Dashboard
The dashboard summarizes runtime readiness. It does not modify configuration; it helps determine whether the complete path from provider key to downstream request is ready.
Workspace states
| State | Meaning | Recommended action |
|---|---|---|
| Incomplete | A model resource or API configuration is missing | Follow the shortest setup path shown on the page |
| Callable | Platform model or connected model resources and an API configuration are ready | Send a request with an access key |
| Needs attention | The basic path exists, but recent requests or upstream health are abnormal | Check the client error and test the upstream |
Main sections
Upstream access
Shows the number and health of enabled upstreams. A warning means at least one enabled upstream failed its latest health check.
Model resources
Shows discovered routable models and their detection status. If an upstream exists but this section is empty, run discovery from Model resources.
API configurations
Shows how many configurations are ready to receive requests.
Recommended check order
- Verify that at least one platform model or connected model resource is routable.
- When using a connected model, verify its upstream is healthy.
- Verify that the API configuration is enabled and has candidates.
- Send a minimal request.
Common states
For Incomplete, inspect model resources and the API configuration. Quickstart may continue without a connected upstream or discovered model, but a configuration with no candidates cannot serve calls. For Needs attention, record the client error and test the selected resource or upstream.
Usage import status
The administrator usage summary distinguishes scan progress lag from execution time. Last successful scan duration measures the most recent successful incremental scan, excluding queue and safety waits. It shows “—” until a measurement is available. A failed scan retains the previous successful duration and reports the failure separately.
Data updated at identifies when the displayed status was collected; refresh the page for the latest status. A completed scan may import no new records. Daily historical reconciliation uses a separate queue and consumer from incremental scans.
Total quarantined records include Needs review and Reviewed · Unsettled. A request is reviewed only after another administrator approves its exception and no billing incidents remain open or under remediation. It remains quarantined, rather than settled or marked free. Use View reviewed incidents to inspect its review history.
When only reviewed quarantines remain and all other checks are healthy, the panel shows Exceptions reviewed and these records no longer trigger repeated alerts. Pending approvals, missing billing incidents and new open incidents still require attention. The page and independent monitor reflect their displayed check times; refresh after approval and allow the next monitoring run to update.
View reviewed incidents opens the Reviewed quarantines filter, which lists approved exceptions associated with those quarantined requests. History includes all closed incidents. One request can have several incidents, so the incident count can differ from the quarantined request count.
Invalid usage source data may never become an imported request. Its unresolved incidents still require attention and trigger alerts after a successful scan. Select the warning to inspect those incidents. The next health check reflects their resolution or approved exception.
Core concepts
Understand Lumoswitch upstreams, credentials, model resources, configurations, keys, and projects. Learn how client-facing models map to routing and failover.
Account and security
Manage Lumoswitch sign-in identities, passwords, and sessions. Verify phone or email changes, revoke unused sessions, and investigate suspicious account activity.