Product Updates · September 10, 2026 · 4 min read
Introducing the Usage Tab: See the Models and Cost Behind Every Turn
See the models and estimated cost behind every turn your Kolo takes. The new Usage tab breaks it down per turn, ties each one to the model and the activity, and exports to CSV.
By Paul Willmes, COO, Head of Product at Kolo
The question every owner eventually asks
Give a whole team AI employees and a new question shows up by the second invoice: what is this actually costing, and where is it going? For a long time the honest answer at our company was a shrug and a total. You could read the bill. You could not read the story behind it. The Usage tab is how we closed that gap, and it is now live inside every Kolo workspace.
You will find it under Conversations, on a tab simply called Usage. The header sets the expectation up front: "Estimated LLM usage for your account." It is a plain, per-turn ledger of the models your Kolo used, the tokens they read and wrote, and what that work is estimated to cost.
One row for every turn
The heart of the tab is a table with one row per turn. A turn is a message you sent plus everything your Kolo did to answer it. Behind a single question, your Kolo may call a model once or several times, pull in a Skill, read a file, and check its work. All of that rolls up into the one row for that turn, so you read the table the way you actually experienced the work, one exchange at a time, instead of drowning in raw model calls.
Each row carries a Date / Time, the LLM Model that answered, the number of Calls the turn made, the Tokens it used (prompt, completion, and cache read and write), and the estimated Cost. When a turn used more than one model, the LLM Model column shows the first and a "+N more" you can hover to see the rest. Each row also tells you where the work came from, whether it was a message in a conversation, a routine running on its own, or a call through the API, so a cost is never floating free of its source.
You can filter the whole view down to a single conversation to see that thread's estimated cost, switch the window between Day, Week, and Month, and export the current window to CSV when you want to slice it in a spreadsheet or hand it to finance.
The numbers at the top
Above the table sits a strip of totals for whatever window you are looking at: Entries (the count of rows below it), Calls (the individual model requests those turns made), Prompt tokens, Completion tokens, Cache read, and Cache write. The gap between Entries and Calls is worth a look on its own, because it shows how much fan-out a single request really produces.
The last tile is the one most people came for: Estimated Kolo Credits. It is the estimated cost of everything in the window, in the same credits your plan is billed in.
An estimate, not a bill
I want to be precise about that number, because precision is the whole point of shipping this. The cost on the Usage tab is an estimate, not your invoice. The tab says so in plain language: "Estimated usage, not a bill. See Plan and Billing for the amount actually charged." When some calls cannot be priced yet, the total is shown as an "at least" figure with a note about how many calls were left out, so the number never pretends to more certainty than it has.
For the amount you are actually charged, Plan and Billing remains the source of truth, one click away from the same screen. The Usage tab is there to explain the shape of your spending, not to replace the bill.
Built on Auto Select
The Usage tab is the natural companion to how Kolo already keeps costs down. With Auto Select, Kolo routes each task to a well-matched model instead of running everything on the most expensive one, and you can pin a specific model for a person, or just for one thread, when you want to. That routing has always been working under the hood. What was missing was a way to see the result. Now you can watch a stretch of routine work land on lighter models and reserve the frontier models for the turns that need them, and you can see it per turn rather than trusting that it happened. If you want the deeper story on how that routing decides, our write-up on Auto Select walks through it.
Why we built it
Most AI tools show you a monthly total and leave you to guess at the rest. As of September 2026, we have not seen another team AI employee break cost down to the individual turn, tie each one to the model that did the work and the activity it came from, and let you export the whole thing. That transparency is not a side feature for us. It is part of the same posture as the exportable Audit Trail: an AI employee that acts on your behalf should also show its work, including what it cost.
For an owner, that turns a vague worry into a decision you can make. You can see which workflows are cheap and which are heavy, whether a routine is quietly running up a bill, and where Auto Select is already saving you money. It is oversight of the spend, in the same read-only spirit as the rest of Kolo.
See it for your own team
The Usage tab is live now, under Conversations, in every workspace. Open it, switch to a Week view, and you will probably learn something about your own team's habits in the first minute. If you want to see it, Auto Select, and the rest of Kolo running for your team, Book a Demo.
Frequently asked questions
Where do I find the Usage tab?
It lives under Conversations, on a tab called Usage. The header reads "Estimated LLM usage for your account," and it opens on the current window with totals across the top and one row per turn below.
Is the cost shown the amount I am billed?
No. It is an estimate in Kolo Credits, labeled "Estimated usage, not a bill." For the amount actually charged, check Plan and Billing, which is linked right from the tab. When some calls cannot be priced, the total is shown as an "at least" figure.
What counts as one row?
One turn: a message you sent plus everything your Kolo did to answer it. A single turn can make several model calls, and they roll up into that one row, with a "+N more" when more than one model was involved.
Can I export the data?
Yes. You can switch between Day, Week, and Month windows, filter to a single conversation, and export the current window to CSV to analyze it in a spreadsheet.