Usage & spend
Organization → Usage & Spend is the breakdown of what your organization actually consumed: what cost what, on which models, over which days. The By agent / By user toggle at the top left switches which axis the page groups by — the same usage, counted two ways.
Who can do what
| Task | User | Agent owner | Admin |
|---|---|---|---|
| View organization usage and spend | — | — | ✅ |
| Break spend down per agent and per model | — | — | ✅ |
| Break spend down per user | — | — | ✅ |
| Export or reconcile against invoices | — | — | ✅ |
Usage is organization-wide, so it lives with the admin. Owning an agent doesn't grant visibility into what other people's agents cost. If you want an agent owner to see spend, make them an admin — there is no narrower grant.
The summary cards
Across the top of the page, for the date range you've selected:
| Card | What it means |
|---|---|
| Raw Token Cost | What the model usage cost at provider rates, before Delegate's fee. |
| Total Spend | What you are billed — raw cost plus the service fee. The card's sub-label states the rate, or reads (no service fee) when your organization is charged none. |
| Total Tokens | Combined tokens, split into input / output underneath. |
| Top Agent / Top User | The single most expensive agent — or person — in the range, and their cost. |
| Active Agents / Active Users | How many agents ran at all, or how many people had spend, with the total task count. |
Raw Token Cost vs. Total Spend is the distinction to internalise: the first is the provider's price, the second is your bill. If you've brought your own key, the usage on that key appears in Raw Token Cost but carries no fee — you already paid the provider.
The per-agent table
One row per agent, most expensive first, with Tasks, Input Tokens, Output Tokens, Utilization, Total Cost, Cost/hr, and Avg/Day. The ordering is fixed, so the agent at the top of the table is always the one to look at first.
Expand a row to see that agent's usage split by model — Model, Provider, Tasks, Input, Output, Cost. This is how you answer "why is this agent expensive?": usually it's a small number of tasks on a large model, and the row-level split shows it immediately.
Cost/hr and Avg/Day are the columns to watch over time. A rising Avg/Day on an agent whose task count is flat means the work per task is growing — often a sign that an agent's instructions have drifted toward doing more than you asked, or that a routine is running more often than intended.
The per-user table
Switch the toggle to By user and the same table regroups by the person the turn was attributed to — most expensive first, with the same task, token, cost and cost-per-task columns, and the same expandable per-model split.
This is the view for "who is spending this?" rather than "what is spending this?". The two filters still apply and compose with it: leave the group-by on By user and pick an agent in the Filter by agent dropdown to see who is driving that one agent's spend.
Three things to know about how spend is attributed:
- A turn belongs to whoever sent the message that started it. Chatting with a shared agent puts that turn on your row, not the agent owner's.
- Scheduled routines are attributed to the agent's owner. Nobody is present when they run, and the owner is who configured them.
- Turns we can't tie to an account appear as Unattributed. The usual cause is a Slack message from someone whose Slack account isn't linked to Delegate. It is a real bucket of spend, counted in every total on the page — it just doesn't belong to a person we can name. To shrink it, get those people to link their accounts.
Cost over time
The Cost Over Time chart stacks daily spend by whichever axis the page is grouped by, so you can see both the total trend and which agent — or person — is driving it. Switch the granularity between day, week, and month, and set the range with the From / To date pickers.
Usage, budgets, and your bill
Three related but distinct things:
- Usage & Spend (this page) — what was consumed, historically.
- Budgets — forward-looking caps that throttle or stop agents before they spend.
- Plans & subscription — what Stripe actually charges, and the invoices behind it.
Usage figures are Delegate's own records. They're the input to your invoice, but the invoice is authoritative; small differences in timing (a task that finishes either side of a day boundary) are normal.
Reading it when something looks wrong
- Someone near the top who says they barely use it. They probably own a busy agent: scheduled work and messages from unlinked Slack accounts land on the owner's row. Switch to By agent, filter to their agents, and you'll see which one.
- An agent you don't recognise near the top. Check who owns it under agent sharing — shared agents run under the organization's budget regardless of who is chatting with them.
- Spend with no matching tasks. Scheduled routines run without anyone present. They appear here like any other usage.
- A blocked agent still showing cost. Work done before a budget stopped it is real usage and is billed.