specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: CData
providerId: cdata
generated: '2026-09-05'
method: searched
source: >-
  https://www.cdata.com/ai/pricing/, https://www.cdata.com/developers/ai/developer-edition/,
  https://docs.cloud.cdata.com/en/Settings/Billing.md, and a full read of the API reference
  under https://docs.cloud.cdata.com/en/API/ plus all seven first-party OpenAPI definitions.
created: '2026-05-04'
modified: '2026-09-05'
supersedes: >-
  The 2026-05-04 bulk-sweep scaffold that previously occupied this file. It asserted
  X-RateLimit-Limit / X-RateLimit-Remaining / X-RateLimit-Reset / RateLimit-Policy response
  headers, 429 throttling, and per-minute limits of 10 / 100 / 1,000 by tier. None of that is
  published by CData. It was invented, and it is removed here.
limit_count: 3
headers: {}
headers_note: >-
  CData publishes NO rate-limit response headers. Searched every response object in all seven
  specs and the whole API reference for X-RateLimit-*, RateLimit-*, Retry-After and 429 — no
  match anywhere. An agent gets no runtime signal about its remaining budget from the wire;
  it has to read the Billing tab in the UI. This is the most consequential gap in CData's
  runtime contract for autonomous consumers.
response_codes:
  throttled: null
  quota_exceeded: null
  note: >-
    No 429 response is declared on any operation in any published specification. What happens
    when a plan's tool-call allowance is exhausted is not documented.
model:
  unit: tool calls
  window: month
  scope: account
  detail: >-
    Consumption is metered as monthly TOOL CALLS against the subscription, not as a request
    rate. There is no published per-second, per-minute or burst limit on any surface —
    neither REST, OData, MCP nor the Management API.
limits:
- tier: developer-edition
  name: Developer Edition monthly tool calls
  scope: account
  metric: tool_calls
  limit: 500
  timeFrame: month
  source: https://www.cdata.com/developers/ai/developer-edition/
  applies: [CData Connect AI MCP, CData Connect AI Management MCP]
- tier: developer-edition
  name: Developer Edition data sources
  scope: account
  metric: data_sources
  limit: 5
  timeFrame: null
  source: https://www.cdata.com/developers/ai/developer-edition/
- tier: standard-and-growth
  name: Connections per source
  scope: per-source
  metric: connections
  limit: 10
  timeFrame: null
  source: https://www.cdata.com/ai/pricing/
  detail: >-
    "Distinct query configurations within each connected system — tables, reports, or
    objects." Business tier is custom. A structural cap rather than a rate limit, but it is
    the only numeric ceiling the paid tiers publish.
undocumented:
  per_request_rate: true
  burst: true
  concurrency: true
  paid_tier_tool_call_allowance: true
  note: >-
    Standard and Growth publish no numeric tool-call allowance; Business pools calls with no
    figure. The account's remaining allowance is visible in the product's Billing tab
    ("The Additional Usage section displays what is remaining in your current plan") but is
    not published per tier and is not exposed through the API.
downstream_limits:
  note: >-
    A material second-order limit that CData does not control: every Connect AI query is
    executed against a third-party system, so the practical ceiling is often the SOURCE
    system's rate limit (Salesforce, NetSuite, Workday…), not CData's. The pricing page
    acknowledges this indirectly — Custom OAuth Apps let a customer authenticate with their
    own registered app so that connections "appear under your credentials with dedicated rate
    limits", i.e. the customer's own quota at the source rather than a shared one.
  mitigation:
  - Query cache (cache large datasets to a customer-provided Postgres database)
  - Metadata caching, with clearConnectionMetadata to invalidate it
  - Custom OAuth Apps for dedicated source-side rate limits
static_egress_ips:
  note: >-
    Not a rate limit, but the related operational fact an integrator needs: Connect AI calls
    out from a fixed set of addresses that must be allow-listed at the data source.
  cidrs: ['52.224.0.160/28', '4.154.117.160/28']
  source: https://docs.cloud.cdata.com/en/Settings/Siem-Endpoints.md
maintainers:
- FN: Kin Lane
  email: info@apievangelist.com