Audit and usage logs
Who did what in your workspace, and every prediction it served — what each log records, the fields of an entry, filters, charts, and CSV export.
The dashboard keeps two logs for each workspace, under Observability. The audit log records account activity: changes to API keys, members, workspaces, and configuration. The usage log records every prediction the workspace served through the API or in batch. Both logs are available to the Creator and Auditor roles.
Both work the same way. Pick the workspace, open Audit or Usage, set the period and filters, open an entry to see its details, and export what you need as CSV.
Audit log
Observability → Audit records account activity:
| Area | What’s recorded |
|---|---|
| API keys | Creation, rotation, renaming, and deletion |
| Members | Invitations, role changes, and removals |
| Workspaces | Creation, settings changes, archival, and unarchival |
| Models and organization | Model alias changes, and organization settings |
The table shows each entry’s Event, Source, Actor Type, Actor Name, and Time. Select a row to open its details: the same fields, with the event shown as Action Type and the exact time as Event At, plus the entry’s identifiers and its full record.
| Field | What it shows |
|---|---|
| Event, or Action Type in the details | The action recorded, such as api-key-recycling for a key rotation |
| Source | Dashboard, API, or System, depending on where the action came from |
| Actor Type | User, API Key, or System, depending on who acted |
| Actor Name | The member’s or the key’s name, or the actor’s ID when there’s no name to show |
| Time, or Event At in the details | How long ago it happened, such as 2 hours ago, and the date for entries older than 30 days. The details show the exact date and time |
| Event ID | The entry’s unique ID, in the details |
| Actor ID | The member’s or the key’s ID, in the details |
| Model and Model ID | For an event about a model, the model it concerns, in the details |
| Action Details | The event’s full record, as JSON, in the details |
For example, the details of a key rotation made in the dashboard read like this, with made-up values:
| Field | Example |
|---|---|
| Event ID | 5b1f0c3e-7a2d-4c8b-9e61-0d4f2a7c9b13 |
| Action Type | api-key-recycling |
| Event At | Nov 3, 2026, 2:14:07 PM |
| Actor Type | User |
| Actor Name | Ana Souza |
| Source | Dashboard |
Usage log
Observability → Usage records every prediction the workspace served, through the API or in batch: its status, who made it, and the model. It shows the volume over time in a chart and the events in a table. The summary bar at the top of the table counts the period’s predictions: the Total, and how many were API Prediction and Batch Prediction.
The table shows each entry’s Event, Status, Model, Actor Type, Actor Name, and Time, and a row’s details add the rest.
| Field | What it shows |
|---|---|
| Event, or Action in the details | The action recorded for the prediction |
| Status | success or error |
| Model and Model ID | The model that served the prediction. The details add its ID |
| Actor Type | User, API Key, or System, depending on who made it |
| Actor Name | The member’s or the key’s name, or the actor’s ID when there’s no name to show |
| Time, or Event At in the details | How long ago it happened, and the date for entries older than 30 days. The details show the exact date and time |
| Type | API Prediction or Batch Prediction, in the details |
| Error | For a failed prediction, the error message, in the details |
| Event ID and Actor ID | The entry’s and the actor’s IDs, in the details |
| Action Details | The prediction’s full record, as JSON, in the details |
For example, a live prediction made with an API key reads like this, with made-up values:
| Field | Example |
|---|---|
| Type | API Prediction |
| Status | success |
| Model | churn-90d |
| Actor Type | API Key |
| Actor Name | credit-engine-prod |
| Event At | Nov 3, 2026, 2:16:41 PM |
Filter and explore
Both logs open on This month. To change the period, pick Last hour, Last 24h, Last 7 days, Last 30 days, or Last 90 days from the period menu, or choose a custom range of dates. The period goes back to This month each time you open the page.
Under Filters, narrow the entries down:
| Log | Filters |
|---|---|
| Audit | Action Type, Source, Actor Type, and Actor Name |
| Usage | Action Type, Status, Actor Type, and Actor Name |
Switch between Logs, the table, and Activity, which charts the same period: events over time, and for the audit log, the most frequent actions with their share of the total.
Export the logs
Exports are evidence for audits and reviews, so make them complete.
Open the log
In the workspace you’re reviewing, open Observability → Audit or Observability → Usage.
Set the period and filters
Choose the period, and narrow it down under Filters if you need to.
Load every entry
In the Logs view, scroll to the bottom of the table until it shows End of results.
Export
Select Export. The file downloads as audit-logs-YYYY-MM-DD.csv or usage-logs-YYYY-MM-DD.csv, dated the day you export.
| Log | CSV columns |
|---|---|
| Audit | Event ID, Event, Source, Actor Type, Actor Name, Event Time |
| Usage | Event ID, Event, Status, Model, Actor Type, Actor Name, Event Time |
Event Time is written as yyyy-MM-dd HH:mm:ss, in your browser’s time zone.
Using the logs
Use the logs for security monitoring, to spot unexpected key, member, or configuration changes. In incident response, they let you reconstruct what happened, by whom, and when: filter by Actor Name to follow one member or one key, and by Action Type to find one kind of change. For audits and reviews, export the evidence as CSV. To track usage, see which keys and models are serving predictions, and how many.
To trace a single API request, send an X-Correlation-Id header with it. Avra links it to the request’s audit events, so support can find the request later. See Live predictions.