AI Consumption brings your company's AI usage into one place. It shows the cost and tokens your AI providers report, which models, people and API keys drive them, and how AI coding tools like Claude Code and Cursor are adopted.
You'll find it in the sidebar under AI → AI Consumption. It has two tabs, Overview and Explorer, and every application that reports AI metrics also gets its own AI Consumption tab.
Supported integrations
AI Consumption shows data from these integrations:
Integration | What it adds |
Anthropic API Platform | Reported cost and tokens for Claude API usage, by model and API key. Also Claude Code adoption: sessions, commits, pull requests and lines of code. |
Claude Enterprise | Token usage and cost for each user (usage-based Enterprise plans only). Set up Claude Enterprise |
OpenAI API Platform | Reported cost and tokens, by model, user and API key. Set up OpenAI |
Amazon Web Services (Amazon Bedrock) | Bedrock cost and tokens from your AWS billing data. |
Cursor | Adoption across your team: lines changed, requests, suggestions and Tab completions. |
To connect one, go to Integrations and filter on the AI Consumption category, or click Connect on the AI Consumption page. Data appears after the first sync.
GitHub Copilot and Microsoft Copilot are coming soon.
Overview
The Overview gives you the big picture for the period you pick:
Time period: This month, Last 3 months, Last 6 months or Last 12 months. It applies to the whole page.
Key figures: Reported AI cost, Total tokens and Active users & API keys, each with a 12-week trend line. Cost and tokens compare with the same days of the previous period.
Reported cost by provider: a stacked chart of what each provider reports over the period.
Top: switch between Models and Users & API keys to see the top 10, with their cost, share of cost and change.
AI Providers & Apps: one card per provider, with its cost or tokens, trend, number of models and top model. Sort by Reported cost, Tokens, Cost increase or Name.
Click Open in Explorer under the chart or the Top list to dig deeper.
Explorer
The Explorer lets you slice the data the way you need:
Cost or Tokens: pick the metric.
Group by: Model provider, Inference provider, Model, Provider feed or Actor (User/Key). Tokens can also be grouped by Country, Cost center, Division or Department, with a choice of department level.
Date range (This month, or the last 3, 6 or 12 months), Time granularity (Daily, Weekly or Monthly) and Bar chart or Area chart.
Filters: Employee, Department and Group, plus Country, Division and Cost center.
Summary: for cost, Reported cost (vs. the previous period), Daily average, Share of SaaS cost and Top cost driver. For tokens, Total, Input, Output and Cached input tokens.
Table: each group's cost or tokens and its share. Tokens are split into input, output, cache read and cache write. Expand a provider row to see its models, or search the table.
Your selections are saved in the page URL, so you can bookmark a view or share it with a colleague.
AI consumption for an application
When an application reports AI metrics, its page gets an AI Consumption tab, and its Overview tab gets an AI consumption (this month) card that links to it.
The tab shows Cost, Tokens and Adoption, but only the groups that app reports. Adoption shows how a tool is actually used, for example:
Claude Code: sessions, commits, pull requests, and lines added and removed.
Cursor: lines changed, requests, suggestions and Tab completions accepted.
Group by the app's own fields, such as Model or Terminal for Claude Code, or by Actor (User/Key), Country, Cost center, Division or Department. Date range, granularity, filters and chart type work as in the Explorer.
Cost on this tab is estimated cost, as reported by the tool. It's separate from the reported cost on the AI Consumption pages.
Key terms
Reported cost: the cost your providers report for AI usage, in the currency they report it in.
Inference provider / Model provider: who ran the model (for example AWS for Claude on Bedrock) vs. who made it (Anthropic).
Provider feed: the integration or account the data came from.
Actor (User/Key): the person or API key behind the usage. An actor marked Unmapped couldn't be matched to an employee.
Unattributed: usage the provider reported without a value for the chosen grouping, such as cost without a model.
FAQ
Why is the latest day incomplete?
Providers publish usage with a delay, so the most recent day can still change after the next sync.
Why does reported cost differ from my invoice?
Reported cost comes from each provider's usage reporting. Invoices can include discounts, taxes, commitments or seat fees. Seat-based licenses are tracked under Licenses.
Why can't I group cost by department?
In the Explorer, the organization groupings (Country, Cost center, Division and Department) are available for tokens. Use the Employee, Department and Group filters to narrow down cost.
Is AI Consumption included in my plan?
Yes. It's available to all workspaces at no extra cost.