Read rate-limit headroom from API response headers

Headers by provider

Capture them at your egress

The only way to see headroom is to read it off responses as they come back. A fetch wrapper does it in one place:

const watched = ["x-ratelimit-remaining-requests", "anthropic-ratelimit-requests-remaining",
  "x-ratelimit-remaining", "x-credits-remaining", "x-totallimit-remaining"];

const original = globalThis.fetch;
globalThis.fetch = async (input, init) => {
  const res = await original(input, init);
  const host = new URL(input instanceof Request ? input.url : String(input)).host;
  for (const h of watched) {
    const v = res.headers.get(h);
    if (v !== null) record(host, h, Number(v)); // your metrics sink
  }
  return res;
};

The Argus SDK is this wrapper with every provider's header names built in, and it never reads bodies or auth headers. Argus turns the readings into headroom per provider, raises rate_limit_low at 10% or less, and answers get_rate_limit_status for your coding agent over MCP.

For providers that send nothing

Count your own requests per window against the published limit on each integration page, and treat the 429 rate as the signal. A rising 429 rate is the warning those providers won't give you.

Questions

Which OpenAI headers show remaining rate limit?

x-ratelimit-remaining-requests and x-ratelimit-remaining-tokens, on every API response.

Which Anthropic headers show remaining rate limit?

anthropic-ratelimit-requests-remaining and anthropic-ratelimit-tokens-remaining, on every API response.

Does Gemini send rate-limit headers?

No. Gemini responses carry no rate-limit headers; limits are per model and tier and only visible in the console. You find out with a 429 RESOURCE_EXHAUSTED.

Related

OpenAI API usageAnthropic API usageOpenRouter API usagePeople Data Labs API usage