OpenAI 429 rate limit

Requests exceed the per-minute token or request limit of your tier. Retrying immediately makes it worse; retrying with jitter and spreading static prompt tokens into the provider cache usually clears it.

1. Send X-TokenZero-Fallback: auto: the gateway retries 429/5xx with exponential jitter and honours short retry-after values, and can fail over to a model and key you name.
2. Let the cache formatter mark the static prefix: cached input tokens count against limits at a lower rate on most tiers and cost up to 90% less.
3. Route simple prompts to the cheaper sibling (X-TokenZero-Cascade: auto) so the frontier model's limit is spent on the prompts that need it.

copycurl -s -X POST https://tokenzero.sh/v1/signup -d '{}'   # workspace + key, no form
npx -y @tokenzero/cli init                                # or: audits the repo, writes .env

copyOPENAI_BASE_URL=https://tokenzero.sh/w//v1
ANTHROPIC_BASE_URL=https://tokenzero.sh/w/
# provider keys stay in your code exactly as they are; nothing else changes

Does TokenZero add its own rate limit?
No per-request limit on the gateway path; public pages and workspace creation are limited per IP to prevent abuse.