What are we paying for?
Seats, API keys and extra usage across several vendors, with no one view of the total.
Holistic AI EndLayer / AI spend & usage
AI spend & usage.
AI spend & usage.
What AI costs, and whose it is.
EndLayer reads token usage from the coding agents on each enrolled device and prices it against a model price table refreshed daily. See the cost by tool, model, person and project, set it beside the seats you pay for, and find the seats nobody uses.
From usage on each device to cost you can attribute
01 / Where usage comes from
Agent usage recordsClaude Code, Codex, Cursor, OpenCode
Signed-in accountsCompany or personal, and the plan
Recorded plansSeats and tiers you pay for
Vendor admin keysThe billed cost, where connected
03 / One tool, accounted for
● 30 days
Claude Code
- Tokens
- 842M across 12 models
- Token cost
- $717, at list prices
- Plan
- Team seats, recorded
- Wasted seats
- 4: holders used less than half the seat's price
Next decision
Reclaim the idle seats, or move heavy users to the plan that fits them.
Illustrative figuresRead → Price → Attribute
Cost before it compoundsAI spend is on
AI spend is on
a dozen bills at once.
What is AI spend management?
AI tools are paid for as seats, as API usage, as extra usage on top of a plan, and sometimes on a personal card. No single invoice says which teams use what, which projects the tokens went to, or which seats nobody touches.
AI spend management puts the usage and the cost in one place, attributes them to the person, project and tool behind them, and shows where the money is not earning its keep.
Who is not using theirs?
Seats bought in a rollout stay on the bill long after the holder stopped using them.
Where did the tokens go?
Without the person and the project, a cost spike is a number nobody can explain or plan for.
What you seeOne cost, four ways
One cost, four ways
to divide it.
The same money, split by the tool that spent it, the person it reaches, the work it was filed under and the provider it went to.
By tool
Each AI app and agent with its plan, whether it acts unattended, its tokens, their value at API prices, extra usage and what it actually costs you.
By person
Each person's usage and cost beside the seats they hold. Where Claude Code or Codex reports the signed-in account, that names the seat and its tier; a personal account is no company seat.
By project
Usage split by the repository the work was done in, from the agent's own session records, in tokens and in seats.
By provider and model
The window's cost per model provider and per model, with the vendor's own invoice shown beside the derived figure once an admin key is connected.
Usage is read from the tools that keep local records. A seat used only on the web is on the bill, but its use is counted as visits, not tokens.
Explore Monitor ↗Cost is derived, and labelled as suchEvery figure says
Every figure says
where it came from.
No device reports a bill. EndLayer prices the tokens it collected and keeps that figure apart from what you were actually charged.
FigureWhere it comes fromRead it as
Token cost
Tokens by model, priced at read time against a model price table refreshed daily.
What the usage is worth at list prices. A model with no published price is named and left out, so the total is a lower bound.
Spend
The vendor's bill where an admin key is connected, or the seats and extra usage of the plan you recorded.
What you actually paid. A tool with neither is left out and named under the figure.
Wasted seats
The plan you recorded, set against each holder's measured use.
A seat whose holder used less than half its price over the window. It reads a dash until a plan is recorded.
How AI spend worksFrom one device
From one device
to one number.
Enrol the device. Read the usage. Price and attribute it. Then decide what changes.
- 01 / Enrol
Start on the device.
Install the Layer 1 Agent. No per-tool integration is needed to start reading usage.
Agent installation ↗ - 02 / Read
Read the usage.
Tokens by day, tool and model over a 30-day window, from the coding agents' local usage records, with the account each tool is signed in with.
The 21 signals ↗ - 03 / Attribute
Price it and name it.
Price the tokens, attach them to the person and the project, and record the plans you pay for so spend and idle seats can be worked out.
Monitor ↗ - 04 / Decide
Change what costs too much.
Reclaim idle seats, settle on the tools you sanction, and hand the period to finance as a report, in PDF or slides.
Govern ↗
Usage, not content.
Usage records carry the day, tool, model and token counts. The prompts and responses are not part of them.
The domain, not the address.
For a signed-in account EndLayer keeps whether it is a company or personal account, the plan and the email domain, never the part before the @.
Per-person cost is a permission.
Who can see what one person costs is a role decision your administrators make. Others see the totals.
Five practical answers
Before you start.
Coverage, accuracy, billing, seats and privacy. These are the details to settle in your evaluation.
Which AI tools does EndLayer measure usage for?
Token usage is read from the local usage records of Claude Code, Codex, Cursor and OpenCode on enrolled devices. The browser extension counts prompts sent from AI websites. Ask for coverage of your exact tools and plans.
Is the cost EndLayer shows what we are billed?
Not by itself. Token cost is the usage priced at list prices, labelled as such. Spend is what you actually paid: the vendor's bill where an admin key is connected, or the seats and extra usage of the plan you record. Both are shown, side by side.
Can EndLayer read our vendor invoices?
Connect an Anthropic or Cursor admin key to read the billed cost and flag seats nobody uses. Connecting a provider's billing needs an owner of the EndLayer organisation.
How does EndLayer decide a seat is wasted?
A seat is wasted when its holder used less than half its price over the window. Seats come from the plans you record and, where Claude Code or Codex reports the signed-in account, from that account and its tier.
Does measuring AI spend mean reading prompts?
No. Usage records carry the day, tool, model and token counts, not the conversation. Prompt and response collection is off by default and is a separate administrator choice.
EndLayer
Give AI room to work.And clear boundaries.
See the Guardian Agent Platform on your own fleet. Find every AI tool, understand what it reaches and costs, and apply the controls your team needs.
One connected layer of control
Devices · Cloud · ApplicationsSurface / Monitor / Govern