Rate Limits

AIsa does not publish fixed request-per-minute (RPM), token-per-minute (TPM), or concurrency tiers, and responses do not carry X-RateLimit-* headers. A 429 Too Many Requests can come from one of the sources below. In every case, wait for Retry-After when it is present and back off before retrying.

Where a 429 comes from

Source	Applies to	How to recognize it
Per-model success limit	Only models that have a limit configured	`error.code = "model_success_rate_limit_exceeded"` with a `Retry-After` header
Upstream provider	Any model or data API	The provider throttled the request and no fallback attempt succeeded; the body follows the endpoint’s error format
Async capacity	Async data API jobs	`{"error": "integration async capacity is unavailable"}` with `Retry-After: 2`

Per-model success limit

Some models cap how many successful calls one account can make to that model within a rolling window of hours. Failed calls do not count toward the limit. Most models have no such limit. When the cap is reached, the gateway returns 429 with a Retry-After header (seconds) and this body:

{
  "error": {
    "message": "model <model-id> is limited to 100 successful calls per 1h0m0s for this user",
    "code": "model_success_rate_limit_exceeded",
    "model": "<model-id>",
    "limit": 100,
    "window_seconds": 3600,
    "retry_after_seconds": 812,
    "reset_at": "2026-10-14T09:30:00Z"
  }
}

The values above are illustrative. reset_at is when the oldest counted call leaves the window and a slot frees up.

Handling 429 responses

Honor Retry-After

When the header is present, wait at least that many seconds before the next attempt. Never retry immediately.

Use exponential backoff with jitter

Without Retry-After, start around 1 second, double the delay on each consecutive 429, add ±25% jitter, and cap the delay. See Retry guidance.

Switch models for success limits

model_success_rate_limit_exceeded is per model. If you cannot wait until reset_at, send the request to a different model.

Queue instead of looping

For batch workloads, queue requests and drain them at a steady rate rather than retrying in a tight loop.

Example: retry on 429

import time
from openai import OpenAI, RateLimitError

client = OpenAI(base_url="https://api.aisa.one/v1", api_key="sk-aisa-...")

def create_with_retry(max_attempts=5, **kwargs):
    delay = 1.0
    for attempt in range(max_attempts):
        try:
            return client.chat.completions.create(**kwargs)
        except RateLimitError as err:
            if attempt == max_attempts - 1:
                raise
            retry_after = err.response.headers.get("retry-after")
            time.sleep(float(retry_after) if retry_after else delay)
            delay = min(delay * 2, 30)

response = create_with_retry(
    model="qwen3.7-flash",
    messages=[{"role": "user", "content": "Hello"}],
)

Need more capacity?

If a per-model limit or upstream throttling blocks a production workload, contact developer@aisa.one with the model or endpoint, your expected peak request rate, and a short description of the use case.

Error Codes

Response formats and retry guidance.

Usage Logs

Per-request status and billing in the dashboard.

Overview

Get Started · For AI Agents

Get Started · For Developers

Essentials

Console

Help

Rate Limits

Where a 429 comes from

Per-model success limit

Handling 429 responses

Example: retry on 429

Need more capacity?

Error Codes

Usage Logs

​Where a 429 comes from

​Per-model success limit

​Handling 429 responses

​Example: retry on 429

​Need more capacity?

​Related

Error Codes

Usage Logs

Where a 429 comes from

Per-model success limit

Handling 429 responses

Example: retry on 429

Need more capacity?

Related