Inworld Realtime API

openapi: 3.1.0
info:
  title: Inworld Realtime API
  description: >
    End-to-end speech-to-speech voice pipeline (STT + LLM + TTS). Connect over
    WebSocket or WebRTC. Event protocol is OpenAI-Realtime-API-compatible, so
    existing OpenAI Realtime clients can swap the base URL. Includes server-side
    and semantic VAD, function/tool calling, MCP server tunneling, Twilio media
    streams, and JWT-based session authentication.
  version: v1
  contact:
    name: Inworld Support
    url: https://docs.inworld.ai/tts/resources/support
  license:
    name: Inworld Terms of Service
    url: https://inworld.ai/legal/terms-of-service
servers:
  - url: https://api.inworld.ai
    description: Inworld Production API
security:
  - BearerAuth: []
tags:
  - name: Realtime
    description: Realtime speech-to-speech sessions.
paths:
  /api/v1/realtime/session:
    get:
      summary: Realtime WebSocket Session
      description: >
        Upgrade to a Realtime WebSocket session. The connection accepts a JSON
        event protocol compatible with the OpenAI Realtime API
        (`session.update`, `input_audio_buffer.append`,
        `input_audio_buffer.commit`, `response.create`,
        `conversation.item.create`, `response.cancel`, etc.). Authenticate with
        a short-lived JWT in the `Authorization: Bearer <jwt>` header, minted
        with a Realtime-only API key.
      operationId: realtimeWebsocket
      tags: [Realtime]
      responses:
        '101':
          description: Switching protocols to WebSocket.
  /v1/realtime/webrtc:
    post:
      summary: Realtime WebRTC Offer Exchange
      description: >
        Exchange a WebRTC SDP offer for an answer to open a realtime audio session.
        The server provisions an audio track wired to the same STT-LLM-TTS pipeline
        used by the WebSocket endpoint.
      operationId: realtimeWebrtc
      tags: [Realtime]
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/WebRtcOffer'
      responses:
        '200':
          description: WebRTC answer returned.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/WebRtcAnswer'
        '4XX':
          $ref: '#/components/responses/ErrorResponse'
components:
  securitySchemes:
    BearerAuth:
      type: http
      scheme: bearer
      bearerFormat: JWT
  schemas:
    WebRtcOffer:
      type: object
      required: [sdp]
      properties:
        sdp:
          type: string
          description: SDP offer string.
        sessionConfig:
          $ref: '#/components/schemas/SessionConfig'
    WebRtcAnswer:
      type: object
      properties:
        sdp:
          type: string
        sessionId:
          type: string
    SessionConfig:
      type: object
      properties:
        model:
          type: string
          description: Realtime model id.
        instructions:
          type: string
          description: System prompt for the agent.
        temperature:
          type: number
        maxTokens:
          type: integer
        voice:
          type: string
          description: Voice id used by the TTS leg.
        inputAudioFormat:
          type: string
          enum: [pcm16, pcmu, pcma]
        outputAudioFormat:
          type: string
          enum: [pcm16, pcmu, pcma]
        turnDetection:
          type: object
          properties:
            type:
              type: string
              enum: [server_vad, semantic_vad, none]
            threshold:
              type: number
            silenceDurationMs:
              type: integer
        tools:
          type: array
          items:
            type: object
        toolChoice:
          oneOf:
            - type: string
            - type: object
        mcpServers:
          type: array
          items:
            type: object
            properties:
              url:
                type: string
              transport:
                type: string
    RealtimeEvent:
      type: object
      description: >
        Base envelope for every WebSocket event. Client-to-server types include
        `session.update`, `conversation.item.create`, `response.create`,
        `input_audio_buffer.append`, `input_audio_buffer.commit`,
        `response.cancel`, `conversation.item.delete`,
        `conversation.item.truncate`, `conversation.item.retrieve`. Server-to-client
        types include `session.updated`, `conversation.item.added`,
        `response.audio.delta`, `response.audio.done`,
        `response.text.delta`, `response.function_call_arguments.delta`,
        `input_audio_buffer.speech_started`,
        `input_audio_buffer.speech_stopped`, `error`.
      required: [type]
      properties:
        event_id:
          type: string
        type:
          type: string
    Error:
      type: object
      properties:
        code:
          type: integer
        message:
          type: string
  responses:
    ErrorResponse:
      description: Error response.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'