Build a Multi-Agent System

Use app.agents.create to create multiple specialized agents, then coordinate their runs from your server.

import { AppClient } from "@uclaw/sdk";

const app = new AppClient({ apiKey: process.env.UCLAW_API_KEY });

const researcher = await app.agents.create({
  title: "Researcher",
  config: {
    instructions: "Research the user's request and produce concise notes.",
    modelTier: "fast",
  },
});

const writer = await app.agents.create({
  title: "Writer",
  config: {
    instructions: "Turn research notes into a polished answer.",
    modelTier: "capable",
  },
});

Run Agents in Sequence

async function collectText(agent: typeof researcher, input: string) {
  const run = await agent.run(input);
  await run.wait({ until: "running", timeoutMs: 60_000 });

  let text = "";
  for await (const event of run.stream()) {
    if (event.type === "text-delta") text += event.delta;
  }
  return text;
}

const notes = await collectText(researcher, "Research durable agent runtimes.");
const answer = await collectText(writer, `Write an answer from these notes:\n${notes}`);

console.log(answer);

Clean Up

await app.agents.delete(researcher.id);
await app.agents.delete(writer.id);

For agent-to-agent delegation inside a single run, UClaw's runtime also includes sub-agent capabilities. Use explicit server orchestration first; introduce sub-agents when you need nested autonomous work.