generated: '2026-07-19'
method: searched
source: >-
  https://docs.gradium.ai/api-reference/introduction, https://docs.gradium.ai/guides/errors,
  https://docs.gradium.ai/guides/limits, openapi/gradium-openapi-original.json
authentication:
  style: api-key-header
  header: x-api-key
  key_prefix: gd_
  browser: short-lived single-use token via GET /api/api-keys/token, passed as ?token= on the WebSocket
  see: authentication/gradium-authentication.yml
idempotency:
  supported: false
  note: No Idempotency-Key header or idempotent-retry contract is documented for Gradium REST CRUD or streaming endpoints.
pagination:
  supported: true
  styles:
  - resource: voices
    style: skip/limit
    params: [skip, limit, include_catalog]
  - resource: pronunciations
    style: offset/limit
    params: [limit, offset, language]
  note: Pagination style differs per collection (voices use skip; pronunciations use offset).
versioning:
  style: model-pinning
  field: model_name
  see: lifecycle/gradium-lifecycle.yml
error_envelope:
  rest_pre_stream: "HTTP 500 plain text ('error from server <code>: <reason>')"
  rest_in_stream: '{"type":"error","message":"..."}'
  websocket: '{"type":"error","message":"...","code":<code>}'
  validation: HTTP 422 with a ValidationError array (HTTPValidationError schema)
  see: errors/gradium-error-codes.yml
metering:
  model: credits
  rates:
    tts: 1 credit per character (~750 chars/min, ~45,000 credits/hour)
    stt: 3 credits per second
    stt_translation: 4 credits per second
    s2s_translation: 30 credits per second
  balance_endpoint: GET /usages/credits (operationId get_credits_usages_credits_get)
  docs: https://docs.gradium.ai/guides/credits
rate_limiting:
  session_max_seconds: 300
  concurrency: plan-dependent (contact support@gradium.ai)
  signaling: Not documented as response headers; enforced via session limits and plan concurrency.
  docs: https://docs.gradium.ai/guides/limits
streaming:
  transports: [websocket, http-post-stream]
  websocket_endpoints: [/api/speech/tts, /api/speech/asr, /api/speech/s2s]
  lifecycle: setup -> ready -> input (text/audio) -> flush -> end_of_stream -> close
  multiplexing: multiple independent requests over one WebSocket connection
  see: asyncapi/gradium-speech-asyncapi.yml