What users spent on the assistant (spec §7.3), counted from the AI request journal (phoenix_kit_ai_requests) rather than from the module's own messages: a row of the user (user_uuid) whose metadata.source is "agent" — every model call of the module passes both, so a call that describes an attachment is counted too. The journal is not pruned by age, so the month never loses rows. The journal keeps a cost in dollars (cost, a numeric that can be finer than a micro-dollar); the sums are in micro-dollars ($1 = 1_000_000): each window is added up in dollars and only then rounded, half up, to a whole micro-dollar (PhoenixKitAgent.Config.cost_to_micro/1) — two rows of $0.0000004 are 1, not 0. Only endpoints that report a cost add to it.

Every row with a cost counts, whatever its status: a call that failed or timed out after the provider billed it is money spent all the same. Today phoenix_kit_ai writes a cost only on its "success" rows (a failed call is logged with cost NULL, which adds nothing), so the sums equal a success-only count. For the same reason the queries do not filter on status = 'success' and cannot use core's partial index phoenix_kit_ai_requests_user_spend_idx; they go by the journal's user_uuid index and the month's window.

The windows are calendar days and months in the site time zone (Settings → time_zone; PhoenixKit.Utils.TimeZone), each from its local midnight to the next one — so a day that has 23 or 25 hours is counted right.

Turns are counted from the module's own messages (turns_today/2): a turn is one user message with a turn_ref.