> ## Documentation Index
> Fetch the complete documentation index at: https://support.i.moneyforward.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Rate limits

> The AI Helpdesk API does not currently enforce rate limits at the HTTP layer. Respect upstream DynamoDB capacity and keep concurrency below ~20 RPS per API key.

## Rate-Limit Policy

The AI Helpdesk API does not currently enforce rate limits at the HTTP layer. It does not return `X-RateLimit-*` headers, nor does it return `429 Too Many Requests`.

However, the upstream data store (DynamoDB) has finite capacity. Sending bulk requests can lead to upstream throttling and latency. **Keep concurrency below approximately `~20 RPS` per API key.**

## Recommended Practices

* For bulk reads, space requests at least \~50 ms apart
* If you parallelize requests, keep concurrency low (about 5–10 in-flight requests per key)
* If you observe latency that suggests upstream throttling, back off with exponential delays
* Cache responses on the client when the same data is fetched repeatedly in a short window

## Polling

When polling session lists or notifications, use the minimum frequency that meets the use case.

| Use case                   | Recommended polling interval |
| -------------------------- | ---------------------------- |
| Unhandled session count    | 30 seconds or longer         |
| Notification badge updates | 30 seconds or longer         |
| General dashboard refresh  | 60 seconds or longer         |

## Future Changes

HTTP-layer rate limiting may be introduced in the future. When that happens, both this page and response headers will surface the limits. We recommend building clients now under the assumption that server-side limits could be enforced later.