Management console
—
| Key | Documents | Other calls | Prompt | Completion | Failed |
|---|
Every consumer gets its own key, so this is where spend actually comes from.
| Key | Requests | Prompt | Completion | Total tokens |
|---|
Leave a limit blank for unlimited. The key is shown once — copy it immediately.
| Name | Key | TPM | RPM | Parallel |
|---|
—
What is running on the GPU node and the settings it was started with.
Checked every 30 seconds against a rolling 24-hour window. Each rule re-arms at most once an hour.
| Rule | Metric | Threshold | Scope | Last fired |
|---|
The GPU node costs money only while it runs. These control when it sleeps and how long a cold request waits.
Used by the overview's cost panel. On-demand list prices; check them against the billing account monthly.
Commercial comparison — USD per million tokens, in / out
Hours when the node is held awake so interactive work never waits on a cold start (roughly 3 minutes). Outside the window it sleeps after the idle timeout and still wakes on demand.
Fill in any combination. Anything left blank is simply skipped.
Boots, shutdowns, key changes and errors.