generated: '2026-07-20'
method: searched
source: >-
  https://docs.modulate.ai/guides/authentication, https://docs.modulate.ai/quickstart,
  https://docs.modulate.ai/faq, and derived from openapi/*.yml + asyncapi/*.yml
authentication:
  style: api-key
  rest_header: X-API-Key
  websocket_param: api_key
  websocket_location: query
  docs: https://docs.modulate.ai/guides/authentication
  cross_ref: authentication/modulate-authentication.yml
transports:
  rest:
    style: multipart-upload
    description: >-
      Batch endpoints accept a multipart/form-data upload with an audio file
      (upload_file) and, for Velma, a JSON config blob. Synchronous — the full
      result is returned in one response.
    max_upload_bytes: 104857600
    timeout_seconds: 60
  websocket:
    scheme: wss
    host: platform.modulate.ai
    description: >-
      Streaming endpoints are WebSocket connections. Send audio chunks as binary
      frames; signal end-of-stream with an empty text frame. Server drains buffered
      audio, delivers outstanding results, sends a `done` message, then closes.
    end_of_stream_signal: empty text frame ("")
    partial_results: >-
      When partial_results=true, the server emits partial_utterance messages that are
      each superseded by the final utterance message.
pagination:
  supported: false
  note: Model endpoints are single-shot per audio file; no collection pagination.
idempotency:
  supported: false
  note: >-
    Audio-processing operations are stateless single-shot inferences with no documented
    idempotency-key contract.
versioning:
  scheme: model-name-in-path
  note: >-
    API versioning is expressed through the model name in the path (e.g. velma-2-*),
    not a global version segment or header. No dated-version or version header is documented.
error_envelope:
  format: json
  schema: ErrorResponse
  cross_ref: errors/modulate-error-codes.yml
rate_limit_signaling:
  documented_headers: false
  rest_statuses: [403, 429]
  websocket_close_code: 4029
  cross_ref: rate-limits/modulate-rate-limits.yml
data_retention:
  default: no audio stored by default
  stored_retention_days: 35
  note: >-
    Audio marked for storage is retained 35 days then permanently deleted. Enterprise
    customers can negotiate custom retention.