AdminGuide
The record of AI use
Emargen’s AI runs on your organisation’s own accounts. The provider’s bill is yours, and it arrives as one number. The record of AI use is how you explain it. It keeps every call to your models, its purpose, who it was for, and what it cost at list prices.
To read everyone’s figures and change prices you need a role that may manage models. Everyone else reads their own month in Your usage.
A month, explained
Section titled “A month, explained”Usage, opened from Models, shows one month at a time. It opens with the month’s cost at list prices, the tokens the models read and wrote, and the number of calls. Then the cost by kind of work and by model, and the five people and the five initiatives that cost most, each with a link to the full list.
Two kinds of work are easy to mix up. Coding agents are the code-reading jobs Emargen runs on your own Agents default, such as a repository’s analysis. Your people’s own assistants, like Claude or ChatGPT connected through agent access, run on their accounts and don’t appear at all.
Say you suspect one big initiative is behind this month’s cost. Usage shows what went on it, next to the month’s total. It also shows how much wasn’t about any initiative: the work in topics, the Inbox and settings. Often that’s most of the month. By person shows whose work it was. So before you ask a team to cut back, look at where the money went.
Choosing a model for a kind of work
Section titled “Choosing a model for a kind of work”A default covers a whole kind of work. Change the Chat default, and the conversations and background jobs that run on it move to the new model from their next call. Anything already running finishes on the old one. Change the Agents default, and every analysis and code check moves.
There’s one exception. A persona can name a model of its own, and conversations on that persona keep it, whatever the Chat default says. A persona that thinks longer before it answers also costs more.
By model on Usage shows what each model came to. By person and by initiative show what that work was for. Maybe a cheaper chat model would suit the whole team, or a cheaper agents model all the code reading.
Before and after a switch
Section titled “Before and after a switch”Compare the two months side by side. Look at the totals, the cost by kind of work and by model, and the people and initiatives. Every month keeps its own address, so you can send a colleague the one you’re looking at.
A lower cost tells you spending fell. It doesn’t tell you why. The amount of work changes from month to month. So does the length of what the models read and wrote, and how much of it came from the cache, which costs less. The month’s tokens, calls and cache share, beside the cost, show how the work changed. They don’t split the change into the price and the rest. So when you report on a switch, say what changed in those terms, not that the model saved the money.
List prices, and yours
Section titled “List prices, and yours”The record counts cost at list prices, and leaves your discounts out. We left them out for now, because list prices are enough to show where the money goes.
If a model’s cost looks wrong, or it shows none, give it your own prices on its provider’s page. Calls from then on are counted at your prices. Calls already recorded keep the prices they were charged at, so last month doesn’t change under you.
It only counts
Section titled “It only counts”The record only counts. No figure on Usage holds work back. What limits how much runs at once is your plan.
The one limit on spending is for flows. On Models, you can give the AI that flows use a monthly limit in dollars. Past it, a person answers instead of the agent. Everything else runs whatever the month has cost.
Some use is left out on purpose. The admin agent and other work on Emargen’s own models are recorded, but Usage doesn’t show them. Cost by topic isn’t shown either, though every call keeps its topic. Private and shared topics are hard to tell apart there, and there are far more topics than initiatives.