1. Your first agent

You're building support for a small shop. This first step is the smallest thing that works: an agent with instructions, no tools yet, answering one question.

Write the agent

An agent needs a name, instructions and a model. scriptedModel (scripted_model in Python) plays back fixed responses in order, so the program runs with no API key and no network — the same model the framework's own tests use.

sqlite(":memory:") is the store: the append-only log this run is written to. In memory it lasts for the process; give it a path and it lasts on disk.

import { agent, scriptedModel, sqlite } from "@threads/core";

const support = agent({
  name: "support",
  instructions: "You answer refund questions. Quote the 30-day refund window.",
  model: scriptedModel({
    responses: [
      {
        content: [{ type: "text", text: "Orders can be returned within 30 days of delivery." }],
        stop_reason: "end_turn",
        usage: { input_tokens: 10, output_tokens: 2 },
      },
    ],
  }),
});

const result = await support.run("What is your refund policy?", { store: sqlite(":memory:") });
if (result.status === "completed") console.log(result.output);
else console.log(result.status);

Run it with bun support.ts or uv run python support.py:

Orders can be returned within 30 days of delivery.

Check the status, not an exception

run() doesn't throw when a run ends early. It returns a result keyed on status, and completed is only one of the answers — a run can also park for an approval, run out of budget, be cancelled, fail, or hand off. That's why the program above prints result.status when it isn't completed, and it is the reason the next tutorial has something to check.

The full list is in Quickstart.

Point it at a real model

Swap the scripted model for a provider and nothing else changes. The key is read from ANTHROPIC_API_KEY, and a store with a path keeps the log on disk so you can read, resume and fork the run later.

import { anthropic } from "@threads/anthropic";

const support = agent({
  name: "support",
  instructions: "You answer refund questions. Quote the 30-day refund window.",
  model: anthropic("claude-sonnet-5"),
});

const result = await support.run("What is your refund policy?", { store: sqlite(".threads") });

Keep the scripted version around. Every tutorial after this one uses it, because a scripted run is free, fast and gives the same answer every time — which is what makes the outputs on these pages checkable.

Next