prem-rs

Hardware attestation SDK for JavaScript and TypeScript. Cryptographically verifies AMD SEV-SNP, Intel TDX, and NVIDIA GPU attestations from an attestation server — runs in Node.js, Bun, Deno and the browser via WASM.

Install

npm install @premai/prem-rs

Quick start

import { ClientBuilder } from "@premai/prem-rs";

const client = new ClientBuilder("https://attestation.example.com")
  .build();

// End-to-end attestation: discovers modules, generates nonces,
// verifies cryptographic signatures and certificate chains
const result = await client.attest();

console.log("CPU:", result.modules().cpu());   // CpuModule.Sev or CpuModule.Tdx
console.log("GPU:", result.modules().gpu());   // GpuModule.Nvidia or undefined

API

ClientBuilder

Creates and configures a Client instance.

const client = new ClientBuilder(url)
  .with_authorization("Bearer <token>")  // optional: set Authorization header
  .with_kds(new Kds(kdsUrl))             // optional: custom AMD KDS cache
  .with_pcs(new Pcs(pcsUrl))             // optional: custom Intel PCS cache
  .build();

Client

High-level (recommended)

These methods handle nonce generation, request, and full cryptographic verification in a single call:

| Method | Description | |---|---| | client.attest(query?) | Full attestation of all available modules (CPU + GPU) | | client.attest_sev(query?) | AMD SEV-SNP attestation only | | client.attest_tdx(query?) | Intel TDX attestation only | | client.attest_nvidia(query?) | NVIDIA GPU attestation only |

Low-level

These methods fetch and parse attestation evidence (discouraged) — useful when you need to inspect raw data or implement custom validation:

| Method | Description | |---|---| | client.request_modules(query?) | List available attestation modules on the server | | client.request_sev(nonce, query) | Fetch raw SEV-SNP attestation report | | client.request_tdx(nonce, query) | Fetch raw TDX quote | | client.request_nvidia(nonce, query) | Fetch raw NVIDIA EAT/JWT token |

QueryParams

Pass custom query parameters to the attestation server. The nonce key is reserved and will throw if used.

const query = new QueryParams()
  .with("model", "my-model")
  .with("version", "1.0");

await client.attest(query);

Sub-module access

Lower-level types from the attestation modules are available under namespaces:

import { nvidia, sev } from "@premai/prem-rs";

// NVIDIA: manual token parsing and verification
const keychain = await nvidia.fetch_keychain();
const token = nvidia.EATToken.parse(rawJwt);
const claims = token.verify(keychain);
claims.validate(nonce);

// AMD SEV-SNP: manual certificate chain verification
const kds = new sev.Kds("https://kcds.prem.io");
const chain = await kds.fetch_certificates(attestation);
attestation.verify(chain, nonce);

Memory management

WASM objects are not garbage-collected automatically. Call .free() when done, or use using (TypeScript 5.2+) for automatic disposal:

using client = new ClientBuilder(url).build();
using result = await client.attest();
// automatically freed at end of scope

Examples

• Bun — minimal CLI attestation
• Vite — browser UI with attestation status

License

See LICENSE.