Usage

The Usage screen charts what your organization's keys have done: spend, requests, tokens and errors over a period, with an optional breakdown. Everyone in the organization can open it.

Every choice you make on this screen is kept in the page address, so a link to it opens the same view.

Choose the time range

Press 24h, 7d, 30d or 90d at the top. The default is 24h. With 90d selected, a note says the figures come from daily summaries and that very old requests are not included.

Break the figures down

group by adds a breakdown table under the charts. Choose one of:

• project
• key
• model
• provider
• app
• tag

For tag, a tag key box appears. Type the name of the tag you send with your requests; the table then has one row per value of that tag.

Choose none to remove the table.

Filter by project

The project list next to the controls limits every figure on the screen to one project. all projects removes the filter.

Compare with the previous period

Tick compare to previous. The Spend and Requests cards then show the change against the period of the same length just before, with the earlier figure beside it.

What the screen shows

Four cards summarise the range:

• Spend: the total cost.
• Requests: the number of requests.
• Tokens: input plus output tokens, with the cached and reasoning counts below.
• Error rate: the share of requests that failed, with the cache hit rate below.

Four charts show the same range over time: Spend (µ$), Requests (successful and failed), Tokens by kind (input, cached, output, reasoning) and Error rate (%).

With a group-by chosen, the table has one row per group with its requests, errors, tokens and spend. Hover over a spend figure for the exact micro-USD value.

Explain a spike

Each row of the breakdown table ends with explain. It opens the request log filtered to that group, so you can see the individual requests behind a figure. Without a group-by, the open the activity log link under the charts opens the unfiltered log.

Export CSV

Press CSV to download what is on screen. With a group-by you get one row per group; without one, one row per time bucket. Each row has the requests, successes, errors, token counts and cost in micro-USD. The button is disabled when there is nothing to export.

The "analytics delayed" banner

Usage figures are analytics. They are written a moment after each request, and can fall behind. When they are more than 60 seconds behind, an amber banner at the top reads "analytics delayed", with the number of seconds, and says the figures below may lag.

The delay affects only what this screen reports. Your balance and the ledger on the Credits screen are always exact. Wait and reload; the missing requests appear once analytics catch up.

Two related signs:

• A chip reading "as of" with a time next to the title means you are looking at figures fetched more than a minute ago.
• If analytics cannot be reached at all, the screen shows an error with a request id and a Retry button. Your requests are still served and charged correctly meanwhile.

Other notes

• If the range has no requests, the screen reads "No requests in the last" followed by the range. Widen the range, or check that your keys are in use.
• With the environment switch in the top bar set to test, a note says that sandbox requests are unbilled and reported separately, and that the figures shown are for live keys.

The same figures are available from the API; see the usage reference.