# Event Backups & Data Retention

:::info
Event Backups is a feature of [Jitsu Cloud](https://use.jitsu.com). Backups are kept for up to 7 days on the Free
plan and up to 90 days on paid plans. [Self-hosted](/docs/self-hosting/) Jitsu doesn't archive events — configure a
warehouse or file-storage destination if you need a raw copy of incoming events.
:::

## What are Event Backups?

Jitsu keeps a raw copy of every event it accepts in a dedicated Google Cloud Storage bucket, separate from your
destinations. If a destination or warehouse fails, drops a table, or silently loses data, the backup is the copy
Jitsu can replay events from. Backups are written as compressed newline-delimited JSON, one file per batch, grouped by
date.

Backups are governed by a **retention window**: backups older than the window are deleted automatically. The window
is set per workspace in **Settings → Data Retention & Backups**. Workspaces on the Free plan keep **7 days** of
backups by default; paid plans default to **90 days**.

## Choosing a retention window

| Window | Availability | When it fits |
| --- | --- | --- |
| **No backups** | All plans | You keep your own raw copy (e.g. a file-storage destination) or must not retain data outside your destinations |
| **7 days** | All plans | Enough to replay a short destination outage |
| **30 days** | Paid plans | Covers a month of destination or warehouse issues |
| **90 days** | Paid plans | Maximum self-serve window — replay a full quarter |

Longer or custom windows are available on Enterprise plans — [contact us](https://jitsu.com/contact).

Changes are applied to the backup bucket within an hour. Shortening the window applies to existing backups too:
anything older than the new window is deleted and cannot be restored.

:::caution
Setting the window to **No backups** stops archiving and deletes existing backups. From then on, Jitsu keeps no copy
of your events beyond the pipeline stages listed below: if a destination or warehouse fails or loses data, those
events cannot be recovered. The console asks for an explicit acknowledgement before saving this setting.
:::

## How long Jitsu keeps your data

Event data passes through several stages of the pipeline, each with its own limited retention. Once data has aged out
of every stage, Jitsu no longer has a copy of it.

| Stage | Retention | What it holds |
| --- | --- | --- |
| Event stream | up to 16 hours | All incoming events, in the internal message queue (Kafka) during normal processing |
| Batched destinations | up to 2 days | Events destined for warehouses and other batch-mode destinations, until delivered |
| Failed events | 7 days | Undeliverable events, in a dead-letter queue for retries and troubleshooting |
| [Identity stitching](/docs/features/identity-stitching) | up to 30 days | When enabled, anonymous events waiting to be associated with a user profile |
| **Backups** | **your retention window** | Raw event backups in Google Cloud Storage — the only copy Jitsu can restore from |
| [Live Events](/docs/features/live-events) | most recent 200,000 entries | Events and function logs per configured entity (source, destination, …) for operational visibility |

## Restoring from a backup

Replays from the backup bucket are performed by the Jitsu team. If you need events restored into a destination,
[contact support](https://jitsu.com/contact) with the workspace, the destination and the time range.