generated: '2026-08-04'
method: searched
source: https://docs.dataloop.ai/docs/rest-api-connection
sources:
- https://docs.dataloop.ai/docs/rest-api-connection
- https://sdk-docs.dataloop.ai/en/latest/
- https://github.com/dataloop-ai/dtlpy/blob/master/dtlpy/entities/filters.py
- https://github.com/dataloop-ai/dtlpy/blob/master/dtlpy/services/api_client.py
notes: >-
  Dataloop publishes no public OpenAPI, so these cross-cutting semantics are taken
  from the provider's API documentation and from the first-party Python SDK (dtlpy),
  which is the authoritative client implementation of the platform contract.
base_url: https://gate.dataloop.ai/api/v1
authentication:
  style: bearer-jwt
  header: 'Authorization: Bearer <JWT>'
  token_sources: [API key (JWT), POST /token with user_credentials, Auth0 M2M client credentials]
  detail: authentication/dataloop-authentication.yml
idempotency:
  supported: false
  header: null
  note: >-
    No idempotency key header or parameter is documented anywhere in the Dataloop
    API docs or implemented in the dtlpy client. Retrying a create is not safe;
    duplicate entities are the expected outcome. No Idempotency pointer is emitted
    for this provider.
pagination:
  style: page-number
  carried_in: request body (DQL query document), not the query string
  request_params:
  - {name: page, description: Zero-based page index, default: 0}
  - {name: pageSize, description: Items per page}
  - {name: sort, description: Sort object, field to ascending/descending}
  - {name: resource, description: Resource the query targets (Item, Annotation, Execution, Task, ...)}
  defaults:
    page: 0
    page_size: 1000
    page_size_executions: 100
    page_size_pipeline_executions: 100
    page_size_dpk: 100
  maximums:
    page_size: 1000
    page_size_executions: 100
    page_size_pipeline_executions: 100
  behavior: >-
    The SDK clamps page_size to the per-resource maximum and logs a warning rather
    than erroring. Listing is exposed as a PagedEntities iterator in dtlpy.
  source: https://github.com/dataloop-ai/dtlpy/blob/master/dtlpy/entities/filters.py
query_language:
  name: DQL (Dataloop Query Language)
  description: >-
    JSON filter documents used to query items, annotations, executions, tasks and
    other resources, and to scope event triggers. Supports field/operator/value
    clauses, AND/OR joins, sorting and projection.
  applies_to: [list operations, event trigger scoping, dataset browsing]
  docs: https://docs.dataloop.ai/docs/triggers
field_expansion:
  supported: false
  note: No documented expand/include or sparse-fieldset mechanism.
metadata:
  supported: true
  description: >-
    Most entities (items, annotations, datasets, tasks) carry a free-form `metadata`
    object with a reserved `metadata.system` namespace managed by the platform and a
    `metadata.user` namespace for consumer data. DQL can filter on metadata paths.
request_tracing:
  request_id_header: null
  note: >-
    No request-id or correlation-id response header is documented. Platform-side
    traceability is provided through the Audit Logs feature in the UI, not through
    a per-response identifier a client can capture.
  audit_logs: https://docs.dataloop.ai/docs/audit-log
versioning:
  style: uri-path
  current: v1
  detail: lifecycle/dataloop-lifecycle.yml
error_envelope:
  content_type: application/json
  shape: '{"status": <int>, "message": "<string>"}'
  variant_observed: '{"message": "<string>"}'
  problem_json: false
  detail: errors/dataloop-problem-types.yml
rate_limiting:
  documented: false
  headers: null
  note: >-
    No published rate limits, no quota documentation, and no RateLimit/Retry-After
    header contract. The only published throughput ceilings are pagination page-size
    maximums. The SDK ships an aiohttp retry client, implying transient failures are
    expected, but the retry policy is client-side and not a documented server contract.
retries:
  client_side: true
  implementation: dtlpy RetryClient (aiohttp)
  error_suppression: >-
    dtlpy exposes global ErrorSuppression filters (by status code, message keyword,
    or path pattern) that downgrade selected platform errors to warnings.
  source: https://github.com/dataloop-ai/dtlpy/blob/master/dtlpy/services/api_client.py
transport:
  https_only: true
  note: Dataloop states it exclusively uses HTTPS for data transfer.
  source: https://docs.dataloop.ai/docs/data-privacy-security
cross_links:
  authentication: authentication/dataloop-authentication.yml
  errors: errors/dataloop-problem-types.yml
  lifecycle: lifecycle/dataloop-lifecycle.yml
  events: asyncapi/dataloop-platform-events.yml
  data_model: data-model/dataloop-data-model.yml
gap:
  summary: >-
    The three conventions an agent most needs are the three that are missing:
    no idempotency contract, no rate-limit signaling, and no request-id header.
    Combined with the absence of a public OpenAPI, an integrator has to read the
    Python SDK source to learn how the API actually behaves.