Skip to main content
Use limits at the narrowest boundary that matches the risk.

​
Project limits

Project settings can define:
  • requests per minute
  • monthly budget in US dollars
A null value means no limit is configured for that field.

​
Key limits

Individual Relay keys can set their own requests-per-minute limit and monthly budget. Key settings can also override project routing configuration. Use key limits to isolate applications, environments, or teams that share a project. A stricter key limit should reduce exposure without changing unrelated keys.

​
Workload-specific budgets

Some Relay operations have separate spend controls:
  • shadow comparison has a monthly budget and sample rate
  • quality evaluations can stop with budget_exceeded
  • custom-router training accepts a per-run budget
  • plan quotas limit custom routers and route evaluations
These controls bound Relay-managed work. Provider-side account limits and invoices remain an independent backstop.

​
Operating practice

  1. Establish a measured baseline in Observe or with a low-volume gateway key.
  2. Set project limits for the whole workload.
  3. Add stricter key limits where applications have different risk.
  4. Assign explicit budgets to shadow, evaluations, and training.
  5. Monitor monthly spend and requests in Relay Usage.
  6. Test the behavior at each limit before relying on it in production.
A budget is an enforcement control, not a forecasting model. Use reports to estimate the effect of a routing change before changing the budget.

Troubleshoot limit failures

Distinguish Relay limits, plan quotas, provider errors, and payment requirements.