Usage & activity
You don't toggle metering on anywhere — qualified inference traffic tied to your API keys automatically feeds AI Studio's Activity and Usage / analytics surfaces and (depending on your subscription) supports estimated spend snapshots. Failures typically log too — handy when you're debugging flaky calls.
Wondering strictly about integrating (URLs, payloads)? That's in Inference API — Quick Start land; here we describe what you're looking at in the product.
What's behind each row?
Roughly captures when, who (organization, project, key), model, outcome, and volume (tokens and other units depend on the modality). If your plan includes metered usage, Studio may show estimated spend in a reporting currency — treat these as indicators, not your invoice.
Failures (4xx/5xx) still usually appear so you can fix integration mistakes; billing impact depends on the plan + model semantics.
From an API request to charts (plain language)
- Client calls Nebul Inference with a Bearer key belonging to a project.
- Nebul runs the inference and captures metering data you'd expect for stewardship + billing pipelines.
- Within moments you can chase single calls in Activity and aggregates in Usage dashboards.
Different screens answer different questions:
| Where in Studio | What it helps with |
|---|---|
| Inference → Activity log | "What literally happened?" — request trail for the currently active project |
| Inference → Usage | "How noisy / pricey was period X, possibly split by model vs key?" — scoped to active project |
| Settings → Organization → Activity log (organization owners) | Organization-wide sibling of Activity plus a Which project though? column |
| Settings → Organization → Usage (organization owners) | Compare projects side by side at organization scale |
They're two views of the same underlying activity — detail vs rollups — not separate silos you need to reconcile by hand.
About "cost" in the UI
- Rough / indicative versus whatever shows up downstream on contractual invoices unless we explicitly mirror something else.
- Per-model quirks abound — some payloads simply won't monetize symmetrically everywhere.
- Some deployments hide currencies outright (bundled / prepaid / pilot programs). Tokens & throughput still behave.
Questions about bill shock versus what's on screen → support portal is the fastest loop.