Your AI bill is going up. So is your carbon footprint.
Your AI bill comes as a single line, from Anthropic and OpenAI alike. TokenClimate shows the detail: who uses what, in which team, on which model. Plus the CO₂ and litres of water behind each session. We set up the dashboard with you, then help you act on it.
What your Anthropic and OpenAI invoices don't tell you.
Three things missing every month when you have to make the call. TokenClimate puts a number on each, team by team.
2 min 16, in French with English subtitles
Anthropic's Analytics API only sees your organisation. Individual Pro and Max subscriptions, API keys outside the organisation and isolated heavy users slip past it.
No CO₂ or water data consolidated by team and by month. The footprint of your AI usage shows up nowhere today.
No way to tell who weighs the most, which project is overrunning or which team has actually adopted the tool.
Your dashboard, updated after every session.
No third-party integration to maintain.
One command per machine, nothing to integrate into your code. Every Claude Code session is measured token by token: input, output and cache.
After each session, the CLI counts the tokens locally and sends them to the dashboard, which converts them into cost, CO₂ and water. No prompt and no line of code leaves the machine.
The dashboard breaks usage down to the developer and the model. Month-on-month comparison, target tracking, PowerPoint export and a scope 3 category 1 line.
Two ways in, depending on what brings you here.
We work with organisations that already spend at least €2,000 a month on AI. It all starts with a 30-minute call; we install the dashboard with you during the engagement.
- ·Admin connectors set up on your Claude, OpenAI and OpenRouter accounts
- ·Cost, CO₂e and water baseline, by team and by person
- ·A verdict per seat: flat-rate plan that pays off, under-used, or usage-based billing
- ·The Opus / Sonnet trade-off, priced in euros and in gCO₂e
- ·Action plan and debrief
- ·An annex ready to attach to a tender with environmental criteria
- ·The AI line of a greenhouse gas inventory, with the method cited
- ·A published, versioned methodology that a buyer can check
The natural next step after Scoping: a monthly review of the figures, follow-up on the recommendations, and the ESRS export (scope 3 cat. 1) once a year; a greenhouse gas inventory can be written from the same export. The report itself is generated in one click whenever you need it, always on up-to-date data. The format is agreed in the proposal, directly with the founder.
Few organisations can show these figures today. You would be among the first to track AI cost, CO₂ and real usage together, and that shows in tenders as much as in board meetings.
Nearly 10 years of carbon & climate work with
Four profiles, one way in for each.
The real total by team and by project, tracked month after month. Who uses what, which licences sit idle. You decide on actual figures.
Each developer's actual usage, model by model, with the input / output / cache split. You see which session blew the month's budget, and what the light users are doing.
CO₂ and water consolidated by team and by month, exportable as a scope 3 category 1 line.
The calculator compares Claude, GPT, Gemini or Mistral on your volumes, no account needed. And the claude-carbon hook measures locally, without a single prompt leaving your machine.
Investment funds: measure Claude usage across the portfolio, company by company. Book a call →
It started with claude-carbon.
claude-carbon, released in April 2026, is the solo open source version. TokenClimate extends the same measurement to the whole team, with a shared dashboard, sustainability emails and tailored support.
A bash Stop hook, a local SQLite database, text and PNG reports through slash commands. The data stays on the developer's machine.
Who I am.
Gaëtan Wittebolle. Nearly ten years in carbon & climate strategy (4 years at EY, 5 as an independent consultant for large French companies). Creator of Carbonara and claude-carbon.
Find me on LinkedIn →
