generated: '2026-07-21'
method: searched
source: https://docs.sarvam.ai/api/getting-started/quickstart
authentication:
  style: api-key-header
  header: api-subscription-key
  bearer_alternative: 'Authorization: Bearer <key>'
  ref: authentication/sarvam-authentication.yml
idempotency:
  supported: false
  notes: No idempotency-key header or documented idempotency contract as of the capture date.
pagination:
  supported: false
  notes: >-
    No cursor/offset pagination surface; list endpoints (e.g. pronunciation
    dictionary) return full collections. Large speech/document workloads use the
    async job model instead of pagination.
async_jobs:
  style: job-poll
  pattern: >-
    Long-running speech-to-text, speech-to-text-translate, and document
    digitization work uses a job lifecycle: initialise -> upload-files ->
    {job_id}/start -> poll {job_id}/status -> download-files.
  poll_defaults:
    batch_stt: 5s
    doc_digitization: 2s
  ref: openapi/sarvam-openapi-original.json
error_envelope:
  shape: '{ "error": { "message": "...", "code": "..." } }'
  auth_status: 403
  ref: errors/sarvam-error-codes.yml
rate_limiting:
  model: token-bucket
  enforcement: per-account (all API keys share one pool)
  granularity: per-API (REST, WebSocket, Vision, LLM have independent limits)
  statuses: [429, 503]
  ref: https://docs.sarvam.ai/api/getting-started/ratelimits
versioning:
  style: model-version-tags
  ref: lifecycle/sarvam-lifecycle.yml
streaming:
  transport: websocket
  channels: [/speech-to-text/ws, /speech-to-text-translate/ws, /text-to-speech/ws]
  ref: asyncapi/sarvam-streaming-asyncapi.yaml
sdk_env_var: SARVAM_API_KEY