AI agent cost tracking
Know which agent spends. Before the invoice does.
See what every agent and every model costs, estimated from the tokens recorded on each run. Set monthly budgets for the workspace and for single agents, and hear about a runaway agent on the day it starts.
The problem
The monthly bill is the wrong place to look.
Most of what an agent costs is tokens, run after run. Each run is cheap, so nobody looks at runs; they look at the bill, and by then a wasteful agent has been wasting for weeks. The waste rarely comes from the model's price. It comes from the agent reading more than it needs.
- Whole documents when a part would do. A full web page fetched to find three lines that changed.
- The same context on every step. A long thread sent back with every tool result.
- The wrong model for the job. A large model sorting emails into five labels.
Usage
Spend by agent, by model, over time.
Usage shows total spend, average cost per run, tokens and runs for the last day, week, 30 or 90 days, against the period before. A trend shows spend across the period, and a table ranks every agent by what it cost.
Owners and admins also see spend by model, and by member for the agents each person owns.
Cost is estimated from the tokens recorded on every run, at each model's list price. It tells you where the money goes; your provider's invoice stays the final word.
Budgets
A limit for the workspace, and for the agents you worry about.
Set a monthly budget for the whole workspace and a smaller one for any single agent. Each shows the month's spend against its limit, and changes colour at 80% and again once it is over.
A budget does not stop an agent. It makes sure someone hears about it: every workspace comes with a Monthly budget exceeded alert in the app, so setting a workspace budget is enough. For an agent's budget, add a rule for that agent, with email or a webhook if you want it to reach further.
With a workspace budget set, the Intelligence page also warns when the month's burn rate is on track to go over it.
Cutting the waste
Point at the waste, then fix it.
- Token Optimizer ranks the fixes. One of the built-in intelligence agents reads recent runs and ranks the prompt changes that would cut the waste, each with its estimated saving.
- Every run shows its tokens. Tokens in and out on every model call, so a step that reads far more than it needs stands out.
- Budgets from day one. The budget alert is already there. Owners and admins set and remove budgets from Usage.
In our own workspace
One agent cost thirty times more than we expected.
One search run cost around $0.60, against about $0.02 for publishing a post. A monthly bill would have hidden that for weeks. Cost per run, per agent, made it obvious right away. The agent is paused until reading is worth that price to us.
Go deeper
How it works, in detail
More in Observe
See what every agent did, what it cost, and when it breaks.
Give one job to an agent this week.
Start with one repetitive workflow, gate the sensitive step behind your approval, and read the first run end to end.