Backend contract

Green compute that stays machine-readable

BatchIn exposes request-level green-compute metadata directly in inference headers and audit rollups. This page mirrors the current backend profile instead of collapsing it into a generic sustainability badge.

Open header referenceOpen audit demo

Configured renewable ratio

92%

Profile value: 9200 bps from the current green-compute config

X-Green-*

X-Green-Region

US clean-power compute regions

X-Green-Energy-Source

renewable

X-Green-Methodology

token-estimate-v1

X-Green-Renewable-Ratio-Bps

9200

X-Green-Energy-MWh

request-specific estimate

X-Green-Carbon-Mg

request-specific estimate

Delivery surface

Where the signal shows up

The current backend emits one profile across request headers, audit rollups, and agent/runtime metadata so operators do not need a separate reporting plane.

Response path

Inference headers stay explicit

  • Chat, completions, embeddings, images, audio, and video expose the base green headers on the live response path.
  • Non-streaming responses can also include per-request energy and carbon estimate headers.
  • Streaming chat keeps the base metadata up front while the estimate headers remain a synchronous-only surface.
Review docs reference

Audit summary

Rollups stay tied to request evidence

  • GET /v1/audit/green/summary aggregates request_count and token_count alongside estimated_energy_kwh and estimated_carbon_kg.
  • The same summary preserves region_scope, energy_source, methodology, and renewable_ratio_bps instead of replacing them with prose.
  • Audit inventory remains user-scoped by default, so green rollups line up with the caller's own verification history.
Open VaaS demo

Runtime handoff

Agents and operators parse the same contract

  • The anonymous playground returns the same routing, cache, and green-compute headers as authenticated inference.
  • The authenticated /mcp/ surface parses region_scope, energy_source, methodology, and renewable_ratio_bps into tool metadata.
  • This keeps REST, SDK, and MCP integrations aligned on one machine-readable signal instead of three separate narratives.
Open docs playbooks

Boundary

Keep the contract narrow enough to trust

Request-level, not a blanket badge

The backend labels the signal with region, energy source, methodology, and renewable ratio. That keeps the estimate boundary visible on every request instead of hiding it behind one static claim.

Methodology is named on purpose

The current profile is token-estimate-v1. Callers can record the methodology identifier together with the request ID and decide how much operational or procurement weight to place on it.

Use it with the surrounding evidence chain

Capture X-Request-Id, the X-Green-* headers, and the audit summary together when rollout, compliance, or sustainability reviewers need the same traceable artifact set.

Rollout use

Carry the same green signal into trust reviews and rollout evidence

Contact rollout teamOpen trust hub
AI Assistant