Mission and vision
Our mission is to make the environmental cost of AI visible and honestly answered: estimate it with the uncertainty stated, put real money into third-party environmental work in response, and give every customer a record that survives scrutiny.
The question we hear most is some version of: why don't you just make it neutral?
Because today, nobody can, and we would rather tell you why than pretend otherwise.
What stands between anyone and a true accounting
A neutrality claim asserts an equation: this much harm, matched by this much good, equals settled. Every term of that equation is broken right now, and not because of anything a customer did wrong.
The labs do not publish the data
No major provider discloses per-token energy, per-site water use, serving hardware, or utilization. Independent evidence and provider disclosures disagree by roughly an order of magnitude, which is why our estimates carry ranges that wide. You cannot balance a book whose left column is uncertain by a factor of ten, and anyone who claims otherwise is rounding uncertainty into marketing.
Nobody outside the labs knows where the compute happens
Which datacenter served your request determines the local grid's carbon intensity and which watershed the cooling drew from. For most API traffic that location is undisclosed, so we estimate under conservative unknown-facility parameters. Precise environmental accounting is a function of place, and the place is hidden.
Parts of the usage itself are invisible
Some providers do not report hidden reasoning tokens or cache dynamics in their usage APIs. Where a count is unreported we treat it conservatively rather than pretending it is zero, but conservative is not the same as known.
Water is local, and projects mostly are not where datacenters are
Restoring a gallon in one basin does not return a gallon to the aquifer a datacenter drew down. The restoration projects available today do real, measurable good, but they are rarely sited in the communities that feel the effects of the compute. We say so on every certificate rather than implying the geography away.
The instruments are not equivalence machines
A Renewable Energy Certificate is a generation attribute, not a carbon instrument. Carbon credits vary in vintage, method, and timing, and some fund reductions years out. These are real, verifiable instruments for funding real work; they are not arithmetic that cancels a footprint.
The law is catching up to all of this, and it should
The FTC Green Guides, California's AB 1305, and the EU's Directive 2024/825 all restrict unsubstantiated environmental status claims, for good reason. The claims that survive scrutiny are activity claims: what was funded, where, identified by project and vintage. Those are the only claims we support.
So we built the honest version
Estimates with the error bars left in, accrued at the conservative end of the range. Certificates retired every month, at least half of every subscription spent on them. A certificate that names projects, vintages, and the exact statements it supports. Not because honesty is a lesser product, but because it is the only one worth building: the record we give you today will still stand when the data improves and the regulations tighten.
Where we intend to take it
None of the obstacles above is permanent, and we are building as if they will fall. These are goals and intentions, stated with their conditions; they are not commitments, timelines, or claims about today's product.
Better data, better methodology
goal · not available todayOur methodology is versioned and built to be revised. As providers publish more (per-token energy, site-level water, regional serving data) or as independent measurement matures, the ranges narrow and every revision ships publicly, with historical estimates preserved under the version that produced them. The per-zone grid rebuild is already structured and waiting on pinned, citable data.
Comprehensive accounting, if the labs open up
goal · not available todayIf providers release usage APIs that state where compute ran and what it drew, we intend to build the comprehensive account this product cannot honestly offer today: location-resolved estimates against real facility data, not conservative defaults. The platform is architected for that day; the ingestion, versioning, and certificate machinery would not need to change shape.
Projects in the communities that host the compute
goal · not available todayWe intend to work toward water, carbon, and renewable energy projects sited in the watersheds and grids where datacenters actually operate, so the funded work lands where the effects are felt. Supplier catalogs do not support this well today. As they improve, and as our volume earns a seat at that table, community-sited projects are the goal we will keep pushing toward.
Independent review
goal · not available todayThird-party review of the methodology (the ISO 14064-3 path) is the intended step before any strengthened claim. Stronger language should be earned through verification, not asserted through copy.
Reduce first; this comes second
The first response to a footprint is to shrink it: smaller models where they suffice, cached context, batched workloads, fewer tokens. Every credible framework puts reduction ahead of funding external projects, and so do we. R3CKON does not make usage cost less or count less; it makes what remains visible and answered. One more honesty: the integrity frameworks that govern corporate climate claims (VCMI, SBTi, ICVCM's supply-side rules) require inventories, published targets, and assurance most small teams do not have, which is exactly why the only claims this platform supports are activity claims about what was funded, never status claims about what it achieves.
Until then, the rule that governs everything on this site stays simple: we say what was estimated and what was funded, and nothing more. The full methodology, with every assumption and its source, is public at r3ckon.com/methodology.