Spend, per task

Every dollar of agent spend has a task and a name attached.

Token usage and tool spend are attributed to the task that incurred them and the person who approved it. Instead of one opaque model bill at the end of the month, you see which work cost what, and where a cheaper model would have done the job.

Get started Read the docs

This week · 5 tasks$36.00

The problem with usage you can't tie to work

Most providers do hand you a usage breakdown, but it collapses to per-API-key or per-team totals. You can see the engineering key burned thousands last month; you can't see whether that went to fixing CI, drafting board decks, or a runaway retry loop. Spend maps to credentials and cost centers, never to the actual unit of work that incurred it. So you can't tell which tasks were worth it, charge cost back to the work that caused it, or spot where a cheaper model would have done the job.

How cost attribution works

Per-task metering

Tokens and tool calls are tallied against the task that triggered them, so spend maps to real units of work.

Owner attribution

Every task carries the human who approved it, making chargeback and budget ownership straightforward.

Model routing

Send low-risk or simple tasks to cheaper models and reserve frontier models for the work that needs them.

Budgets & alerts

Set ceilings per team or workflow and get warned before a loop or a bad prompt burns the budget.

Spend stops being a surprise and starts being a lever you control.

It's one of six power-ups you get from approving the task instead of the tools.

Fail-closed by default

Containment

Agents start with zero access. Every tool is granted per-task and revoked when the task ends. A compromised prompt can't reach what the task never needed.

See every action