> ## Documentation Index
> Fetch the complete documentation index at: https://docs.prefect.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Rate limits and data retention

> Understand Prefect Cloud's API rate limits, request size limits, and data retention policies.

## API rate limits

<Note>
  **Rate limits are subject to change.**

  Contact Prefect support at [help@prefect.io](mailto:help@prefect.io) with questions about rate limits.
</Note>

Prefect Cloud applies rate limits at the account level to ensure system stability. When limits are exceeded, endpoints return a `429` response with a `Retry-After` header.

Rate limits are organized into two buckets. All API paths within a bucket share a single counter. There are no sub-limits per resource type.

### API Requests

Every HTTP call to the Prefect Cloud orchestration API counts as one request toward this bucket, including:

* Creating, reading, updating, and deleting **flows**, **flow runs**, **task runs**, **deployments**, **work pools**, **work queues**, **blocks**, **artifacts**, **variables**, and **concurrency limits**
* **Flow run and task run state transitions** (each `set_state` call counts as one request)
* **Reading events and logs** (filter and count queries)

<Note>
  Concurrency limit operations count toward this bucket. When a task uses a concurrency limit tag, the SDK makes API calls to acquire and release the concurrency slot. This means a task with a concurrency limit tag generates additional API requests beyond the task run state transitions themselves.
</Note>

### Logs + Events

Every log message and every event submitted to Prefect Cloud counts as one request toward this bucket, including:

* **Log writes**: Each individual log message counts as one request (not each HTTP batch). This includes all logs sent by flows, tasks, and Prefect infrastructure (such as worker and agent logs).
* **Event submissions**: Each event submitted through the `/events` HTTP endpoint counts as one request.
* **Task run events via WebSocket**: The Prefect SDK streams task run lifecycle events (such as `Running`, `Completed`, and `Failed` state changes) to Prefect Cloud over a WebSocket connection. Each event counts as one request toward this bucket. These events are emitted automatically by the SDK for every `@task` execution and are not manually created.

### Limits by plan

| Plan       | API Requests (per minute) | Logs + Events (per minute) |
| ---------- | ------------------------- | -------------------------- |
| Hobby      | 625                       | 2,000                      |
| Starter    | 1,250                     | 2,800                      |
| Team       | 2,500                     | 8,000                      |
| Pro        | 5,000                     | 40,000                     |
| Enterprise | Custom                    | Custom                     |

For Enterprise limits, [contact Prefect's sales team](https://www.prefect.io/pricing).

### Monitor your usage

View your account's rate limit usage in the Prefect Cloud UI under **Account Settings > Rate Limits**. The dashboard shows current usage against your limits and highlights the top contributors to request volume, such as specific API keys, workspaces, or API routes.

### Reduce request volume

If you're approaching your rate limits, consider the following:

* **Reduce flow run volume** by batching work into fewer runs or using task runs within a single flow instead of launching many small flows.
* **Optimize logging** by adjusting log levels or reducing the volume of log output from your flows.
* **Use webhooks and automations** instead of polling the API for state changes.
* **Upgrade your plan** if your workload consistently exceeds your current limits.

### SDK retry behavior

The Prefect SDK automatically retries rate-limited requests up to 5 times, using the delay specified in the `Retry-After` header. You can customize this behavior through [client settings](/v3/develop/settings-ref#clientsettings).

## Request size limits

Prefect Cloud enforces a **5 MiB** (5,242,880 byte) maximum on request body size for all API endpoints.
Requests that exceed this limit receive a `413 Content Too Large` response. The `Content-Length` header is
required on every `POST`, `PUT`, and `PATCH` request; omitting it returns a `411 Length Required` response.

This limit applies to the entire HTTP request body, which includes JSON encoding overhead. Common scenarios
where you may encounter this limit:

* **Large artifacts**: Table or Markdown artifacts with many rows or large cell values.
* **Bulk log writes**: Batches of log entries that together exceed 5 MiB.
* **Flow run inputs**: Large payloads passed through `flow.send_input()` or `send_input()`.

### Reduce request size

If your requests are approaching or exceeding the 5 MiB limit:

* **Paginate or chunk large artifacts** by splitting data across multiple smaller artifact calls instead of one large request.
* **Store large results externally** and save a reference (such as a URL or storage key) as the artifact value instead of the full dataset.
* **Reduce log volume per batch** by lowering the `PREFECT_LOGGING_TO_API_BATCH_SIZE` setting so each HTTP request carries fewer log entries.
* **Trim flow run input payloads** by passing references to external storage instead of large inline values.

## Query limits

Event queries are limited to a 14-day window per request. You can query any 14-day period within your retention period, but each individual query cannot span more than 14 days. If your plan's retention period is shorter than 14 days, the query window is limited to your retention period.

## Metadata retention

Prefect Cloud retains flow run, task run, and artifact metadata for a limited time based on your plan. This applies to all workspaces in your account.

### Retention periods by plan

* **Hobby and Starter**: 7 days
* **Team**: 14 days
* **Pro**: 30 days
* **Enterprise**: Custom retention periods

Metadata is retained from creation until the specified period expires. For flow and task runs, retention is calculated from when the run reaches a [terminal state](/v3/develop/manage-states/#state-types). Subflow runs are retained independently from their parent flows.

For custom retention periods, [contact Prefect's sales team](https://www.prefect.io/pricing).