0:00
The gap: inference bills show tokens and dollars.
Sustainability teams may also need carbon and water estimates for inference workloads. CarbonLayer puts those modeled estimates beside the request receipt.
Screen 1 · the homepage
Prototype walkthrough · real API receipts
See CarbonLayer in three minutes — without a meeting.
The sample HTTP call reaches the real API and the displayed receipt is a real captured response. Model output and dispatch are simulated. Environmental values are modeled estimates; region targets and latency are seeded snapshots, not physical devices. The physical edge network is in development.
No demo call is needed. Use the written walkthrough, then send a workload if you want to discuss a pilot.
Captured receipt · 880 tokens
A real API response from the prototype; environmental values are modeled estimates and latency is seeded.
CO₂e · modeled estimate
5.94 g
Water · modeled estimate
76.05 mL
Saved · modeled baseline
136.6 g
Latency · seeded
78 ms
Sample output is simulated; this page does not make an inference call.
Beat by beat · 3:00 total
The three-minute tour, written out.
Each displayed API value is from a captured real receipt. The model output and dispatch are simulated, environmental values are modeled estimates, and target latency is seeded.
0:25
The request: one POST with a familiar JSON shape.
Send a model label, prompt, token budget, and latency preference to the prototype API. Model output and dispatch are simulated; the region target and latency are seeded snapshots.
Screen 2 · the docs quickstart
0:55
The receipt: a real response with modeled estimates.
The sample POST reached the real API and returned a persisted receipt. Its model output and dispatch are simulated, and its carbon, water, and energy values are modeled estimates.
Screen 3 · the response contract and captured receipt
1:35
The ledger: a key-scoped view of requests.
GET /api/savings is authenticated and scoped to the API key. The captured example returned HTTP 200 with three dispatches; it is not a public anonymous ledger.
Screen 4 · the captured key-scoped savings response
2:05
The fleet: seven seeded region targets.
GET /api/network is public and returns the current seeded target snapshot. Environmental values are modeled, latency is seeded, and these targets are not physical devices.
Screen 5 · the public seeded network snapshot
2:35
The next step: start free or send a workload.
The Free tier is listed at $0 per month with 50,000 calls per month. Start with an account, or send a workload through the existing contact form for a pilot discussion.
Screen 6 · the pricing snapshot
Prototype boundary
No language-model provider is called for this demo. The API receipt is real, while generated model output and dispatch are simulated. Read the methodology and provenance notes for the calculation boundaries.
Annotated screens
The real pages behind the walkthrough.
These are local captures of the existing CarbonLayer home, docs, and pricing routes. No screenshot from the unavailable reference asset paths is used.
Screen 1 · home
Prototype status comes before the action.
The current home page states that carbon and water figures are modeled and that the physical edge network is in development.
Screen 2 · API reference
The request shape is explicit.
The docs show the request body and identify claude-sonnet as a simulated model label. The prototype does not call a language-model provider.
Read the methodology
Screen 3 · pricing
Plan details stay on the pricing page.
Pro and Team are optional, availability-qualified tiers. Hosted checkout depends on Stripe Connect configuration.
See current pricingReal API calls · captured 23 Sep 2026
Three receipts, with their limits stated.
The sample HTTP calls and recorded receipts are real. Model output and dispatch are simulated. Carbon, water, energy, and baseline comparisons are modeled estimates; each latency is a seeded target value.
Request · POST /api/inference
Uses the CarbonLayer API origin and a placeholder bearer key. Replace the key after signup.
bash
curl -sS -X POST https://carbonlayer.polsia.io/api/inference \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"prompt":"Draft a status update for a 12% emissions drop in Marseille.","model":"claude-sonnet","tokens":880,"preferLowLatency":false}'Captured receipt · abridged JSON
Values are from a real API response; model output and dispatch are simulated.
json
{
"model": "claude-sonnet",
"tokens": 880,
"energyKwh": 0.216,
"carbonGrams": 5.94,
"waterMl": 76.05,
"latencyMs": 78,
"baselineCarbonGrams": 129.6,
"savedCarbonGrams": 136.6,
"device": {
"name": "Oslo Harbour Hydro Cluster",
"region": "eu-north-1",
"carbonIntensity": 25,
"renewableMix": 0.82,
"carbonIntensitySource": "modeled"
},
"carbonSource": "modeled",
"waterSource": "modeled",
"energySource": "modeled",
"source": "synthetic"
}carbonSource, waterSource, energySource, and the device intensity source are modeled. The response source is synthetic; the target and its latency are seeded.
5.94 g CO₂e
Modeled estimate
76.05 mL water
Modeled estimate
136.6 g saved
Modeled baseline comparison
78 ms
Seeded target latency
The three captured calls
| Example · model | Tokens | CO₂e · modeled | Water · modeled | Saved vs baseline · modeled | Latency · seeded |
|---|---|---|---|---|---|
| Marseille status updateclaude-sonnet | 880 | 5.94 g | 76.05 mL | 136.6 g · 351.7 mL | 78 ms |
| EU grid-intensity summaryclaude-haiku | 120 | 0.54 g | 6.9 mL | 12.4 g · 32.0 mL | 78 ms |
| Support-ticket classifiercl-compact-v1 | 60 | 0.30 g | 3.9 mL | 6.9 g · 17.8 mL | 78 ms |
Marseille status update
claude-sonnet
Tokens
880
CO₂e · modeled
5.94 g
Water · modeled
76.05 mL
Saved · modeled
136.6 g · 351.7 mL
Latency · seeded
78 ms
EU grid-intensity summary
claude-haiku
Tokens
120
CO₂e · modeled
0.54 g
Water · modeled
6.9 mL
Saved · modeled
12.4 g · 32.0 mL
Latency · seeded
78 ms
Support-ticket classifier
cl-compact-v1
Tokens
60
CO₂e · modeled
0.30 g
Water · modeled
3.9 mL
Saved · modeled
6.9 g · 17.8 mL
Latency · seeded
78 ms
The captured key-scoped GET /api/savings response returned HTTP 200 and three dispatches. Its authenticated ledger is scoped to the key that made those requests. For the calculation boundaries behind each modeled value, read the methodology.
GET /api/network · public seeded snapshot
Seven region targets, one honest table.
This is the captured snapshot shown by the public API. Grid intensity and renewable mix are modeled estimates. Latency values are seeded for region targets; the physical edge network is in development, so these are not physical devices.
Open the current /api/network JSON snapshot
| Region target | Country | Grid intensity · modeled | Renewable mix · modeled | Latency · seeded |
|---|---|---|---|---|
| Oslo Harbour Hydro Clustereu-north-1 | Norway | 25 g/kWh | 81.6% | 78 ms |
| Quebec Hydro Edgeus-east-1 | Canada | 37.5 g/kWh | 71.1% | 92 ms |
| São Paulo Hydro Roofsa-east-1 | Brazil | 36.0 g/kWh | 79.1% | 184 ms |
| Tasmania Wind + Hydrooc-east-1 | Australia | 51.8 g/kWh | 74.1% | 220 ms |
| Columbia Gorge Wind Podus-west-2 | United States | 94.1 g/kWh | 56.1% | 110 ms |
| Andalucía Solar Labeu-west-2 | Spain | 180.2 g/kWh | 21.3% | 150 ms |
| Karnataka Renewable Bridgeap-south-1 | India | 261.8 g/kWh | 42.5% | 168 ms |
Oslo Harbour Hydro Cluster
eu-north-1 · Norway
Grid intensity · modeled
25 g/kWh
Renewable mix · modeled
81.6%
Latency · seeded
78 ms
Quebec Hydro Edge
us-east-1 · Canada
Grid intensity · modeled
37.5 g/kWh
Renewable mix · modeled
71.1%
Latency · seeded
92 ms
São Paulo Hydro Roof
sa-east-1 · Brazil
Grid intensity · modeled
36.0 g/kWh
Renewable mix · modeled
79.1%
Latency · seeded
184 ms
Tasmania Wind + Hydro
oc-east-1 · Australia
Grid intensity · modeled
51.8 g/kWh
Renewable mix · modeled
74.1%
Latency · seeded
220 ms
Columbia Gorge Wind Pod
us-west-2 · United States
Grid intensity · modeled
94.1 g/kWh
Renewable mix · modeled
56.1%
Latency · seeded
110 ms
Andalucía Solar Lab
eu-west-2 · Spain
Grid intensity · modeled
180.2 g/kWh
Renewable mix · modeled
21.3%
Latency · seeded
150 ms
Karnataka Renewable Bridge
ap-south-1 · India
Grid intensity · modeled
261.8 g/kWh
Renewable mix · modeled
42.5%
Latency · seeded
168 ms
The public endpoint returns its current seeded snapshot, which may change as the snapshot is refreshed. Values in this walkthrough are the captured example.
Pricing snapshot · captured 23 Sep 2026
Free to start. Three tiers, no sales step.
The snapshot below preserves the displayed Free, Pro, and Team prices. See the current pricing route for availability and billing details; hosted checkout depends on Stripe Connect configuration.
Free
Prototype with authenticated API access and modeled region-target estimates.
$0/ month
50,000 calls / month
Prototype estimates are modeled; the physical edge network is in development.
Pro
For shipping teams whose inference volume is part of their sustainability work.
$29/ month
5M calls / month
Prototype estimates are modeled; the physical edge network is in development.
Team
For platform workloads and renewables-sensitive batch experiments.
$99/ month
25M calls / month
Prototype estimates are modeled; the physical edge network is in development.
Team CSV export is on the roadmap, not an available feature. Review the current pricing page.
Workload → pilot
Send a workload, not a meeting request.
Share a few details through the existing contact form. CarbonLayer can use them to discuss a pilot against the seeded prototype fleet; the estimates remain modeled.
Send a workloadFour lines to include
Copy these details into the contact form message.
- 1Modelclaude-sonnet, claude-haiku, cl-compact-v1, or the model label your stack uses.
- 2Tokens per dayA rough estimate is enough to size the modeled carbon and water calculations.
- 3RegionsNote where latency matters, or say “anywhere clean” if it does not.
- 4TriggerDescribe the workload timing, such as nightly batch or interactive requests.
Free access is available at /signup.