generated: '2026-09-06'
method: searched
source: https://docs.dremio.com/dremio-cloud/api/
docs:
- https://docs.dremio.com/dremio-cloud/api/
- https://docs.dremio.com/dremio-cloud/api/oauth-token
- https://docs.dremio.com/dremio-cloud/help-support/limits
- https://docs.dremio.com/dremio-cloud/sql/commands/rollback-table
- https://docs.dremio.com/dremio-cloud/sql/commands/vacuum-table
base_urls:
- name: Dremio API
  url: https://api.dremio.cloud/v0/
  note: US control plane. EU control plane is https://api.eu.dremio.cloud.
- name: Dremio Login / OAuth authorization server
  url: https://login.dremio.cloud/
- name: Iceberg Catalog REST
  url: https://catalog.dremio.cloud/api/iceberg/v1/
  note: Open Catalog exposed through the Apache Iceberg REST Catalog specification.
- name: Arrow Flight SQL
  url: grpc+tls://data.dremio.cloud:443
  note: Not REST — the JDBC/ODBC/ADBC data path.
- name: Self-managed REST v3
  url: https://{hostname}/api/v3
  note: Dremio Enterprise / Community; the hostname is the customer's coordinator.
authentication:
  style: bearer
  header: 'Authorization: Bearer <token>'
  token_types:
  - personal-access-token
  - oauth2-access-token
  exchange: >-
    Dremio recommends exchanging a PAT for an OAuth access token before production use.
  see: authentication/dremio-intelligent-lakehouse-platform-authentication.yml
idempotency:
  supported: true
  coverage: full
  mechanism: request-body parameter
  parameter: requestId
  location: body
  value: a client-generated UUID
  methods:
  - POST
  retention: 24 hours
  retention_note: An idempotency key can be reused after 24 hours in a new POST operation.
  behavior: >-
    The server uses requestId to identify retries of the same request and returns the same
    response without repeating the operation.
  docs: https://docs.dremio.com/dremio-cloud/api/
  coverage_rationale: >-
    Graded full rather than partial. The mechanism is documented against POST generally — not
    against a named subset of operations — and Dremio's docs state that PUT, GET and DELETE are
    naturally idempotent, so the whole mutating surface has replay protection. The caveat worth
    knowing is the shape: this is a body parameter, not an Idempotency-Key header, so it cannot
    be applied by an HTTP client or a proxy without touching the payload, and the docs say a
    successful POST for entities REQUIRES it.
concurrency:
  mechanism: optimistic
  field: tag
  description: >-
    Retrieving an entity returns a `tag` attribute holding a UUID version identifier. An update
    must include that tag, so a concurrent modification is rejected rather than silently
    overwriting.
pagination:
  styles:
  - style: cursor
    request_param: pageToken
    response_fields:
    - pageToken
    - nextPageToken
    response_field_note: >-
      The continuation token comes back under different names by endpoint — pageToken for Data
      Maintenance, nextPageToken for Folder and Reflection Summary. An agent cannot assume one.
    page_size_params:
    - maxChildren
    - maxResults
    endpoints:
    - Source
    - Folder
    - Data Maintenance
    - Reflection Summary
    - Search
    - Scripts
    - User Defined Functions
  - style: offset
    request_params:
    - limit
    - offset
    endpoints:
    - Job Results
    - Scripts
  consistency: >-
    Pagination style is NOT uniform across the API — cursor on catalog-shaped resources, offset
    on job results. This is the single biggest ergonomics wrinkle in the contract.
field_selection:
  include_exclude:
    params:
    - include
    - exclude
    endpoints:
    - Catalog
    - Source
    - Folder
    - User Defined Functions
  filtering:
    params:
    - filter
    - search
    - createdBy
    - ownedBy
    endpoints:
    - Data Maintenance
    - Reflection Summary
    - Search
    - Scripts
  sorting:
    param: orderBy
    endpoints:
    - Reflection Summary
    - Scripts
identifiers:
  scheme: uuid
  description: >-
    Projects, sources, datasets, roles and version tags are 128-bit UUIDs rendered as 36
    characters. There are no typed id prefixes.
timestamps:
  format: ISO 8601 UTC
  pattern: YYYY-MM-DDTHH:mm:ss.sssZ
  example: '2024-01-31T09:50:01.012Z'
versioning:
  scheme: uri-path
  current: v0
  note: >-
    Dremio Cloud is at /v0/; the self-managed product is at /api/v3. The two version strings
    belong to different products, not to a migration path between them.
  see: lifecycle/dremio-intelligent-lakehouse-platform-lifecycle.yml
errors:
  envelope: json
  documented_status_codes:
  - 200
  - 400
  - 401
  - 403
  - 404
  note: >-
    Dremio documents response status codes per endpoint rather than publishing a central error
    registry, and does not use application/problem+json (RFC 9457).
  see: errors/dremio-intelligent-lakehouse-platform-problem-types.yml
rate_limit_signaling:
  documented_headers: none
  note: >-
    Dremio publishes hard numeric limits (1,200 API calls/minute and per-endpoint ceilings) but
    documents no RateLimit-*/X-RateLimit-*/Retry-After response headers and no exhaustion status
    code. An agent has to know the limits ahead of time; it cannot read remaining budget off a
    response.
  see: rate-limits/dremio-intelligent-lakehouse-platform-rate-limits.yml
dry_run_mode:
  supported: true
  surface: cli
  flag: --dry-run
  description: >-
    The Dremio Developer CLI accepts --dry-run to validate a destructive command before
    executing it; the docs demonstrate it on `dremio reflection refresh`.
  rest_equivalent: none-documented
  docs: https://docs.dremio.com/dremio-cloud/ai-integration/cli/
reversibility:
  grade: verified
  summary: >-
    Dremio's strongest reversal guarantee is at the data layer, not the control layer. Table
    writes are Apache Iceberg snapshots and can be rolled back to any retained snapshot; the
    retention window is stated. Control-plane deletes (catalog entities, Reflections, tokens)
    have no documented undo.
  surfaces:
  - surface: table data writes (INSERT / UPDATE / DELETE / MERGE / CTAS)
    reversal: ROLLBACK TABLE <table> TO { SNAPSHOT '<id>' | TIMESTAMP '<ts>' }
    operation_id: null
    window: >-
      As long as the target snapshot still exists. Snapshots are expired by VACUUM TABLE EXPIRE
      SNAPSHOTS, whose older_than default is five days before the command runs — so the
      documented default retention floor is five days, extendable with older_than/retain_last.
    window_source: https://docs.dremio.com/dremio-cloud/sql/commands/vacuum-table
    docs: https://docs.dremio.com/dremio-cloud/sql/commands/rollback-table
    privileges: ADMIN role, table owner, or INSERT/UPDATE/DELETE on the table
  - surface: running query / job
    reversal: cancel
    operation_id: cancelJob
    window: while the job is running
    docs: https://docs.dremio.com/dremio-cloud/api/job/
  - surface: catalog entity delete
    reversal: none-documented
    operation_id: deleteCatalogEntity
    window: null
    note: >-
      No undelete or restore is documented for a deleted catalog entity in Dremio Cloud. The
      self-managed product documents an Open Catalog backup/restore procedure, which is an
      administrator recovery path, not an API-level undo.
  - surface: reflection delete
    reversal: recreate
    operation_id: deleteReflection
    window: null
    note: A deleted Reflection must be recreated; there is no restore.
  caution: >-
    No window here was inferred. The five-day figure is the documented default of VACUUM TABLE's
    older_than parameter and nothing else; a deployment that runs VACUUM with a shorter
    older_than has a shorter window.
metadata:
  wikis_and_labels: >-
    Datasets carry wikis (rich text/Markdown) and labels, both writable through the catalog API.
  see: https://docs.dremio.com/dremio-cloud/manage-govern/wikis-labels
request_tracing:
  request_id_header: none-documented
  note: >-
    requestId is an idempotency key in the request body, not a correlation id echoed in a
    response header. Jobs are traced by job id instead.