Waxell · Rate Limits

Waxell Rate Limits

Waxell exposes two related rate-limiting surfaces. The Observe REST API enforces tenant-scoped request limits (per `wax_sk_` API key) and returns HTTP 429 when exceeded. Independently, the Waxell governance engine itself emits "throttle" decisions through the policy_check endpoint to slow down or block agents under runtime cost or rate-limit policies — agents see this as a `PolicyViolationError`. Numeric ceilings are not publicly documented.

3 Limits Throttle: 429 Quota: 429
Rate LimitingAI Agent GovernanceObservability

Limits

Observe API requests (per API key) key
varies
see /docs/observe/api/rate-limits (gated)
Tenant-scoped throttle attached to the wax_sk_ key; numeric ceilings not published.
Policy check throttling account
varies
governed by configured Rate Limit / Cost Management policies
Governance policies can return a `throttle` decision that is enforced inside the SDK.
Runtime cost-budget throttle account
varies
configurable budget, raises PolicyViolationError when exceeded
Cost policies block or throttle runs when configured token / dollar budgets are exceeded.

Policies

PolicyViolationError on throttle/block
Throttle and block decisions raise `PolicyViolationError` inside the Waxell SDK; callers should catch and back off.
Decision actions
The policy engine returns one of seven actions — allow, warn, redact, throttle, block, skip, retry — only the latter three are throttle-equivalent.
Mid-execution governance
With `mid_execution_governance=True`, policies are re-evaluated between agent steps so long-running agents can be throttled mid-run.
Kill switch
Operators can flip a kill switch policy that immediately blocks all runs for an agent or tenant.
Retry guidance
Standard exponential backoff is recommended on HTTP 429; `retry_after_seconds` may be returned on policy decisions.

Sources