2718 The PCM audio format. Only a 24kHz sample rate is supported.

2719

2720 - `Rate int64`

2721

2722 The sample rate of the audio. Always `24000`.

2723

2724 - `const RealtimeAudioFormatsAudioPCMRate24000 RealtimeAudioFormatsAudioPCMRate = 24000`

2725

2726 - `Type string`

2727

2728 The audio format. Always `audio/pcm`.

2729

2730 - `const RealtimeAudioFormatsAudioPCMTypeAudioPCM RealtimeAudioFormatsAudioPCMType = "audio/pcm"`

2731

2732 - `type RealtimeAudioFormatsAudioPCMU struct{…}`

2733

2734 The G.711 μ-law format.

2735

2736 - `Type string`

2737

2738 The audio format. Always `audio/pcmu`.

2739

2740 - `const RealtimeAudioFormatsAudioPCMUTypeAudioPCMU RealtimeAudioFormatsAudioPCMUType = "audio/pcmu"`

2741

2742 - `type RealtimeAudioFormatsAudioPCMA struct{…}`

2743

2744 The G.711 A-law format.

2745

2746 - `Type string`

2747

2748 The audio format. Always `audio/pcma`.

2749

2750 - `const RealtimeAudioFormatsAudioPCMATypeAudioPCMA RealtimeAudioFormatsAudioPCMAType = "audio/pcma"`

2751

2752 - `NoiseReduction RealtimeTranscriptionSessionCreateResponseAudioInputNoiseReduction`

2753

2754 Configuration for input audio noise reduction.

2755

2756 - `Type NoiseReductionType`

2757

2758 Type of noise reduction. `near_field` is for close-talking microphones such as headphones, `far_field` is for far-field microphones such as laptop or conference room microphones.

2759

2760 - `const NoiseReductionTypeNearField NoiseReductionType = "near_field"`

2761

2762 - `const NoiseReductionTypeFarField NoiseReductionType = "far_field"`

2763

2764 - `Transcription AudioTranscription`

2765

2766 - `Delay AudioTranscriptionDelay`

2767

2768 Controls how long the model waits before emitting transcription text.

2769 Higher values can improve transcription accuracy at the cost of latency.

2770 Only supported with `gpt-realtime-whisper` in GA Realtime sessions.

2771

2772 - `const AudioTranscriptionDelayMinimal AudioTranscriptionDelay = "minimal"`

2773

2774 - `const AudioTranscriptionDelayLow AudioTranscriptionDelay = "low"`

2775

2776 - `const AudioTranscriptionDelayMedium AudioTranscriptionDelay = "medium"`

2777

2778 - `const AudioTranscriptionDelayHigh AudioTranscriptionDelay = "high"`

2779

2780 - `const AudioTranscriptionDelayXhigh AudioTranscriptionDelay = "xhigh"`

2781

2782 - `Language string`

2783

2784 The language of the input audio. Supplying the input language in

2785 [ISO-639-1](https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) (e.g. `en`) format

2786 will improve accuracy and latency.

2787

2788 - `Model AudioTranscriptionModel`

2789

2790 The model to use for transcription. Current options are `whisper-1`, `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`, `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`. Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.

2791

2792 - `string`

2793

2794 - `type AudioTranscriptionModel string`

2795

2796 The model to use for transcription. Current options are `whisper-1`, `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`, `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`. Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.

2797

2798 - `const AudioTranscriptionModelWhisper1 AudioTranscriptionModel = "whisper-1"`

2799

2800 - `const AudioTranscriptionModelGPT4oMiniTranscribe AudioTranscriptionModel = "gpt-4o-mini-transcribe"`

2801

2802 - `const AudioTranscriptionModelGPT4oMiniTranscribe2025_12_15 AudioTranscriptionModel = "gpt-4o-mini-transcribe-2025-12-15"`

2803

2804 - `const AudioTranscriptionModelGPT4oTranscribe AudioTranscriptionModel = "gpt-4o-transcribe"`

2805

2806 - `const AudioTranscriptionModelGPT4oTranscribeDiarize AudioTranscriptionModel = "gpt-4o-transcribe-diarize"`

2807

2808 - `const AudioTranscriptionModelGPTRealtimeWhisper AudioTranscriptionModel = "gpt-realtime-whisper"`

2809

2810 - `Prompt string`

2811

2812 An optional text to guide the model's style or continue a previous audio

2813 segment.

2814 For `whisper-1`, the [prompt is a list of keywords](https://platform.openai.com/docs/guides/speech-to-text#prompting).

2815 For `gpt-4o-transcribe` models (excluding `gpt-4o-transcribe-diarize`), the prompt is a free text string, for example "expect words related to technology".

2816 Prompt is not supported with `gpt-realtime-whisper` in GA Realtime sessions.

2817

2818 - `TurnDetection RealtimeTranscriptionSessionTurnDetection`

2819

2820 Configuration for turn detection. Can be set to `null` to turn off. Server

2821 VAD means that the model will detect the start and end of speech based on

2822 audio volume and respond at the end of user speech. For `gpt-realtime-whisper`, this must be `null`; VAD is not supported.

2823

2824 - `PrefixPaddingMs int64`

2825

2826 Amount of audio to include before the VAD detected speech (in

2827 milliseconds). Defaults to 300ms.

2828

2829 - `SilenceDurationMs int64`

2830

2831 Duration of silence to detect speech stop (in milliseconds). Defaults

2832 to 500ms. With shorter values the model will respond more quickly,

2833 but may jump in on short pauses from the user.

2834

2835 - `Threshold float64`

2836

2837 Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A

2838 higher threshold will require louder audio to activate the model, and

2839 thus might perform better in noisy environments.

2840

2841 - `Type string`

2842

2843 Type of turn detection, only `server_vad` is currently supported.

2844

2845 - `ExpiresAt int64`

2846

2847 Expiration timestamp for the session, in seconds since epoch.

2848

2849 - `Include []string`

2850

2851 Additional fields to include in server outputs.

2852

2853 - `item.input_audio_transcription.logprobs`: Include logprobs for input audio transcription.

2854

2855 - `const RealtimeTranscriptionSessionCreateResponseIncludeItemInputAudioTranscriptionLogprobs RealtimeTranscriptionSessionCreateResponseInclude = "item.input_audio_transcription.logprobs"`

2856

2857### Realtime Transcription Session Turn Detection

2858

2859- `type RealtimeTranscriptionSessionTurnDetection struct{…}`

2860

2861 Configuration for turn detection. Can be set to `null` to turn off. Server

2862 VAD means that the model will detect the start and end of speech based on

2863 audio volume and respond at the end of user speech. For `gpt-realtime-whisper`, this must be `null`; VAD is not supported.

2864

2865 - `PrefixPaddingMs int64`

2866

2867 Amount of audio to include before the VAD detected speech (in

2868 milliseconds). Defaults to 300ms.

2869

2870 - `SilenceDurationMs int64`

2871

2872 Duration of silence to detect speech stop (in milliseconds). Defaults

2873 to 500ms. With shorter values the model will respond more quickly,

2874 but may jump in on short pauses from the user.

2875

2876 - `Threshold float64`

2877

2878 Activation threshold for VAD (0.0 to 1.0), this defaults to 0.5. A

2879 higher threshold will require louder audio to activate the model, and

2880 thus might perform better in noisy environments.

2881

2882 - `Type string`

2883

2884 Type of turn detection, only `server_vad` is currently supported.