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

# Audit Log

> A tamper-evident record of everything that happens on your instance

The **Audit Hub** is the admin view into RelayHub's audit log: an immutable, hash-chained record of every state-changing action on your instance. It exists for compliance (SOC 2 style evidence), for incident investigation, and for answering "who changed this and when".

Open it from **Audit Hub** in the sidebar (administrators only).

## The Event Log

The **Events** tab lists audit events newest first, with the time, event type, action, affected resource, and the event's hash. Click any row to expand the full detail, including the event payload.

Filter by:

* **Event type**: Auth, Chat, Model, Agent, Tool, File, Settings, Admin, or System
* **Date range**: from and to dates

Events record what changed, who did it, and from where. Sensitive values are scrubbed before logging, so the audit trail itself never leaks secrets or PII.

## Chain Verification

Every event includes a cryptographic hash that chains to the previous event, which makes tampering detectable: removing or editing any historical event breaks the chain from that point forward.

The badge at the top of the page shows the current state:

* **Chain Verified** (green): the full chain checks out
* **Chain Broken at #N** (red): the chain fails at a specific event, which warrants investigation
* **Not Verified** (amber): verification has not been run recently

Click **Verify Now** to run a verification on demand.

## Exporting

Click **Export** to download audit events as an archive, optionally limited to a date range. Use this for auditor requests or offline retention.

## Wizard Build Health

The **Wizard** tab tracks the health of the AI skill-creation wizard: builds started, succeeded, and failed, cache hit rate, retry rate, and build duration percentiles over a selectable window (1 hour to 30 days). A recent-failures table shows when builds failed and why. Use it to spot systemic issues with skill generation before users report them.