Limits and cost controls

Run admission applies an HTTP envelope limit, versioned JSON request bounds, per-minute rate limiting, and an optional organization monthly run cap.

Request bounds

The released engine configuration may lower or raise the effective JSON bounds within the platform ceilings. A zero or omitted configuration value inherits the platform default.

Bound	Platform default	Maximum configurable ceiling	Failure status
HTTP request envelope	1,048,576 bytes	Fixed	`413`
Canonical JSON bytes for `input` plus `offerings`	262,144 bytes	1,048,576 bytes	`413`
Unicode characters in one string	16,384	65,536	`400`
Items in one array	100	1,000	`400`
Fields in one object	100	1,000	`400`
Object or array depth	12	32	`400`

The first deterministic violation is returned before run creation, cap reservation, queue insertion, or provider access.

{
  "code": "request_bound_exceeded",
  "message": "Request exceeds max_array_items",
  "details": {
    "bound": "max_array_items",
    "path": "$.input.items",
    "limit": 100,
    "actual": 101
  }
}

The details object always contains bound, path, limit, and actual.

Requests per minute

Engine-scoped routes use the engine’s configured RPM when it is non-zero. Other public routes, and engines without an override, use the organization default. A rejected request returns:

{
  "code": "rate_limited",
  "message": "Rate limit exceeded"
}

The response is 429 and includes Retry-After in seconds. If the rate-limit backend is unavailable, request processing currently continues and the degraded enforcement interval is recorded operationally. The monthly cap below remains the atomic admission backstop when configured.

Monthly run cap

Every successfully created run reserves one unit in the current UTC calendar month. Live, sandbox, and deterministic fixture runs use the same organization cap. Concurrent admission performs the cap check and reservation in the run creation transaction. At the configured limit, the API creates no run and returns 429:

{
  "code": "run_cap_exceeded",
  "message": "Monthly run cap reached for this organization. Contact support to raise the limit."
}

Existing organizations without a configured cap are uncapped. Partner credentials are issued only after an explicit launch cap and RPM are set.

Provider and output declarations

Released configs can declare max_provider_calls up to 100 and max_output_tokens up to 1,000,000. The configuration validator accepts zero as not yet enforced. These declarations do not replace the admission controls above, and a launch contract must state the actual configured values before credentials are issued.

Escalation

To request a limit change, contact support@usenexio.com with the environment slug, engine slug, expected peak RPM, expected monthly run count, and the relevant run_id or error timestamp. Do not retry a monthly-cap error in a tight loop.

Start Here

Guides

API Reference

Limits and cost controls

Request bounds

Requests per minute

Monthly run cap

Provider and output declarations

Escalation

​Request bounds

​Requests per minute

​Monthly run cap

​Provider and output declarations

​Escalation

Request bounds

Requests per minute

Monthly run cap

Provider and output declarations

Escalation