# Elyra

> Elyra is a private inference gateway on Elysium. Prompts run inside hardware-attested
> enclaves the operator can't read — the gateway runs in Intel TDX, verifies the GPU's
> attestation before forwarding, and inference executes on NVIDIA Confidential Computing.
> Every response returns with an Ed25519-signed receipt covering the model, the session,
> commitments to the exact request and response, and the bill. Receipt digests are
> anchored on Elysium in Merkle batches every ten minutes, each with an inclusion proof.

## Key facts

- OpenAI-compatible API: base URL `https://api.elyra.technology/v1` — chat completions and streaming, same SDKs.
- A receipt per response: named in the `x-receipt-id` response header. Change one character of the request or response and verification fails.
- Five hops, no trusted middleman: Agent (SDK attestation check) → Gateway (Intel TDX) → GPU enclave (NVIDIA CC) → Receipt (Ed25519) → Elysium (Merkle anchor).
- Suppliers: Phala (GPU TEEs), NEAR AI, Chutes (on Bittensor), RedPill.
- Open weights, list prices: every endpoint is billed to the micro-dollar on the receipt.

## Links

- Site: https://elyra.technology/
- App: https://app.elyra.technology/
- Chat completions: https://api.elyra.technology/v1/chat/completions
- llms.txt: https://elyra.technology/llms.txt