Skip to main content
@singularity-layer/grid is the official TypeScript/JavaScript client for SGL Grid. Typed methods over the grid’s HTTP API; zero dependencies; native fetch (Node ≥ 18, modern browsers, edge runtimes).
npm@singularity-layer/grid
Sourcegithub.com/Singularity-Layer/sgl-network-sdk-ts
LicenseMIT

​
Install

npm install @singularity-layer/grid
# pnpm add @singularity-layer/grid · yarn add @singularity-layer/grid

​
Quick start

import { GridClient } from "@singularity-layer/grid";

const grid = new GridClient();              // public reads need no key

const res = await grid.chatCompletions({
  model: "<model-id>",                       // see grid.models()
  messages: [{ role: "user", content: "What is confidential compute?" }],
});

console.log(res.choices[0].message.content);

​
Configuration

const grid = new GridClient({
  apiKey: "scg_...",                          // required for job submission / billed calls
  baseUrl: "https://grid.x402compute.cc",     // override the orchestrator URL
  timeout: 30_000,                            // request timeout in ms (default 60000)
});
  • No key → public reads (capacity, models, pricing) work.
  • API key → required for submitJob, getJob, getAttestation, and billed chat. Create keys in the dashboard (see Billing).

​
Discovery (no auth)

const cap = await grid.capacity();
console.log(`${cap.active_nodes} nodes online`);

const models = await grid.models();
models.forEach((m) => console.log(`${m.id} — ${m.sgl_node_count} nodes`));

const pricing = await grid.pricing();
pricing.forEach((p) => console.log(`${p.model}: $${p.price_per_1k_input_tokens_usd}/1k input`));

​
Chat completions

const grid = new GridClient({ apiKey: "scg_..." });

const res = await grid.chatCompletions({
  model: "<model-id>",
  messages: [
    { role: "system", content: "You are concise." },
    { role: "user", content: "Summarize TEEs in one sentence." },
  ],
  max_tokens: 256,
  temperature: 0.7,
});

​
Jobs + attestation

const grid = new GridClient({ apiKey: "scg_..." });

const job = await grid.submitJob("<model-id>", {
  messages: [{ role: "user", content: "Summarize quantum computing" }],
});

const result = await grid.getJob(job.job_id);
if (result.status === "completed") console.log(result.result);

const att = await grid.getAttestation(job.job_id);   // verify the enclave
console.log(`Verified: ${att.verified}, TEE: ${att.tee_type}`);
See Confidential inference for what the attestation proves.

​
Error handling

import {
  GridClient, SGLAPIError, SGLAuthError, SGLConnectionError, SGLNotFoundError,
} from "@singularity-layer/grid";

try {
  await grid.getJob("nonexistent");
} catch (err) {
  if (err instanceof SGLNotFoundError) { /* 404 */ }
  else if (err instanceof SGLAuthError) { /* 401/403 — bad/missing key */ }
  else if (err instanceof SGLConnectionError) { /* unreachable / timeout */ }
  else if (err instanceof SGLAPIError) { console.log(err.statusCode, err.message); }
}

​
Or use the OpenAI SDK

The grid is OpenAI-compatible — skip this client if you just want drop-in chat: 
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://grid.x402compute.cc/v1", apiKey: "scg_..." });
Use @singularity-layer/grid for grid-native features (capacity, pricing, jobs, attestation). See the API guide and the Python SDK.