Cloudflare Workers AI calls need admission through the shared free-ai-gateway budget before each provider attempt. Add the existing NEURON_BUDGET binding to this Worker and guard every Workers AI binding call with conservative UTF-8 input reservations, a bounded output-token limit, strict current-day receipts, and fail-closed behavior. Preserve BYOK and do not add provider fallback on budget denial. Validate retries reserve separately and malformed/exhausted budgets make zero inference calls.
Cloudflare Workers AI calls need admission through the shared free-ai-gateway budget before each provider attempt. Add the existing NEURON_BUDGET binding to this Worker and guard every Workers AI binding call with conservative UTF-8 input reservations, a bounded output-token limit, strict current-day receipts, and fail-closed behavior. Preserve BYOK and do not add provider fallback on budget denial. Validate retries reserve separately and malformed/exhausted budgets make zero inference calls.