Model Runs
Model Runs is the admin observability screen for model invocations recorded by Roster.
Only admins can view model runs. PII visibility follows the Model runs settings in Settings.
The table shows:
| Column | Meaning |
|---|---|
| Started | When the model run began. |
| Agent | Agent name such as roster.resolver. |
| Model | Provider and model name. |
| Status | Run status, with success and error states highlighted. |
| Mode | live for live/production execution or test for test execution. |
| Latency | Recorded model run latency. |
Filters
Section titled “Filters”The table supports:
- Search across model run fields
- Date: all, today, last 7 days, last 30 days, or custom range
- Status
- Mode
Details Drawer
Section titled “Details Drawer”Open a run to inspect:
- Started and completed time
- Latency
- Token usage: input, output, total
- Estimated cost
- Actor name/email or identity ID
- Provider and model
- Reasoning effort
- App version
- Run, resolve request, API token, MCP request, provider request, and trace IDs
- Input and output payloads
- Error details
- Tool calls, including input/output payloads, status, latency, and errors
If PII settings disable model-run input/output, tool payloads, actor details, or error details, those fields are redacted or omitted from the screen.
Relationship to Model Providers
Section titled “Relationship to Model Providers”Model Providers describes how the deployment chooses and secures provider credentials. Model Runs shows what actually ran, how long it took, how many tokens it used, and whether it succeeded.