Skip to main content
Two independent controls govern how much work your integration can do: a per-key request rate limit and your tenant’s credit balance. This page explains both and the behavior to expect when you reach either one.

​
Request rate limits

Each API key is rate limited to roughly 100 requests per second. The limit is applied per key, so scoping separate keys to separate workloads gives each workload its own budget. When you exceed the limit, requests are rejected with an HTTP 429 status and a JSON message. This is a temporary, expected signal — not an error in your integration. The correct response is to back off briefly and retry. 
{
  "error": "rate_limited",
  "message": "Too many requests. Retry after a short delay."
}
Implement exponential backoff with jitter on 429 responses: wait a short interval, then double it on each retry (for example 250 ms, 500 ms, 1 s) up to a sensible ceiling. Adding a small random jitter prevents many clients from retrying in lockstep. Most HTTP client libraries offer this as a built-in retry policy.

​
Credits

Every message your integration runs consumes credits from the tenant’s balance. Credits cover the model usage behind chats and agent runs. When a tenant’s balance is exhausted, AutoSage protects you from unexpected partial work with predictable behavior:
  • The next message is refused with a clear error rather than running with an incomplete result.
  • Scheduled agent runs are skipped while the balance is empty.
Topping up the tenant’s credits immediately restores normal operation — new messages succeed and scheduled runs resume on their next trigger. 
{
  "error": "insufficient_credits",
  "message": "This tenant has no remaining credits. Top up to resume."
}
Monitor your credit balance proactively instead of waiting for a refusal. Poll the billing endpoints on a schedule and alert your team when a tenant approaches a low-balance threshold, so top-ups happen before any message is refused or a scheduled run is skipped.
Rate limits and credits are independent. A request can succeed under the rate limit and still be refused if the tenant is out of credits, and vice versa. Handle both signals in your client.

​
Next steps

Billing API

Read balances and usage to monitor credits programmatically.

Best practices

Reliability, security, and multi-tenant patterns for production.