Skip to help content
AICD API
Planned per-plan limits

Usage limits

See what will count against a plan, the monthly limits planned for each plan, and when the allowance resets.

Current status

Paid access is not live. The counting rules and limits published here are the accepted policy for the paid release and take effect when checkout opens. They are not an offer to buy service today.

What counts

When paid access opens, one REST request or one MCP tool call counts once when it is admitted, including work that later fails. A request refused before admission, such as a failed authentication or a rejected limit, is not counted.

All keys on the account share one quota, and REST and MCP draw on the same admitted requests. A bounded REST batch is one admitted request, and the batch endpoint accepts at most 100 record IDs. A top-level MCP JSON-RPC batch array is not supported and is rejected.

A REST response body is capped at 1,000,000 bytes. An MCP request body is capped at 64 KiB, and an MCP tool result is capped at 16 KiB.

Recognized subscription keys on an account are also subject to an abuse limit of 1,000 requests per minute and 100 per second, including requests that are refused. Usable keys and revoked or expired keys are counted separately, so a compromised key you rotate or revoke does not use up your new key's capacity. This is separate from paid usage. If the limit is reached, retry after the time in Retry-After. If the limit cannot be checked, requests return 503.

Monthly limits by plan

Each plan has its own hard limits for one paid month. The allowance does not carry over to the next period, and it is not pooled across plans.

  • Builder: 10,000 admitted requests and 1 decimal GB of response bytes per paid month. 60 requests per minute, a burst of 5 per second, 2 concurrent requests, and up to 100 rows per response.
  • Pro: 50,000 admitted requests and 5 decimal GB of response bytes per paid month. 120 requests per minute, a burst of 10 per second, 4 concurrent requests, and up to 250 rows per response.
  • Ultra: 200,000 admitted requests and 20 decimal GB of response bytes per paid month. 300 requests per minute, a burst of 20 per second, 8 concurrent requests, and up to 500 rows per response.

When a limit is reached

  • Traffic over a plan limit returns 429 with a quota-specific error code and Retry-After. Wait for the number of seconds in Retry-After before you retry.
  • There is no automatic overage charge, and no plan exceeds its limit silently.
  • The allowance resets only in a new paid period. Ending the subscription does not reset it, and a refund or dispute revokes access without resetting the used allowance, so the account needs a new paid period.
  • A plan cannot be changed mid-period, and plan changes are not available at launch.

Recorded usage

The account shows the selected plan and the recorded usage for the selected billing period: requests used and response bytes used. Those figures come from the trusted billing record, so they report what was counted rather than what a later call may use.

When the usage record cannot be read, the account says the usage details are unavailable instead of showing zero.