Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →A relay website can load its homepage normally while its AI API calls fail. The page and the API are often served through different routes, handlers, upstream services, or resource pools, so a working homepage tells you only that the page route is healthy. The API call can still return 400, 404, 429, or 503 because a different component produced that response. To fix it, diagnose the failing API request itself and identify which hop generated the status code.
Why a working homepage proves nothing about the API
A browser request to the site root typically reaches the web frontend. An API call may go to a different host, a different path prefix, a gateway, or an upstream model provider, and each of these can have its own routing rules, deployment, and capacity. A 200 response for the root URL confirms only that the root route answered.
As an Amazon Associate I earn from qualifying purchases.
Judge the API by a request to the API itself. Record the exact URL, method, headers, body, status, response body, and timestamp, then test that request on its own. Do not treat the homepage, or a site that loads in a browser, as evidence that the API works.
Recommended Free Tools
What each status code means for an API call
The four codes point to different causes, and they call for different fixes and different retry behavior. The definitions below follow MDN’s HTTP response status codes reference.
#1 Best Overall
400 Bad Request
The server decided the request was malformed or could not be processed as sent. The problem lies in the request, not in the account’s quota or the service’s health, so resending an identical request will usually fail the same way. Check the following:
- The method and path match the API’s current documentation, including any version segment.
- The Content-Type header matches the body. A JSON body sent without
application/json, or with a malformed payload, is a common cause. - The JSON parses and the field names and value types match what the endpoint expects.
- No proxy, relay layer, or client library has rewritten the headers or truncated the body.
404 Not Found
The requested resource was not found at the address given. In AI APIs, the path can be correct while the model name, deployment name, or resource identifier is wrong. The host or path prefix may also point to an older deployment that no longer contains the route. Some servers return 404 instead of 403 so that a client cannot discover a resource it is not allowed to access, which means a 404 on an identifier you believe is valid can also indicate an access problem. Check the host, the path prefix, the route mapping, the deployed version, and the resource identifier against the API’s current documentation.
429 Too Many Requests
The server’s rate-limiting policy has judged that the client sent too many requests within a period. Find out what the limit is scoped to. It may apply per key, per user, per IP address, or per model, and the scope is set by the provider, not the relay. Check the provider’s own quota documentation rather than assuming a number from elsewhere.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
As a reference point for the rate-limit concept only, Cloudflare’s Error 429 page, last updated August 14, 2026, documents a global Cloudflare API limit of 1,200 requests per five-minute period per user. That figure belongs to Cloudflare’s own API. It does not describe any AI API, and it should not be applied to one.
Rank #3
- If the response includes a
Retry-Afterheader, wait at least that long before retrying. - If the relay pools upstream capacity across many users, your requests can be throttled by traffic you cannot see. Reducing your own concurrency and request rate is the lever you control.
503 Service Unavailable
MDN’s 503 reference states: “The HTTP 503 Service Unavailable server error response status code indicates that the server is not ready to handle the request.” The condition is generally temporary. Maintenance and overload are the usual causes. A 503 can come from the origin service or from an intermediary such as a CDN, so the source matters for the fix. If the response carries a Retry-After header, it gives an estimated recovery time. Rate limiting aimed at a specific client belongs under 429, not 503.
Working through a failing API call
- Capture the evidence. Save the exact API URL, method, request body with secrets removed, relevant headers, response status, response body, any request or trace identifier the response includes, and the timestamp with its time zone.
- Send a minimal, known-valid request directly to the API host. If it succeeds, the failure lies in the original request’s content, path, or client configuration. If it fails the same way, the problem is in the service path.
- Branch by status. For 400, compare the request with the documented contract. For 404, verify the host, route, and resource identifier. For 429, check the provider’s quota and scope, then honor
Retry-After. For 503, determine where the response originated. - Retry only when the status and the evidence support it. Resending an unchanged 400 or 404 request will not help. A 429 or 503 may clear on its own, but a retry loop without backoff can extend the outage for everyone sharing the same capacity.
Telling origin errors from intermediary errors
A 502 or 503 can be produced by the application behind the relay, or by the CDN or proxy in front of it. Cloudflare’s documentation draws this distinction for its own network. Its Error 502 or 504 page, last updated July 20, 2026, asks reporters for a timestamp with time zone, the affected URL, and /cdn-cgi/trace output. Its Error 503 page, last updated April 23, 2026, treats the content of the error page as one signal for whether Cloudflare or the origin generated the response.
Quick Recap
- If the origin generated the 503, check its logs for overload, exhausted connection pools, and whether maintenance mode is active.
- If the intermediary generated the error, the origin may be healthy while the path to it is not, so origin logs alone will not explain the failure.
- These steps reflect Cloudflare’s own diagnostics. Other CDNs, gateways, and relay stacks expose different headers, error pages, and identifiers, so use the equivalents your stack provides.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
