Send LLM chat requests through Cloudflare AI Gateway's OpenAI-compatible endpoint with caching, retries, and gateway authentication

Documented steps

1. Read the current docs first: https://developers.cloudflare.com/ai-gateway/get-started/ and https://developers.cloudflare.com/ai-gateway/usage/chat-completion/ . The older Universal Endpoint page (https://developers.cloudflare.com/ai-gateway/usage/universal/) now carries a deprecation banner: 'The Universal Endpoint is deprecated. Use the OpenAI-compatible endpoint for new integrations, and Dynamic Routing for fallbacks, retries, and conditional routing.'
2. Create an API token with 'AI Gateway - Read' and 'AI Gateway - Edit' permissions (add 'Workers AI - Read' if you will call Workers AI models through the gateway).
3. Create a gateway in the dashboard: AI > AI Gateway > Create Gateway. Gateway names are capped at 64 characters. A default gateway is auto-created on the first authenticated request to a third-party provider.
4. Multi-provider calls go to the OpenAI-compatible endpoint: POST https://gateway.ai.cloudflare.com/v1/{account_id}/{gateway_id}/compat/chat/completions with the model expressed as '{provider}/{model}', e.g. 'openai/gpt-5.2' or 'anthropic/claude-sonnet-4-5'.
5. Example: curl -X POST https://gateway.ai.cloudflare.com/v1/{account_id}/{gateway_id}/compat/chat/completions -H 'cf-aig-authorization: Bearer {CF_AIG_TOKEN}' -H 'Content-Type: application/json' -d '{"model":"openai/gpt-5.2","messages":[{"role":"user","content":"Hello"}]}'
6. For a single provider in its native wire format, use the provider-specific path instead: https://gateway.ai.cloudflare.com/v1/{account_id}/{gateway_id}/{provider} (e.g. .../openai/chat/completions, .../openai/responses).
7. For Workers AI models you may also use the account-scoped OpenAI-compatible path: POST https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1/chat/completions with 'Authorization: Bearer {CF_API_TOKEN}' and header 'cf-aig-gateway-id: {gateway_id}'.
8. Turn on Authenticated Gateway in gateway Settings, mint an authentication token with 'Run' permission, and send it as 'cf-aig-authorization: Bearer {token}' on every gateway.ai.cloudflare.com request.
9. Enable 'Cache Responses' in gateway Settings. Override per request with 'cf-aig-skip-cache: true', 'cf-aig-cache-ttl: <seconds>', and 'cf-aig-cache-key: <key>'; read the 'cf-aig-cache-status' response header to confirm HIT or MISS. Reference: https://developers.cloudflare.com/ai-gateway/features/caching/
10. Set a per-request deadline with 'cf-aig-request-timeout' (milliseconds). For fallbacks and retries, configure Dynamic Routing (gateway > Dynamic Routes > Add Route, or the JSON config) and attach a 'retries' property to model nodes: https://developers.cloudflare.com/ai-gateway/features/dynamic-routing/
11. Optionally enable rate limiting in gateway Settings (limit, interval, fixed or sliding window); requests over the limit receive HTTP 429.