Skip to content

THE ORCHESTRATION LAYER FOR GPU INFERENCE

One API. Every frontier model.20% under list.

Change one base URL and keep your code. Graphene routes every request across owned, partner and spot GPU capacity — metering, billing and residency policy handled underneath. Billed per token. No subscriptions, no seats.

The thinnest layer. The biggest yield.

One API. Every frontier model.

Claude Opus 4.8GPT-5.5Llama 4Gemini 3Mistral LargeDeepSeek R2

Browse models →

Graphene at a glance

Under every provider's list price
20%
Usage-based billing, no seats
Per-token
To get started
$0
Base URL change to migrate
1 line

Everything you need to ship inference

  • Policy-based routing

    Set routing rules and quality floors; every request stays on the models and lanes you allow, across owned, partner and spot GPU capacity.

  • Per-token metering and budgets

    Usage is metered per token with budget controls, so spend is visible on every request before you commit traffic at scale.

  • Workspace governance

    Roles, audit surfaces and compliance profiles keep inference accountable across your whole workspace.

  • Sovereign routing profiles

    Residency-aware routing profiles are on the roadmap for workloads that must stay in-region — configuration, not a separate stack.

Call the API in one cURL

Replace $PORTAL_ORIGIN, <workspace_id>, and <YOUR_API_KEY> with your workspace values — no secrets ship in this example.

curl -sS "$PORTAL_ORIGIN/v1/workspaces/<workspace_id>/completions" \
  -H "Authorization: Bearer <YOUR_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{"model":"public-catalogue-default","messages":[{"role":"user","content":"Hello"}]}'

Where to next

  • Own GPUs?

    Turn idle capacity into inference revenue with a 60-day, no-exclusivity pilot.

    Explore supply →
  • Building agents?

    Machine-readable pricing, capability manifest and x402-ready settlement rails are on the roadmap. Early access for agent builders.

Get started

Create a workspace, issue your first API key, and route inference through Graphene with the same contracts you see in the docs.