Skip to content
Appmake
Get started
Kalle API docs

Rate limits

Rate limits keep the Kalle API fast for everyone. Each workspace has limits per model, on both requests and tokens per minute.

How limits work

  • Limits apply per workspace, so all keys in a workspace share them.
  • Each model has its own limits. Busy traffic on Kalle Turbo does not slow down Kalle Turbo Ultra.
  • Requests and tokens are counted separately. You can hit the token limit with a few large requests.
  • Your current limits are shown in Kalle API in the Appmake console.

Headers

Every response tells you where you stand:

HeaderMeaning
x-ratelimit-limit-requestsRequests allowed per minute for this model.
x-ratelimit-remaining-requestsRequests left in the current window.
x-ratelimit-reset-requestsWhen the request window resets, as an ISO 8601 time.
x-ratelimit-limit-tokensTokens allowed per minute for this model.
x-ratelimit-remaining-tokensTokens left in the current window.
x-ratelimit-reset-tokensWhen the token window resets, as an ISO 8601 time.
retry-afterOnly on 429 responses: the seconds to wait before you try again.

When you hit a limit

Kalle answers with status 429 and an error of the type rate_limit_error. Wait for the number of seconds in retry-after, then try again. Without the header, back off exponentially with some randomness, so many clients do not retry at the same moment:

TypeScript
async function withRetry(send: () => Promise<Response>, attempts = 5): Promise<Response> {
  for (let attempt = 0; ; attempt++) {
    const response = await send();
    if (response.status !== 429 || attempt === attempts - 1) return response;
    const retryAfter = Number(response.headers.get("retry-after"));
    const wait = retryAfter > 0 ? retryAfter * 1000 : 2 ** attempt * 500 + Math.random() * 250;
    await new Promise((resolve) => setTimeout(resolve, wait));
  }
}

Staying under your limits

  • Watch the remaining headers and slow down before you reach zero.
  • Set max_tokens close to what you need. The token limit counts it before the answer is written.
  • Use Kalle Turbo for high-volume, simple work and save the bigger models for the requests that need them.
  • Queue background jobs and run them at a steady pace instead of all at once.