Usage and analytics

What the Console's Dashboard and Usage pages show: requests, tokens and cost, usage by model and by key, and the metadata-only request log.

The Console has two places to watch your inference traffic. Dashboard sums up the last 14 days. Usage lets you pick a period, split it by model and by key, and read the request log. Both read the gateway's own metering, which is the same record your credit is charged from.

Early access. The inference gateway is deployed but switched off. Safeguard switches it on per workspace. Until it is on for yours, these pages have no gateway figures to show.

Dashboard

The Dashboard covers the last 14 days, counted in UTC days up to today.

Card	What it shows
Requests	Every logged request in those 14 days. The total cost appears under the count, as $… spent.
Input tokens	Input tokens across those requests.
Output tokens	Output tokens across those requests.
Active API keys	Your own active inference keys. Keys other members created are not counted.

Under the cards, Requests per day draws one bar per UTC day. Before your first request, the chart says there are no requests yet and points you to API keys and the Workbench.

Usage

Pick the period at the top of the page: Last 7 days, Last 14 days, Last 30 days or Last 90 days. The default is 30 days. Periods are UTC days ending today.

Totals

Requests, Input tokens, Output tokens and Cost for the period. Requests includes requests that ended with an error.

By model and By key

Two tables split the period's totals. Each row has Requests, Input, Output and Cost.

By model has a row per model id. When a fallback model answered a request, the request counts under the fallback. See Reliability and fallbacks.
By key has a row per key, shown by its name and masked form, such as production backend · sg-api-1a2b3c4d…9f8e. The name and mask are the ones the key had when the requests were made, so a renamed or revoked key stays recognizable. Workbench runs use no key, so they are grouped as Console workbench. A key that no longer exists shows as Deleted key. The Console never shows a key's numeric id.

The API keys page shows each key's spend for the current month in its This month column.

Request log

The request log lists requests newest first, 50 at a time. Load older requests pages back through the whole history, whatever period you picked above. The Console notes above the table: "Metadata only. Prompts and outputs are not shown here."

Column	What it shows
Time	When the gateway recorded the request, at the end of the answer, in your browser's time zone. The server sends it in UTC.
Model	The model that answered. For a request that failed, the model you asked for.
Key	The key the request used, named as in By key. Console workbench for a run in the Workbench.
Input	Input tokens, as the provider counted them.
Output	Output tokens. est. next to the number means the provider reported no usage, so the tokens and the charge are the gateway's estimate.
Cost	What the request was charged, in USD. Amounts under a dollar show more decimal places.
Latency	Milliseconds from the start of the provider call to the end of the answer, including any retry or fallback.
Status	OK, or the error code.

The status codes you are most likely to see:

Status	What happened
`upstream_error`, `upstream_timeout`, `upstream_auth`, `upstream_rate_limited`	The provider failed, did not answer in time, refused Safeguard's credentials or rate limited the call. If no answer was produced, the cost is $0. If a streamed answer broke partway, the tokens already produced are charged.
`upstream_rejected`	The provider refused the request as invalid. The cost is $0.
`client_disconnected`	Your client closed a streamed response before it finished. The gateway kept reading the provider's answer to the end and charged for all of it.

Some requests never reach the log. When the gateway refuses a request before calling a provider, it costs nothing and is not logged: your client gets the error directly. That covers a missing or revoked key, the API being switched off, an unknown or unavailable model, too little credit, a spend limit, a rate limit and an invalid request body. See Troubleshooting.

How the figures are counted

Tokens are the provider's own counts. Cost is those tokens times the model's list price, with no markup. See Pricing and billing.
The Dashboard and Usage count UTC days. Monthly figures in the Console use UTC calendar months.
Usage, the request log and the credit figures cover the whole workspace. Every member who can open the Console sees all of it, not only their own requests. The exceptions are your keys: the API keys page and the Dashboard's Active API keys card show only keys you own.
The Console never shows prompts or outputs. Safeguard stores them only if a tenant admin or super admin turns on Store prompts and outputs. See Privacy and data.