Kensho Scribe Real Time API

asyncapi: 2.6.0
info:
  title: Kensho Scribe Real Time API
  version: 2.0.0
  description: |
    Real-time streaming transcription WebSocket API from Kensho Technologies
    (a wholly-owned S&P Global subsidiary). Companion to the Kensho Scribe v2
    batch REST API. Streams uncompressed PCM audio chunks to the server and
    receives partial and final transcripts back over the same WebSocket.

    This AsyncAPI document covers only the publicly documented Real Time
    WebSocket surface at wss://scribe.kensho.com/ws. S&P Global's other public
    surfaces (LLM-ready API, Extract, NERD, Scribe batch, Capital IQ Pro,
    Marketplace) are REST, file/SFTP, Snowflake share, or Databricks share and
    are not modeled here.
  contact:
    name: Kensho Technologies (S&P Global)
    url: https://kensho.com
    email: [email protected]
  license:
    name: Commercial - Kensho Terms of Service
    url: https://kensho.com/terms
  x-source:
    - https://docs.kensho.com/scribe/v2/developer-guide
    - https://docs.kensho.com/scribe/v2/real-time-api-specification
    - https://docs.kensho.com/authentication
servers:
  production:
    url: scribe.kensho.com/ws
    protocol: wss
    description: |
      Production Kensho Scribe Real Time WebSocket endpoint. Clients must
      complete OIDC/OAuth 2.0 authentication out-of-band (keypair or refresh
      token grant) to obtain the access token before opening the WebSocket,
      then send an Authenticate message as the first frame.
    security:
      - bearerToken: []
defaultContentType: application/json
channels:
  /:
    description: |
      Single bidirectional WebSocket channel. All client and server messages
      are JSON objects discriminated by the `message` field. The expected
      lifecycle is Authenticate -> StartTranscription -> AddData (repeating) ->
      EndOfStream, with the server streaming AddTranscript messages while data
      is processed and EndOfTranscript when complete. Audio must be sent in
      segments of at most 15 seconds.
    publish:
      operationId: sendClientMessage
      summary: Messages sent from client to Kensho Scribe Real Time
      message:
        oneOf:
          - $ref: '#/components/messages/Authenticate'
          - $ref: '#/components/messages/StartTranscription'
          - $ref: '#/components/messages/AddData'
          - $ref: '#/components/messages/EndOfStream'
    subscribe:
      operationId: receiveServerMessage
      summary: Messages sent from Kensho Scribe Real Time to client
      message:
        oneOf:
          - $ref: '#/components/messages/Authenticated'
          - $ref: '#/components/messages/TranscriptionStarted'
          - $ref: '#/components/messages/DataAdded'
          - $ref: '#/components/messages/AddTranscript'
          - $ref: '#/components/messages/EndOfTranscript'
          - $ref: '#/components/messages/Error'
components:
  securitySchemes:
    bearerToken:
      type: http
      scheme: bearer
      bearerFormat: JWT
      description: |
        Access token obtained via the Kensho OIDC/OAuth 2.0 flow (keypair grant
        for production or refresh-token grant for development). The token is
        delivered to the server inside the Authenticate message, not as an
        HTTP Authorization header.
  messages:
    Authenticate:
      name: Authenticate
      title: Authenticate
      summary: First client message; supplies the Kensho access token.
      payload:
        $ref: '#/components/schemas/AuthenticatePayload'
    Authenticated:
      name: Authenticated
      title: Authenticated
      summary: Server acknowledgement that the access token was accepted.
      payload:
        $ref: '#/components/schemas/AuthenticatedPayload'
    StartTranscription:
      name: StartTranscription
      title: StartTranscription
      summary: Configures the audio format and opens a transcription session.
      payload:
        $ref: '#/components/schemas/StartTranscriptionPayload'
    TranscriptionStarted:
      name: TranscriptionStarted
      title: TranscriptionStarted
      summary: Server confirmation that a transcription session is active.
      payload:
        $ref: '#/components/schemas/TranscriptionStartedPayload'
    AddData:
      name: AddData
      title: AddData
      summary: |
        Sends one chunk of base64-encoded PCM audio. Chunks must represent
        at most 15 seconds of audio and carry a monotonically increasing
        sequence_number.
      payload:
        $ref: '#/components/schemas/AddDataPayload'
    DataAdded:
      name: DataAdded
      title: DataAdded
      summary: Server acknowledgement that an audio chunk was accepted.
      payload:
        $ref: '#/components/schemas/DataAddedPayload'
    AddTranscript:
      name: AddTranscript
      title: AddTranscript
      summary: Server-pushed transcript segment for previously sent audio.
      payload:
        $ref: '#/components/schemas/AddTranscriptPayload'
    EndOfStream:
      name: EndOfStream
      title: EndOfStream
      summary: Client signal that no further audio will be sent.
      payload:
        $ref: '#/components/schemas/EndOfStreamPayload'
    EndOfTranscript:
      name: EndOfTranscript
      title: EndOfTranscript
      summary: Server signal that all transcripts for the stream have been emitted.
      payload:
        $ref: '#/components/schemas/EndOfTranscriptPayload'
    Error:
      name: Error
      title: Error
      summary: Server error response carrying a human-readable reason.
      payload:
        $ref: '#/components/schemas/ErrorPayload'
  schemas:
    AudioFormat:
      type: object
      description: |
        Audio format descriptor sent in StartTranscription. The publicly
        documented configuration is RAW PCM signed 16-bit little-endian at
        16 kHz mono.
      required:
        - type
        - encoding
        - sample_rate_hz
        - num_channels
      properties:
        type:
          type: string
          enum:
            - RAW
          description: Container type for the streamed audio.
        encoding:
          type: string
          enum:
            - pcm_s16le
          description: Signed 16-bit little-endian PCM.
        sample_rate_hz:
          type: integer
          enum:
            - 16000
          description: Sample rate in hertz. Documented value is 16000.
        num_channels:
          type: integer
          enum:
            - 1
          description: Number of audio channels. Documented value is 1 (mono).
    AuthenticatePayload:
      type: object
      required:
        - message
        - token
      properties:
        message:
          type: string
          enum:
            - Authenticate
        token:
          type: string
          description: Kensho OIDC/OAuth 2.0 access token.
    AuthenticatedPayload:
      type: object
      required:
        - message
      properties:
        message:
          type: string
          enum:
            - Authenticated
    StartTranscriptionPayload:
      type: object
      required:
        - message
        - audio_format
      properties:
        message:
          type: string
          enum:
            - StartTranscription
        audio_format:
          $ref: '#/components/schemas/AudioFormat'
    TranscriptionStartedPayload:
      type: object
      required:
        - message
      properties:
        message:
          type: string
          enum:
            - TranscriptionStarted
        request_id:
          type: string
          description: Unique identifier for the active transcription session.
    AddDataPayload:
      type: object
      required:
        - message
        - audio
        - sequence_number
      properties:
        message:
          type: string
          enum:
            - AddData
        audio:
          type: string
          format: byte
          description: |
            Base64-encoded PCM audio chunk. Must represent no more than 15
            seconds of audio at the configured sample rate.
        sequence_number:
          type: integer
          minimum: 0
          description: |
            Monotonically increasing chunk index starting at 0. Used by the
            server to detect dropped or out-of-order chunks.
    DataAddedPayload:
      type: object
      required:
        - message
      properties:
        message:
          type: string
          enum:
            - DataAdded
        sequence_number:
          type: integer
          minimum: 0
          description: Sequence number of the acknowledged chunk.
    AddTranscriptPayload:
      type: object
      required:
        - message
        - transcript
      properties:
        message:
          type: string
          enum:
            - AddTranscript
        transcript:
          type: object
          description: |
            Transcript segment object emitted by the server. The exact field
            set is not exhaustively documented in the public specification;
            consumers should treat unknown fields permissively.
    EndOfStreamPayload:
      type: object
      required:
        - message
        - last_sequence_number
      properties:
        message:
          type: string
          enum:
            - EndOfStream
        last_sequence_number:
          type: integer
          minimum: 0
          description: Sequence number of the final audio chunk sent.
    EndOfTranscriptPayload:
      type: object
      required:
        - message
      properties:
        message:
          type: string
          enum:
            - EndOfTranscript
    ErrorPayload:
      type: object
      required:
        - message
        - reason
      properties:
        message:
          type: string
          enum:
            - Error
        reason:
          type: string
          description: Human-readable description of the failure.