generated: '2026-07-20'
method: searched
source: https://docs.api.pruna.ai/guides/quickstart
notes: >-
  Cross-cutting request/response semantics for the Pruna P-API, from the docs
  quickstart and the OpenAPI. The API is HTTP+apikey with an async
  submit/poll/download flow and an optional synchronous mode.
authentication:
  style: api-key
  location: header
  parameter: apikey
  docs: https://docs.api.pruna.ai/guides/quickstart
  see: authentication/pruna-ai-authentication.yml
model_selection:
  header: Model
  description: >-
    The target model is selected per request via the `Model` request header
    (e.g. p-image, p-image-edit, p-video). One API surface, ~20 models.
sync_vs_async:
  default: asynchronous
  async_flow:
    - POST /v1/predictions returns a prediction id + get_url
    - GET /v1/predictions/status/{id} polls until succeeded/failed
    - GET /v1/predictions/delivery/{path} downloads the output
  sync_mode:
    header: 'Try-Sync: true'
    description: >-
      When set, POST /v1/predictions blocks up to 60 seconds and returns the
      completed result inline instead of requiring a poll.
idempotency:
  supported: false
  notes: >-
    No idempotency-key header is documented or present in the OpenAPI. Predictions
    are inherently non-idempotent (each POST enqueues a new generation).
pagination:
  supported: false
  notes: No list/collection endpoints; nothing to paginate.
file_uploads:
  endpoint: POST /v1/files
  encoding: multipart/form-data (content field)
  reference: >-
    Uploaded files return an id and a get URL that is passed into prediction
    inputs (images[]). Files carry expires_at.
error_envelope:
  media_type: application/json
  shape: '{ "error": { "code": string, "message": string, "details": string } }'
  see: errors/pruna-ai-problem-types.yml
rate_limiting:
  signalled_via: HTTP 429
  documented_limits:
    - scope: trainer models
      limit: 5 requests/minute
  see: rate-limits/pruna-ai-rate-limits.yml
versioning:
  scheme: uri-path
  current: v1
  see: lifecycle/pruna-ai-lifecycle.yml
content_delivery:
  notes: >-
    Generated assets are delivered from /v1/predictions/delivery/{path} with
    Cache-Control and Content-Disposition headers; content expires (HTTP 410
    when gone).