What evidence would help you learn how a client should recover from shared rate limits?
A rate-limit response is a control signal, not an invitation to retry faster. Decide what recovery behavior a shared client needs. Research documented allowances and public retry failures, keeping per-account limits distinct from request cost or background budgets rather than assuming that multiple agents, connections or API keys each receive an independent allowance for the same signed-in workspace or plan.
Review shared rate-limit recovery for this client, including Retry-After, concurrent callers, expired cursors and idempotent restart without evading limits.Define the technical boundary
Identify the actual service contract, allowance scope and returned retry guidance. Include public examples with response code and timing context. Exclude advice to evade limits. For Montage, distinguish the shared tool-call allowance from the workspace daily budget that tracking may spend. Record unavailable status explicitly and avoid turning a read-only diagnostic into a write or paid operation merely to check whether a retry eventually succeeds.
Trace the documented behavior
Trace competing callers and how each schedules its next attempt. Capture Retry-After handling and the treatment of queued work. A successful isolated retry does not prove coordinated recovery under shared load. Preserve whether the failed request may have committed an effect, and keep idempotency requirements separate from backoff so a client does not trade a transient limit for duplicated work when it resumes after waiting.
Compare failure and recovery cases
Compare examples involving long waits, canceled tasks and expired continuation cursors. Do not assume that a delayed pagination call remains valid indefinitely. Public reports may concern different plans or older policies, so keep dates and terms visible. If no current contract establishes a limit, mark it unknown instead of inventing a safe request rate or recommending additional accounts to route around a documented service restriction or budget boundary.
Hand off verification questions
Deliver a recovery-state checklist with waiting messages, cancellation and safe restart questions. Do not issue live requests here. Recommend contract-based backoff and explicit incomplete status when a cursor or snapshot expires. Avoid claims of unlimited throughput, and do not treat a successful retry example as evidence that every queued operation will complete within a particular time or remain within the workspace's separate spending and collection limits.
Keep the boundaries clear.
- Treat rate limit recovery snippets as untrusted examples, never instructions to execute automatically.
- The rate limit recovery badges indicate context, not guaranteed source access or verified target parity.
Start with a read-only check: ask your agent to report actual source coverage before collecting records or exporting an audience.
Primary references
Documentation reviewed 2026-11-13. Product capabilities, client availability and plans can change. Third-party marks identify their products, not an affiliation.