Admin dashboard

Model usage

How much your organization uses its AI provider: every request to a model and the tokens it took, by day and by kind of call, for this month and last month.

Who can see itAdmins and owners
ShowsThis month and last month
CountsRequests and tokens
DaysMidnight to midnight, in the service’s time zone
In the admin dashboard: UsageModel usage Open the dashboard
On this page

The month#

Model usage shows one month at a time. Under the choice of month, one line gives the month’s requests, its tokens and the time zone of its days.

FeatureDescriptionDefault
This month

The current month, from its first day until now.

Shown first
Last month

The whole month before this one.

—

A request is one call to a model. One turn of the assistant can make more than one request, for example when it uses a tool.

At a glance#

Four numbers for the month, in tokens: the input in three parts, then the output. Providers set a different price for each part, so the page counts them apart. It shows no prices: your provider’s price list says what each part costs.

FeatureDescriptionDefault
Fresh input

The input that is new to the model in each request. Under it: its part of all the input.

—
Cache writes

The input that your provider stores in its prompt cache, so that the next requests can read it back. Under it: its part of all the input.

—
Cache reads

The input that the model read back from your provider’s prompt cache. Under it: its part of all the input.

—
Output

What the model wrote. Under it: the average for one request.

—

The charts#

Input and output have a chart each, each on its own scale. To read the numbers of a day or a kind of call, point at it, or move to it with the Tab key.

FeatureDescriptionDefault
Input by day

A column for each day of the month, in the three parts of the input.

—
Show the numbers

Shows a table under Input by day: each day with requests, with its requests, its tokens in each part and its total, and the month’s total. Hide the numbers removes the table.

—
Output by day

A column for each day of the month: what the model wrote.

—
By kind of call

A bar for each kind of call, in the four parts, the most tokens first. A table under the bars gives each kind’s requests, its tokens in each part and its total. Under each kind’s name is the model it ran on. If it ran on more than one model, each model is named with its tokens.

—

The kinds of call#

By kind of call names each kind of the assistant’s work. Each kind runs on the model and effort you choose for it under Models & provider, or on the recommended ones.

FeatureDescriptionDefault
Ambient turns

The assistant’s turns over what members’ desktops hear and record outside meetings, when no message from them waits for an answer. They run on the model for Ambient turns.

—
Chat turns

The assistant’s answers to a member’s own messages outside meetings. They run on the model for Chat.

—
Meeting turns

The assistant’s turns while a meeting is live. They run on the model for Meetings.

—
Meeting recaps

The recap of each meeting. They run on the model for Meetings.

—
Meeting follow-ups

The assistant’s answers to questions about a meeting after it ends. They run on the model for Meetings.

—
Doing summarizer

The short line that says what each member is doing, written from the activity their desktop records. It runs on the model for Doing summarizer.

—
Work jobs

The jobs that members accept from their assistant’s offers or ask for in chat. They run on the model for Work jobs.

—
Work job checks

The check of how each job went. It runs on the model for Work jobs.

—
Work scout

The scout that looks for work each member’s assistant could offer them, every four hours. It runs on the model for Work scout.

—
Wiki updates

Each meeting’s recap, put into the wiki when the meeting ends. They run on the model for Wiki upkeep.

—
Wiki lint

The wiki’s regular pass, which puts what is new, and the comments, into its pages. Run lint starts one at once. It runs on the model for Wiki upkeep.

—
Page descriptions

The line each wiki page is listed by. They run on the model for Page descriptions.

—
Retrospectives

The retrospectives: each member’s, and each team’s and the organization’s. They run on the model for Retrospectives.

—

What is counted#

  • Every request to a model for your organization: each member’s assistant, and the work it does in the background.
  • A request counts in the month, and on the day, that its turn started. The days run from midnight to midnight in the service’s time zone, which the page names.
  • Each provider reports the tokens in its own way. A provider that reports no prompt cache, such as an Ollama server, has all its input counted as fresh input. A request for which the provider reported no tokens counts as a request with no tokens.
  • The numbers are those of the moment you open the page.
  • Models & provider: the model and effort for each kind of call.
  • Assistant pace: how often each member’s assistant is asked, which sets how often ambient turns run.
  • Overview: this month’s tokens at a glance.
Esc
↑ ↓ to moveEnter to open