Who calls this API
Agent scenarios.
One pay-per-call API.
Every scenario below requires sub-hourly grid data — annual averages can't see the 22-minute windows where it matters. Each maps to one or more of the three signals we sell: generation mix, carbon intensity, and spot price.
Inference scheduler
The agent calls /mix/KPX/live every 5 min, sees CI
falling from 450 → 312 gCO₂/kWh as solar ramps up, and defers a non-urgent batch by 22 min.
A 30% emissions cut on the same workload — invisible to any annual-average API.
GET /api/mix/KPX/live
→ { ci_g_per_kwh: 312, mix: { solar: 0.34, ... } }
GPU workload router
Training jobs hit /spot/ERCOT/HB_NORTH
and /spot/CAISO/TH_NP15, see ERCOT at $18/MWh
vs CAISO at $94/MWh, and route the next checkpoint to a Texas region. Per-call arbitrage on wholesale
electricity, not retail tariffs.
GET /api/spot/ERCOT/HB_NORTH/live
→ { price_usd_per_mwh: 18.42, currency: "USD" }
EV / Tesla charging agent
Two Superchargers, 3 km apart. The agent sees one is on 70% solar at 11:00 (CI 142, $32/MWh), the other on 60% gas at 18:00 demand-peak (CI 480, $210/MWh). It picks the clean+cheap one. Per-charge: cleaner kWh, lower cost, and a receipt that proves both — useful for fleet ESG reports and retail-grade carbon claims.
GET /api/mix/CAISO/live + /api/spot/CAISO/TH_SP15/live
→ choose Supercharger A (low CI, low LMP)
24/7 CFE proof · DePIN slashing · CBAM
Every Grid402 response is a signed receipt: which 5-min interval, which zone, what mix, what CI, what spot price. Use it to prove a workload ran on carbon-free electricity, slash DePIN nodes on dirty grids, or file hourly CBAM reports without a $50k consultant.
GET /api/mix/{ISO}/at?ts=2026-04-25T14:30Z
→ signed JSON receipt for that 5-min interval
Why per-call, not annual
8 of the top 10 x402 agents run on anycast.
We pulled the top 10 agents on x402scan today. 8 of 10 run on globally-anycast infrastructure (Cloudflare Workers, Vercel Edge, GCP global load balancer). The same agent transaction lands in MISO when a Chicago user calls it, ERCOT when a Houston user calls it, and Frankfurt when a Berlin user calls it. Annual carbon accounting cannot capture this. Per-call grid receipts are the only mechanism that can — and they're the foundation a new shape of demand response has been waiting for.
The two fixed-region agents — Heurist on GCP us-central1 (MISO), AInalyst on DO Singapore — already need exactly what Grid402 sells.
The grid-operator framing
This is demand response — for autonomous compute.
Demand response (DR) is the grid-industry term for shifting load away from the hours when supply is stressed, dirty, or expensive — toward hours when it's clean and cheap. It's a $10B+ industry that utilities, grid operators (CAISO, ERCOT, PJM, ENTSO-E TSOs), and governments actively subsidize, because the alternative is grid-side intervention that's far more costly:
- Forced renewable curtailment — grid operators turn off wind and solar generators when there's too much supply (oversupply periods). The generator loses revenue; the planet loses clean MWh that already existed. ERCOT and CAISO routinely curtail GWh-scale renewable output every spring.
- Spinning reserves — fossil peakers kept burning fuel idle, ready to ramp. Costs $30–100/MWh of nothing, just for insurance against demand spikes.
- Forced load shedding — rolling blackouts. ERCOT February 2021 winter storm: $130B+ in damages, hundreds of deaths, all because demand couldn't be shifted in time.
- Peaker plant fires — diesel/gas peakers brought online for a few hours at $200–2000/MWh marginal cost, just because there was no other way to balance.
Classical DR aggregators (AutoGrid, Voltus, OhmConnect, EnergyHub) avoid these interventions by signing annual contracts with industrial fleets, HVAC operators, and battery owners — paid via SCADA integrations and monthly settlement. None of them can enroll an anycast autonomous agent whose "site" is a different region every request, whose owner is a wallet, and whose load is one HTTP call.
Grid402 is the primitive that makes per-call demand response possible: every agent request carries a grid receipt (mix + emissions + price for the responding region, on-chain), so the agent — or the platform routing it — can shift the next call to a cleaner hour or a greener region. No contracts. No SCADA. No human enrollment. Just per-call HTTP, USDC settlement, and grid signals delivered as the natural unit of an autonomous workload's decision.
Why governments and grid operators care: every MWh shifted out of a stressed hour is a MWh that doesn't require curtailment, doesn't require a peaker, doesn't risk a blackout. DR has historically been bottlenecked by enrollment friction. Anycast agents bypass enrollment by paying per call.