The bill
Who uses what?
Your AI runs through API keys or Bedrock, the bill keeps rising, and nobody can assign it to a team. So it is €300 per head per month, and whether that is too much or too little, nobody knows.
You bring AI
into your organisation.
We bring organisation
into your AI.
A budget per team replaces the cap per head, rules replace the ban. Measured and enforced in your infrastructure, not at the provider.
Problem
Does this sound familiar?
The data
What may the agent do?
Nothing may leave the building, and an agent must not touch the laptop. So officially there is no agent. The developers have one anyway, on a private account.
How it works
The meter and the guardrail are the same part.
Every call from people, tools and agents passes through a gateway in your account. That is where it gets counted, and that is where the rules apply.
A budget per team and application. It replaces the cap per head and is derived from actual usage.
The data class picks the model. Internal data stays with EU inference or on your own hardware. Frontier models are used only with zero retention, and the decision is documented.
Pseudonymisation before the model, filtering before the output. Need-to-know is enforced by the system.
Tool permissions per agent. What an agent may read, write and execute is your decision.
An audit entry per call. Who, what, with which model, for what. Ready whenever someone asks.
What you see: cost and hours per task, cost per outcome, adoption per department, rules and violations. In one dashboard, on your side.
Our stance
Your data. Your infrastructure. Your decision.
The provider's region is the beginning, not the answer. What goes into a prompt and what an agent may touch is decided before the provider, on your side. EU inference is the default. Where a frontier model is needed, with zero retention and documented. Where on-premise AI makes sense, we run the full numbers for hardware and operations.
- Zero Data Retention
- EU inference
- In your own account
How it runs
Three steps, one principle. Measure first, then decide.
- 01
Cost Picture
The bill per key, team and model: who carries the usage, where the cap bites, what is paid for twice. One intervention with an amount per finding, at a fixed price.
- 02
Gateway rollout
The gateway in your account, with budgets, data classes, agent permissions and audit. The dashboard goes on, the cap comes off. Fixed price as a share of your AI spend.
- 03
Handover
Your team runs it. Tokenometrics stays as a licence, self-hosted. We come back when a new model ships, or the auditor does.
After that, if you want it
Process analysis, prototypes with kill criteria, implementation with a success component. Only once the gateway shows where it pays.
Data and agents
You decide what the agent may see, not the provider.
The same rules that apply towards the outside apply inside the building.
An agent that may read everything on the laptop is not a productivity tool but an open window. So every agent gets exactly the permissions its task needs, and not one path more. Data is pseudonymised before a model sees it, and filtered before it reaches a person who should not.
What the EU AI Act demands in oversight, logging and traceability falls out of that as a by-product, not as a project afterwards.
Experience
Our insights
AI was never a hype for us. It is our craft.
Contact
Running a cap? Let's do the numbers.
Cost Picture at a fixed price, a clear result, no obligation.
An answer within 24 hours.