B7RDevelopers
Get startedLimits and retries
Get started / Limits and retries

Limits and retries

View .md

Calls per minute

Each account has a limit of calls per minute, shared by all its keys: more keys never raise it. The limit is an account setting, so read it from limits.calls_per_minute in GET /v1/me instead of writing it into your code. Separate caps per address and per key protect the API from floods.

Over a limit, the answer is 429 rate_limited with a Retry-After header in seconds. A 429 means nothing changed and nothing was charged: wait that long, then send the same request again.

Request size

PartCap
Path512 characters
Query string1024 characters
Body262144 bytes (256 KiB)

A body over its cap answers 413 request_too_large. A path or query over its cap is refused with a 4xx problem whose retryable is false.

The form of a request

The API reads requests in one exact form, so that nothing can be read two ways. HTTP clients send this form by default.

  • Paths exactly as in the reference: no trailing /, no //, no . or .. segments, no % encoding.
  • Query keys are lowercase letters, digits and _; values are letters, digits and ., _, ~, :, -. Each key once.
  • A GET has no body. A POST sends JSON with Content-Type: application/json (; charset=utf-8 is fine) and no Content-Encoding.
  • A path outside the API answers 404 not_found, a known path with the wrong method 405 method_not_allowed, another content type 415 unsupported_media_type, and anything else out of form 400 invalid_request.

When to retry

Every error carries retryable. Retry only when it is true: that is 429 and the 5xx codes. A 4xx with retryable: false fails the same way every time until you change the request.

  • Honour Retry-After when it is present.
  • Otherwise back off: about 1, 2, 4 and 8 seconds, each with some randomness, and stop after 5 attempts.
  • Retry a POST with the same Idempotency-Key and the same body. See Safe retries.
  • Set a client timeout of about 30 seconds. The API answers or gives up with 504 upstream_timeout before that.
Retry with backoff
async function callB7r(url, options = {}, attempts = 5) {
  for (let attempt = 1; ; attempt++) {
    const res = await fetch(url, { ...options, signal: AbortSignal.timeout(30_000) });
    if (res.ok) return res.json();
    const problem = await res.json();
    if (!problem.retryable || attempt === attempts) throw Object.assign(new Error(problem.title), { problem });
    const retryAfter = Number(res.headers.get("retry-after"));
    const backoff = 2 ** (attempt - 1) * 1000 * (0.5 + Math.random());
    await new Promise((r) => setTimeout(r, retryAfter > 0 ? retryAfter * 1000 : backoff));
  }
}
import random, time, requests

def call_b7r(method, url, attempts=5, **kwargs):
    for attempt in range(1, attempts + 1):
        res = requests.request(method, url, timeout=30, **kwargs)
        if res.ok:
            return res.json()
        problem = res.json()
        if not problem.get("retryable") or attempt == attempts:
            raise RuntimeError(f"{problem['code']}: {problem['title']}")
        retry_after = int(res.headers.get("Retry-After", "0") or 0)
        time.sleep(retry_after or 2 ** (attempt - 1) * (0.5 + random.random()))

Titles and headings of every page. llms-full.txt holds the full text.