Every prompt has a footprint.
We're building the ledger.
Token Offset accounts for the environmental cost of artificial intelligence, for individuals who want to use AI without the guilt, and for organizations getting ahead of what's coming.
AI's share of the world's electricity is climbing fast, and the people using it rarely pay for its footprint.
Data centers, driven mostly by AI workloads, are projected to consume by 2030, rivaling small industrialized nations.
For a 400-token Le Chat response in Mistral's lifecycle assessment when upstream power-generation water is included.
Carry a paid AI offset, chosen at checkout with no default since ONCE added the option in May 2026. When it takes one click, some people pay.
“The greatest threat to our planet is the belief that someone else will save it.”
From tokens to offsets in three steps.
Each step uses published numbers you can check. 90% of every offset payment goes to our climate partner, and Token Offset keeps 10% to run the service.
Estimate your usage
Pick the band that fits, from Casual to Extreme, or connect a read-only key from Claude, OpenAI, Cursor, Grok or OpenRouter and we'll sample your real usage. Token counts only; we never see your prompts.
See what that costs the planet
We turn your estimated tokens into Wh, mL of water, and grams of CO₂ using coefficients drawn from published research, which we publish with their sources.
Cover it monthly
A single subscription routes funds to vetted carbon removal, watershed restoration, and grid decarbonization projects, with receipts.
Where do you fit?
AI usage spans six orders of magnitude across users. We calibrated four bands against public studies, operator disclosures, and real agent traces so the math feels true to your reality.
Occasional questions to Claude and ChatGPT
A few thousand tokens on use days, or roughly a few dozen short conversations per month.
Average human knowledge worker, daily AI use
20,000 to 250,000 tokens per workday for writing, research, analysis, and repeatable GPT/project workflows.
Coding agents and multi-step workflows
Hundreds of millions of tokens per month for daily coding agents, long-context tasks, retries, and context replay.
Internal devs, eval pipelines, agent fleets
Up to 210 billion tokens in a single week at the very top of the distribution.
1,000 tokens ≈ 750 words · Sources & calibration
Find your number.
Don't want to estimate? Pick any monthly amount you're comfortable with. Every contribution is allocated the same way.
How much AI do you use?
Pick the description that fits you. We'll estimate a 100% contribution for the carbon, water, and grid load behind your tokens.
- Annual CO₂e
- 36.0 kg
- Trees / year
- 1.7
How we estimate1.0 Wh · 6 mL · 1.2 g CO₂e per 1k output tokens
- Energy
- 1.0 Whper 1,000 output-equivalent tokens
- Water
- 6 mLper 1,000 output-equivalent (site + upstream)
- CO₂e
- 1.2 gper 1,000 output-equivalent (lifecycle)
- Cache
- 2%of an output token per cache read (fresh input 25%)
- ~70%Durable removal credits (biochar, enhanced weathering, mineralization)
- ~15%Watershed restoration & grid decarbonization
- ~10%Verification, audit, and registry fees
- ~5%Research fund (open methodology, peer review)
- Microsoft (2025): Energy use of AI inference0.31 Wh typical / 3.91 Wh long query
- Requesty (2026): Coding-agent cache hits92% of Claude Code input served from cache
- Jegham et al. (2025): How Hungry is AI?GPT-4o: 0.42 Wh / short query
- Google (2025): AI at Google Scale0.24 Wh, 0.26 mL / Gemini prompt
- Epoch AI (2025): ChatGPT energy0.30 Wh / GPT-4o short query
- Li et al. (2023): Making AI Less ThirstyLifecycle water methodology
- U.S. EPA eGRID0.359 kg CO₂e/kWh national avg
- Mistral AI (2024): Large 2 LCA1.14 g CO₂e / 400-token query
- CDR.fyi + FrontierDurable CDR pricing ranges
Rate 50c / 1M output-equivalent tokens · Full methodology at /docs/methodology
The math is the same for one person or ten thousand.
Use AI without the guilt.
If you've ever paused before sending a 200-message thread to ChatGPT, this is for you. Pay a transparent monthly contribution and know that the carbon, water, and load you generated are accounted for.
- Pick a usage band in 60 seconds, or sharpen it with a read-only key from Claude, OpenAI, Cursor, Grok or OpenRouter
- From $4.99 / mo. Pay what fits your usage
- Your contributions and footprint in one dashboard
Get ahead of the regulation that's coming.
The EU AI Act already requires providers of general-purpose models to document their energy use, and California's SB 253 makes large companies report their emissions. Measuring AI usage now makes those reports easier to file.
- Org dashboard with usage and footprint
- An opt-in offset line for your own customers
Run your account from the assistant you already use.
Connect Claude, ChatGPT, Cursor or any MCP client to Token Offset and ask it to check this month's estimate, change your plan, log the tokens an agent used, or disconnect an app. You sign in with your Token Offset account and choose whether it can make changes or only read.
Setup guides and tool referenceServer address
https://mcp.tokenoffset.com/mcp
- Sign-in
- OAuth with your Token Offset email
- Tools
- 24 account tools
Works with
- Claude
- ChatGPT
- Cursor
- VS Code
- Windsurf
- Claude Code
- Any remote MCP client
No greenwashing and no vibes-based credits.
You get receipts.
Transparent coefficients
Our energy, water, and emissions estimates are published with sources and revised quarterly. Nothing is invented to make a graph look impressive.
A published 90/10 split
Token Offset is a for-profit company and isn't a charity. 90% of each offset payment goes to our climate partner and we keep 10% to run the service. Payments aren't tax-deductible.
Customer-owned data
We never see, store, or train on your prompts. We measure token counts and nothing else.
Receipts you can show
Your dashboard shows what you have paid and how much of it has gone to our climate partner.
Ready when you are.
Sign up with your email, pick a usage band, and set your monthly offset in about a minute. No card needed to look around.
Get started