generated: '2026-08-02'
method: searched
source:
- https://docs.roboflow.com/reference/platform/rest-api.md
- https://docs.roboflow.com/reference/platform/rest-api/async-tasks.md
- https://docs.roboflow.com/deployment/production-checklist.md
- https://docs.roboflow.com/reference/errors-and-status-codes.md
authentication:
  style: workspace-scoped API key, or OAuth 2.1 bearer token
  placements:
  - query parameter api_key
  - JSON request body field api_key
  - 'Authorization: Bearer header'
  detail: authentication/roboflow-authentication.yml
idempotency:
  supported: false
  note: Roboflow documents no idempotency key header or parameter on any surface, and none appears in the published Inference
    Server OpenAPI. Write safety is handled by async task polling (each long-running write returns a task id) rather
    than request-level idempotency. Recorded as absent — no Idempotency pointer is emitted.
pagination:
  style: per-endpoint
  note: No single documented cross-API pagination contract. Model-evaluation image predictions are documented as paginated;
    the REST API reference documents pagination per endpoint rather than as a platform convention.
async_operations:
  pattern: task-id polling
  description: Long-running REST operations return a task ID; poll it for running / completed / failed status. Poll roughly
    every 5 seconds; processing may take up to 30 seconds to start.
  mcp_tool: async_tasks_get
  docs: https://docs.roboflow.com/reference/platform/rest-api/async-tasks.md
resource_addressing:
  style: hierarchical path
  hierarchy:
  - /:workspace
  - /:workspace/:project
  - /:workspace/:project/:version
  - /:workspace/:project/:version/:format
  - /:workspace/workflows/:workflow
  - /:workspace/groups
  - /:workspace/trash
hosts:
- host: https://api.roboflow.com
  role: Management — workspaces, projects, versions, training, workflows, datasets, vision events, trash
- host: https://serverless.roboflow.com
  role: Hosted inference — running a trained model or workflow against an image or video
- host: https://detect.roboflow.com
  role: Hosted Inference API (model predictions)
- host: https://roboflow.cloud
  role: Dedicated Deployments management
error_envelope:
  format: json
  top_level_field: error
  optional_field: hint
  rfc9457: false
  detail: errors/roboflow-error-codes.yml
retries:
  retryable:
  - 429
  - 5xx
  non_retryable:
  - 400
  - 401
  - 402
  - 403
  - 404
  - 423
  strategy: exponential backoff with jitter (1s, 2s, 4s, 8s + random offset), capped attempts
rate_limit_signaling:
  status: 429
  headers: null
  detail: rate-limits/roboflow-rate-limits.yml
cold_starts:
  note: Serverless loads models on demand; first call after idle (~10 min) pays a warmup cost. Set generous client timeouts
    and prime the model. Observe x-model-cold-start and x-processing-time headers.
versioning:
  scheme: unversioned path
  note: The REST API exposes no /v1-style version segment; the resource hierarchy is workspace/project/version where
    "version" is a dataset version, not an API version. Inference Server releases are versioned semantically (info.version
    in the OpenAPI).
  detail: lifecycle/roboflow-lifecycle.yml
agent_conventions:
  cli_json_mode: cli/roboflow-cli.yml
  mcp_server: mcp/roboflow-mcp.yml
  agent_skills: skills/_index.yml