Controller
A daily spend review you don’t have to run.
Every morning the controller reviews your agency’s spend, flags what is off and proposes the fix. Nothing changes until you click Apply.
No monthly subscription. Add funds when you’re ready to run.
- highBURN RATEBatiplus / Tender scraper
On pace for $212 against a $150 budget
Proposal: review the Oct 9 change, or raise the budget to $220.
DismissApply - mediumMODEL EFFICIENCYDurand / Site reports
claude-sonnet-5 used for calls under 200 tokens
Proposal: allow claude-haiku-4-5 only for this project.
DismissApply - lowIDLE BUDGETAtelier Lefèvre / Quotes
Has a $50.00 budget but spent $0.04 in 30 days
Proposal: budget $50.00 → $5.00.
DismissApply
Nobody has time to review AI spend every day.
Drift goes unnoticed
A project that costs a little more each week never triggers an alarm, until the month is over.
Premium models on small tasks
A workflow built on the best model keeps using it for jobs a cheaper one would handle.
Budgets set once, never revisited
A generous budget on a quiet project is exposure: a leaked key could spend all of it.
Small leaks add up across clients.
One inefficient workflow is a rounding error. Ten of them, across ten clients, is margin you gave away without deciding to. Finding them means reading reports nobody opens.
is how often spend changes, and almost never how often it gets reviewed.
the clients means ten times the projects, models and budgets to keep an eye on.
the month closes is when most agencies find out what went wrong.
A reviewer that reads every number, every day.
The controller computes its findings from your real spend, then writes them up in plain language with a proposed fix for each.
A digest you can read in a minute
Each morning you get a short summary of what needs attention, ordered by what matters most.
- Burn rate against each budget
- Cost drift from one week to the next
- Idle budgets and repeated blocked calls
Latest digest
The main item today: Batiplus / Tender scraper [1] is on pace to spend $212 this month against a $150 budget. Its daily cost doubled on Oct 9.
Also worth a quick fix: Durand / Site reports [2] runs a premium model on short summaries, and Atelier Lefèvre [3] has a $50 budget it barely uses.
Runs daily. Nothing changes until you apply it.
Run nowFindings with a one-click fix
Every finding comes with a concrete proposal: a new budget, a per-call cap, a cheaper allowed model. Apply it or dismiss it.
- Apply a proposal in one click, or dismiss it
- Nothing changes until you decide
- A history of what was applied and dismissed
- appliedApplied Oct 14
Budget $50.00 → $5.00
Atelier Lefèvre / Quotes
- appliedApplied Oct 12
Allowed models → claude-haiku-4-5
Durand / Site reports
- appliedApplied Oct 9
Max per call → $0.25
Batiplus / Tender scraper
- appliedApplied Oct 6
Client markup 0% → 30%
Moreau Charpente
It never invents numbers
Findings are calculated from your spend data first. The written digest can only say what those findings contain.
Daily, or when you ask
The review runs every morning, and you can run it again whenever you want a fresh look.
Available to your agents
Your own agent can run the review, read the findings and apply one through keyone’s MCP server.
Live in minutes.
01
Run your clients through keyone
The controller works from the spend keyone already records.
02
Read the morning digest
Open the spend review page, or get the summary by email.
03
Apply what makes sense
Accept a proposed fix in one click, and dismiss the rest.
Go deeper.
CarineController: questions.
Pick one client.
Know what their AI costs.
Connect a project, set its budget, and see its usage in one place.
No monthly subscription. Add funds when you’re ready to run. Or book a demo
