Observability & spend

What ran, what it cost, and what went wrong.

Most AI tools tell you the answer. Almost none tell you what happened. The Hub keeps an account-wide record of every AI action taken on your behalf — traced, timed and priced to the token.

Account-wide, not per-project

Every workflow run across every project, plus Studio, in one view. If you only ever look at one project, you miss the workflow in another that is doing the spending.

Measured, not estimated

Token counts come back from the model provider on every single call, so cost is measured rather than guessed from the length of the text. The number you see is the number you were charged.

Failures you can diagnose

A failed run records the step that failed and what it was given — including, for web steps, which addresses were tried and what happened to each.

Where cost surprises actually come from

Nearly every unexpected bill has one of two causes: a scheduled workflow running more often than intended, or one grounded on far more project material than it needs. The by-workflow breakdown finds both in about ten seconds.

By workflow. Runs, failures, average step time, tokens and cost, per workflow.
By model. More capable models cost more per token; you can see which ran.
Studio separately. AI you used while building, distinguished from what ran on its own.

By workflow — last 30 days

Weekly client briefing4 runs · 1.9s avg · $0.42

Filing monitor31 runs · 2.4s avg · $6.10

Competitor digest93 runs · 3.1s avg · $11.88

Scheduled hourly — likely more often than intendedcheck

One meter, not two. The spend shown here is the same spend that draws down your account credit. Billing shows the balance; observability shows where it went.

The Hub is ready. Access is by invitation.

We work with a deliberately small group of researchers — close enough that the Hub is built around how the work is actually done. Tell us what you research and we'll be in touch.

Request an invitation