See every AI call in the usage report

Run AI usage in Analytics to check calls, tokens, failures and what a key costs you.

In short

Every model call is recorded: who made it, which feature, provider and model, which key, whether it worked, the tokens and the time. The AI usage report in Analytics adds them up. Use it to check your provider's bill or find a key that keeps failing.

Before you start

Who: anyone who can run reports sees their own calls. A site admin sees the whole platform; an organization's admins see its calls in its AI settings.
Plan: Analytics and AI must both be on your plan.
First: nothing. The default period is the last 30 days.

Steps

Open Analytics › Reports and choose "AI usage".
Pick the period; the default is the last 30 days.
Run it and read model calls, success rate, failed calls, tokens in and out and average latency.
Below, read the split by feature and model and the most recent failures.

Result: you see how many calls each feature made and which failed.

Good to know

The prompt and the answer are never recorded - only that a call happened, its size, time and the start of any error.
It shows your own calls. A site admin sees the whole platform's, with an estimated cost.
Each key test counts as a call.
An organization's admins see the organization's AI calls in its AI settings, with whose key answered and what it cost.

If something goes wrong

The report is empty: no model was called in the period. Widen the period.
Many failed calls on one key: press "Test" on that key in Settings › AI and fix what it reports.