Skip to main content
SGL Grid is decentralized, confidential AI inference — think “OpenRouter, but decentralized and private.” Models are served by independent, hardware-attested TEE nodes; you call them through an OpenAI-compatible API and pay per request.

​
Why the Grid

  • OpenAI-compatible — /v1/chat/completions; existing SDKs work by changing the base URL.
  • Confidential — prompts are end-to-end encrypted and run inside a TEE; responses carry an attestation. See Confidential inference.
  • Pay-per-call — credits or x402 USDC, no subscription. See Pricing.
  • Decentralized — served by staked operators, not one provider.

​
How a request flows

1

Pick a model

Browse models served on the network.
2

Authenticate

API key, session, or per-request signature — see Wallet & auth.
3

Send

Call the OpenAI-compatible endpoint (or use the playground).
4

Get a verifiable, private reply

The job runs in a TEE; the response is sealed to you and tagged with the node’s attestation.

​
In this section

Models

What’s served, and how to choose.

Playground

Try models in the browser.

API

OpenAI-compatible quickstart.

Confidential inference

E2E encryption + attestation.

Pricing

How cost is calculated.