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

# History & revert

> See every change made to your agent — who, when, what — and roll back the most recent change per field.

The **Verlauf** tab on each agent's detail page is a per-agent audit log. Every field-level change is recorded with the actor, timestamp, before/after values, and (for AI-driven edits) the chip indicating the [Builder assistant](/user-guide/chatbot/builder-assistant) was involved.

## What gets logged

| Action              | What you see in the history                                                                                                                    |
| ------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------- |
| **Creation**        | "Morty P created the entity" with timestamp.                                                                                                   |
| **Field edit**      | "Morty P changed *Begrüßung* from *Hi there* to *Hallo, wie kann ich helfen?*" with before/after chips.                                        |
| **Long-value edit** | Same as above, but with an **expand** toggle that opens a side-by-side before/after diff (used for prompt edits — they're too long to inline). |
| **Builder edit**    | Same as a field edit, plus a small builder avatar overlaid on the operator's avatar and a **via Builder** chip after the operator name.        |
| **Revert**          | A new history row with a **Zurückgesetzt** amber chip, showing the rollback as its own audited action.                                         |

## Dual avatars on AI-driven changes

When Pam (the [Builder assistant](/user-guide/chatbot/builder-assistant)) makes a change on your behalf, the history row shows:

* **Your avatar** as the main face (you initiated the change).
* **Pam's avatar** as a small overlay in the bottom-right of your avatar.
* A **via Builder** chip next to your name.

This makes it instantly clear which changes came from the AI versus changes you made directly in the form. The actor on the row is always *you* — Pam acts on your behalf, never as a separate entity.

## Revert a change

Each history row whose change is the **most recent for its field** shows a **↶ Rückgängig** button. Clicking it:

1. Opens a confirm modal: *"Diese Änderung an X wirklich zurücksetzen?"*
2. On confirm, writes the *before* value back to the agent.
3. Records the rollback as a new audit row tagged **Zurückgesetzt**.

<Info>
  Only the latest change per field is revertable. Older entries don't show the button because reverting an older change would either no-op (if it was already overwritten) or skip intermediate edits silently — both confusing. To roll back further, revert from the top of the list down.
</Info>

## What's NOT in the history

* **Knowledge bucket content** — adding/removing entries inside a bucket is logged on the bucket, not the agent. The agent only logs the *link/unlink* event.
* **Conversation messages** — those live in the [Konversationen](/user-guide/chatbot/playground) tab. Audit is for the agent's configuration, not its outputs.
* **Read events** — opening the agent or running Pam without making changes leaves no audit trail. Only writes are logged.

## Compliance

The audit log is retained indefinitely on the account. For GDPR / SOC2 contexts you can export the log via the [conversations API](/api-reference/conversations) (the activity endpoint is documented under the conversations group — same auth model).