ReAct agent

A tool-calling loop on any LightChain model - no native function-calling needed. The model thinks, calls a tool, reads the observation, repeats. Run it LIVE with your wallet below (each reasoning step is one user-paid inference), or copy the Node setup.

NetworkMainnetchain 9200

Thought

Tool call

Observation

Answer

loop until done, capped at maxIterations

Build the agent

TaskThe goal the agent reasons toward.

Tools

Toggle the tools the model may call. Add your own in code.

Max iterations4 steps maxHard cap on reasoning steps (1-8). Each step is one user-paid inference.

Reasoning trace

Run the agent to watch it think, call tools, and answer - live.

The live agent drives the SDK's own Agent loop with your wallet as the inference backend - the same code below, in the browser.

The same agent in Node

TypeScript

import { Agent } from "lightnode-sdk";

const agent = new Agent({
  network: "mainnet",
  privateKey: process.env.PRIVATE_KEY,
  maxIterations: 4,
  tools: [
    {
      name: "add",
      description: "Add two numbers",
      args: { a: "number", b: "number" },
      handler: ({ a, b }) => Number(a) + Number(b),
    },
    {
      name: "now",
      description: "Current ISO timestamp",
      args: {},
      handler: () => new Date().toISOString(),
    },
  ],
});

const { answer, steps } = await agent.run("What is 21 + 21, and what time is it now?");
console.log(answer);
console.log(steps.map((s) => s.kind)); // thought / tool_call / answer

Scaffold it into your project: npx lightnode add agent.