The offer, on one page

You will know what your AI use costs and what it emits, team by team, then we work out with you how to bring it down. The first review arrives within one month if your teams are on Claude Code or Codex.

How a TokenClimate engagement works.

We work with organisations that already spend at least €2,000 a month on AI. It all starts with a 30-minute call.

  1. Assessment

    Day 1

    Connection

    We list your teams' AI tools and connect the ones whose usage we can measure. The dashboard fills up from the first key you enter.

    1 hof conversation, and one access key per tool

  2. Month 1

    First review

    Cost and emissions by team and by tool, with the assumptions behind each figure. In one month on Claude Code and Codex, longer depending on the other connectors needed.

    30 minto present it, with your leadership team if you like

  3. Monitoring

    Then, every month

    Report and review

    A written report: what changed in your data and in the market (models, prices, energy), and costed ways to cut.

    45 mina month to pick which options to try

    The next report measures the effect of the options you tried.

Tools and commitments

Connected today: Claude Code, the Claude API, the OpenAI API, Codex on an API key and OpenRouter. Another tool gets connected if we can measure its usage. Chat (claude.ai, ChatGPT) is estimated from licences, with a wider range. All connectors (in French)

  • We read your usage volumes without changing anything. No conversation is read.
  • Figures are grouped by team: nobody is named unless you choose to.
  • Hosting in Europe, and your data exported whenever you want.

The dashboard we open for you

It computes cost and CO₂e by team and by model from your Claude and OpenAI volumes. Shown here with sample data: click the filters.

[tokenclimate]Acme
Last 30 days

If you are below the threshold or not ready to talk yet, the calculator and the model sheets are free, with no account, and the demo shows the full dashboard on sample data.

If TokenClimate stops

What you download while the service runs stays with you: your data, the method as markdown. The open-source measurement tool does not depend on us at all. Here is what the product allows today, and what it does not allow yet.

Your data downloads as files
While the service runs, an admin downloads from the call log a CSV of every usage row in the chosen period: model, tokens by class, cost, CO₂e, water and factor version, up to 50,000 rows per file. Each member gets their own data as JSON from their personal page.
Measurement carries on, machine by machine
claude-carbon, the open-source plugin (MIT licence) TokenClimate grew out of, logs each Claude Code session in a SQLite database on the developer's machine, with none of our servers. Its repository holds the factors for Claude models, not those of other providers. It sees one machine at a time: the team view stops with the service. The code on GitHub →
Keep the method as markdown
The equations, the factors for each model and their sources are published and versioned on tokenclimate.com. The markdown version, /en/methodology.md, downloads and sits with your exports. The methodology →
The calculation can be redone from these files
Each CSV row carries the model, its four token classes and the version of the factors applied, not the factors themselves. With the method kept as markdown, a spreadsheet is enough to redo a row's CO₂e, or to apply another method to the same tokens.

Still missing: a single export of the whole organisation. The CSV does not carry each row's team, and team membership and settings do not leave as a data file.

What people ask us before starting.

Anthropic and OpenAI already have a dashboard. What do you add?
Each one only shows its own tool. We bring yours together by team, with the CO₂e next to the cost, and go through the figures with you every month.
Are your carbon figures reliable?
Anthropic and OpenAI do not publish the energy of their models, so the carbon figure is an estimate. We tell you what it rests on. The cost comes from your actual volumes. The calculation method →
Which tools connect?
Connected straight away: Claude Code, the Claude API, the OpenAI API, Codex on an API key and OpenRouter. Codex when it runs on an OpenAI API key is counted with the OpenAI API. For any other tool, we first check that your account gives access to its usage. If it doesn't, we don't connect it and you don't pay for it. Chat (claude.ai, ChatGPT) is estimated from licences, with a wider range than measured tools.
Couldn't we do this in-house?
Yes, if your teams have the time. Our method is already built, and it is public.
What if TokenClimate stops?
While the service runs, your usage rows download as CSV and the method as markdown: with these files, the calculation can be redone in a spreadsheet. The claude-carbon plugin stays open source and measures each machine without our servers. There is no single export of the whole organisation yet.
How much does it cost?
It depends on your tools and how much you spend on AI. We send you the price in writing after the 30-minute call.

Thirty minutes to see where you stand

You tell us how your team uses AI and what you want to know. We tell you where we can help. If there is a fit, a written proposal follows. You leave with next steps, even if we do not work together.