The single home for AI spend. At the top, the company total: total token spend for the week, month and year (all functions). Below, three breakdowns: limits — weekly per-employee token caps and a monthly alert on the organization's budget; by person — agents and chat in tokens (who spends how much); by model — the bill in dollars, including indexing and search (where it goes). The panorama and the by-model breakdown come from the AI resource registry (model_usage); per-person spend comes from run and message journals. How accounting works — cost-accounting logic. Per-user cost in dollars, trends and export — v2.
The platform's full spend for the calendar window — all functions: indexing, RAG search, chat, agents. A derived SUM over the period from model_usage, with no dedicated counter table. Broader than the by person breakdown — there it's only agents and chat attributable to people.
The monthly tile is checked against the budget alert threshold. Trends, charts and export — v2.
Per employee — shared across all their agents. Spend is summed from the run journal; once past the cap, runs wait for the reset on Sunday 00:00. Empty means no limit.
Token budget · chat2
500,000
Per employee — the week's spend on personal conversations. In v1 it is informational: it shows "used against the limit" but does not block requests — a hard rejection past the cap arrives in v2. Empty means no limit.
Per organization · monthly bill
Budget alert3
$200 / mo
The monthly AI bill for the whole organization. Once the threshold is reached, a notification is raised; requests are not blocked. The toggle can't be turned on until an amount is set. The spend itself, by model, is in the by model breakdown below.
Token caps apply to every employee; the alert applies to the whole organization
1
The weekly agent budget is blocking: past the cap, runs are marked skipped and wait for the week to reset. It used to live on the AI agents screen and moved here with spend.
2
The weekly chat budget is informational: tracking and an indicator, no blocking. Both amounts are weekly, the window is calendar-based, and the reset is Sunday 00:00 in the organization's zone — shared with the agent budget.
3
The budget alert is the only monetary control in v1: the amount and toggle live in a singleton (ai_monthly_budget · ai_budget_alert_enabled), and a CHECK prevents enabling it without an amount. On reaching the threshold, a Budget-type notification is raised — the channel is owned by the notifications module. The monthly bill by model is in the by model breakdown below.
Per-user cost in dollars, caps as a limit is approached, and chat blocking — v2.
4
Filters — the Period · Role facets. Period defaults to "this week" (the window checked against the caps); past weeks and the month are available. Search and filter behavior is
shared across all lists.
5
The Agents and Chat columns are the two line items attributable to a person; Total is their sum, and it is the default sort (descending). Indexing and RAG search are excluded here — that is the platform's system spend, visible in the by model breakdown below.
6
A pill on agent spend means the gate is exceeded: new runs wait for the week to reset. On chat over the informational cap, a gray marker with no blocking.
Agent Engine.
7
Clicking a row opens the
employee detail: agents by agent, chat by model.
8
Input and output are counted separately: for cloud chat an output token costs several times more (data model). Embedding has no output. This is org level: a sum across the whole organization, not tied to a person — unlike the by-person breakdown above.
9
By default the top 5 by cost are shown; "Show all" expands the rest. Models without a price settle at the tail. Source — model_usage (aggregate by model · function · day).
The monthly threshold for this bill and the alert are in limits above.