01
Count what actually ran
Every answer is logged with its actual model, token usage, estimated cost and energy impact. We do not apply one convenient average to everything.
Our impact
Better answers shouldn’t cost the earth.
AI has a physical footprint. We make ours visible, reduce it before offsetting it, and tie tree funding to real usage.
- Small-first
- model routing for everyday questions
- 1 tree
- funded monthly for every paid member
- Monthly
- planting evidence as funded batches land
The operating principles
Reduce first. Measure honestly. Fund the rest.
02
Use the smallest capable model
Everyday questions start with a smaller efficient model. Larger or specialised models are reserved for requests that genuinely benefit from them.
03
Under-claim, never greenwash
Our impact numbers are estimates because providers do not publish per-token electricity use. Assumptions are visible and deliberately conservative.
The estimate
Transparent assumptions, not invented precision.
Per-message energy combines model size, token counts, data-centre overhead and grid carbon intensity. Providers do not expose direct electricity measurements, so the result is an estimate—not a meter reading.
- 1.2
- data-centre overhead assumption
- 400g
- CO₂e per kWh grid assumption
- Per response
- model and token-aware accounting
Trees are the visible promise.
Every paid Treechat member funds one newly planted tree for every month of membership.
Partner names, planting locations and certificates will be published as each funded batch lands. Until then, we will not claim trees that have not been paid for.
Read our impact notes →