OpenGradient TypeScript SDK

A TypeScript/JavaScript SDK for performing LLM chat and completion via OpenGradient's TEE (Trusted Execution Environment) with x402 payment protocol support.

Installation

npm install opengradient-sdk

Requirements

• Node.js 18+ (for global fetch)
• A funded EVM wallet on Base (settlement happens in OPG on the Base network via x402)

Quick Start

import { Client, TEE_LLM } from "opengradient-sdk";

const client = new Client({
  privateKey: process.env.PRIVATE_KEY!, // EVM private key (with or without 0x prefix)
});

// Non-streaming chat
const result = await client.llm.chat({
  model: TEE_LLM.CLAUDE_HAIKU_4_5,
  messages: [{ role: "user", content: "Hello!" }],
  maxTokens: 100,
});
console.log(result.chatOutput?.content);
console.log("payment hash:", result.paymentHash);

Streaming chat

import { Client, TEE_LLM } from "opengradient-sdk";

const client = new Client({ privateKey: process.env.PRIVATE_KEY! });

const stream = client.llm.chat({
  model: TEE_LLM.CLAUDE_HAIKU_4_5,
  messages: [{ role: "user", content: "Stream me a haiku." }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta.content ?? "");
}

Tool / function calling

const result = await client.llm.chat({
  model: TEE_LLM.GPT_5,
  messages: [{ role: "user", content: "What's the weather in Paris?" }],
  tools: [
    {
      type: "function",
      function: {
        name: "get_weather",
        description: "Get current weather for a city",
        parameters: {
          type: "object",
          properties: { city: { type: "string" } },
          required: ["city"],
        },
      },
    },
  ],
});
console.log(result.chatOutput?.tool_calls);

Completion

const result = await client.llm.completion({
  model: TEE_LLM.CLAUDE_HAIKU_4_5,
  prompt: "The capital of France is",
  maxTokens: 20,
});
console.log(result.completionOutput);

OPG Token Approval

Before making LLM requests, your wallet must approve OPG token spending via the Permit2 protocol. ensureOpgApproval only sends an on-chain transaction when the current allowance drops below the threshold, so it's safe to call on every server startup:

import { privateKeyToAccount } from "viem/accounts";
import { ensureOpgApproval } from "opengradient-sdk";

const account = privateKeyToAccount(process.env.PRIVATE_KEY as `0x${string}`);

// Only sends a tx when allowance < 5 OPG, then approves 100 OPG so
// subsequent restarts are free. Defaults approveAmount to 2 * minAllowance.
const result = await ensureOpgApproval(account, 5, 100);
console.log("allowance after:", result.allowanceAfter, "tx:", result.txHash);

The wallet must hold OPG on Base mainnet. Override the RPC with the BASE_MAINNET_RPC environment variable if you don't want to use the default public node.

End-to-end example

import { privateKeyToAccount } from "viem/accounts";
import { Client, TEE_LLM, ensureOpgApproval } from "opengradient-sdk";

async function main() {
  const privateKey = process.env.PRIVATE_KEY as `0x${string}`;

  // 1. Make sure the wallet has approved Permit2 to spend OPG.
  //    No-op when the allowance is already above the threshold.
  const account = privateKeyToAccount(privateKey);
  await ensureOpgApproval(account, 5, 100);

  // 2. Run a TEE-secured chat completion settled in OPG via x402.
  const client = new Client({ privateKey });
  try {
    const result = await client.llm.chat({
      model: TEE_LLM.CLAUDE_3_5_HAIKU,
      messages: [{ role: "user", content: "Hello!" }],
      maxTokens: 100,
    });
    console.log(result.chatOutput?.content);
    console.log("payment hash:", result.paymentHash);
  } finally {
    await client.close();
  }
}

main();

x402 Settlement Modes

import { X402SettlementMode } from "opengradient-sdk";

await client.llm.chat({
  model: TEE_LLM.GPT_5,
  messages: [{ role: "user", content: "Hi" }],
  x402SettlementMode: X402SettlementMode.BATCH_HASHED, // default
});

• PRIVATE — payment-only settlement; no input/output hashes posted on-chain (most privacy-preserving).
• BATCH_HASHED — aggregates multiple inferences into a single Merkle-tree settlement of input/output hashes (most cost-efficient, default).
• INDIVIDUAL_FULL — records full input data, output data, and verification on-chain per request (highest gas, maximum auditability).

Development

npm install      # install deps
npm run lint     # ESLint over src/
npm test         # Jest unit tests
npm run build    # tsc → dist/
npm run format   # prettier --write

CI runs lint, test, and build on Node 18 and 20 — see .github/workflows/ci.yml.

Available models

See TEE_LLM in src/types.ts for the full list. Highlights:

• OpenAI: GPT_5, GPT_5_MINI, GPT_5_2, GPT_5_4, GPT_5_4_MINI, GPT_5_4_NANO, GPT_5_5, GPT_4_1_2025_04_14, GPT_4_1_MINI, GPT_4_1_NANO, O3, O4_MINI
• Anthropic: CLAUDE_HAIKU_4_5, CLAUDE_SONNET_4_5, CLAUDE_SONNET_4_6, CLAUDE_OPUS_4_5, CLAUDE_OPUS_4_6, CLAUDE_OPUS_4_7
• Google: GEMINI_3_FLASH, GEMINI_3_1_PRO_PREVIEW, GEMINI_3_1_FLASH_LITE_PREVIEW, GEMINI_2_5_FLASH, GEMINI_2_5_PRO, GEMINI_2_5_FLASH_LITE
• xAI: GROK_4, GROK_4_FAST, GROK_4_1_FAST, GROK_4_1_FAST_NON_REASONING, GROK_4_20_REASONING, GROK_4_20_NON_REASONING, GROK_CODE_FAST_1
• ByteDance: SEED_1_6, SEED_1_8, SEED_2_0_LITE