Skip to content

Your agent doesn't need Claude at 3am.

Most agents use the same expensive model for everything — midnight batch jobs, repeated lookups, simple status checks. Windfall auto-routes each call to the right model, caches what it's seen before, and runs it on the cleanest energy available.

Two lines to switch. Up to 100 free requests. Every call attested onchain.

Get free requests
agent.py
# Change two lines. Everything else stays the same.
client = OpenAI(
base_url="https://windfall.ecofrontiers.xyz/v1",
api_key="wf_your_key_here"
)
response.windfall
{
"node": "nbg1", "location": "Nuremberg, DE",
"renewablePercent": 72.3,
"carbonIntensityGCO2": 128,
"costUsd": 0.004,
"cached": false,
"engagement": "warm"
}
OpenAI SDK compatible 200+ models Base L2 EAS attested

What changes

Before
→ Every call goes to Claude at $3/M tokens
→ Low-priority batch job? Same price.
→ Repeated prompt? Same price.
→ Unknown carbon footprint
↓
After Better
→ Auto-selects DeepSeek V3 (91% cheaper) — or specify any model you want
→ Repeated prompt → Cache hit ($0)
→ Every call routed to cleanest energy available
→ Carbon intensity attested onchain via EAS

What you'd save*

Drag the slider to your agent's monthly volume.

100 25,000 100,000
Your effective monthly cost with Windfall
$70.00/mo
Smart routing + ~30% cache hit rate (cache hits are free)
Claude direct
$150.00
Windfall (before cache)
$100.00
Total saved
$80.00/mo
Carbon avoided
3.0 kg

Based on $0.006/req Claude direct vs $0.004/req Windfall default (DeepSeek V3). Premium models at $0.008/req.

*Estimates based on current pricing. Actual savings depend on model mix, cache hit rate, and request volume. See /terms for details.

Get started in 60 seconds

1
Get your API key
curl -X POST https://windfall.ecofrontiers.xyz/api/keys \
  -H "Content-Type: application/json" \
  -d '{"label": "my-agent"}'
2
Point your agent at Windfall
# In your .env
OPENAI_BASE_URL=https://windfall.ecofrontiers.xyz/v1
OPENAI_API_KEY=wf_your_key_here
3
That's it. Windfall handles the rest.

Up to 100 free requests to start — more if you link a Base wallet or Basename. Auto-selects models, routes to cleanest energy, caches responses. Check your savings at /dashboard.

# Or use curl directly — specify any model
curl -X POST https://windfall.ecofrontiers.xyz/v1/chat/completions \
  -H "Authorization: Bearer wf_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-4-5-20250929",
    "messages": [{"role": "user", "content": "Hello from Windfall"}]
  }'

Omit model to let Windfall auto-select. Add X-Priority: low for background tasks.

How it works

Every request flows through four stages — from classification to verifiable proof.

1
Engagement classifier

Analyzes message frequency, conversation depth, and prompt complexity to classify each request as hot, warm, or cold. Cold requests get routed to cheaper models automatically. No configuration needed.

2
Energy oracle

Polls Electricity Maps every 5 minutes for real-time energy prices, carbon intensity, and renewable percentages across all nodes. Routes requests to wherever power is cheapest or greenest — you choose the mode.

3
Response cache

SHA-256 hash of your messages. If Windfall has seen this exact prompt before, you get an instant cached response at zero cost and zero latency. Agents repeat themselves more than you'd think.

4
Onchain attestation

Every batch of requests gets an EAS attestation on Base — proving where the compute ran, the energy price, and the carbon intensity at that moment. Verifiable green inference, not just a claim.

Any model. Your choice.
DeepSeek V3 default Claude 4.5 Sonnet GPT-4o Claude Opus 4 Llama 3.3 70B + 200 more via OpenRouter

Windfall picks the cheapest model that fits each request. Most of the time that's DeepSeek V3 at 91% less than Claude. When you need a specific model, just pass it — model: "anthropic/claude-sonnet-4" — and Windfall routes it through.

Green by default

Every call routes through the cleanest energy available. Not because you configured it — because the oracle does it automatically. Real-time data from Electricity Maps, updated every 5 minutes.

Cheapest lowest $/kWh
Greenest highest renewable %
Balanced cost + carbon
DE Nuremberg loading...
—
$/kWh
—
gCO2/kWh
—
% renewable
FI Helsinki loading...
—
$/kWh
—
gCO2/kWh
—
% renewable
We're hunting for green energy

Have a GPU server powered by renewables? We're actively expanding to locations with cheap, clean energy. Run a node and join the mesh — more nodes means better routing for everyone.

Choose your mode per-request with the X-Routing-Mode header. Default: greenest. ~120 gCO2 saved per 1,000 requests vs US-East.

Start routing smarter.

Up to 100 free requests. No credit card. Works with any OpenAI SDK client.

Get your API key View live status