In Cloudflare Agents, a TypeScript class defines behavior; Wrangler binds that class to Cloudflare resources. With Think, methods select the model, prompt, tools and limits. Each named instance can also keep its own configuration in SQLite.

An agent definition has two parts: application code and deployment configuration. A class extending Agent uses the runtime directly. A class extending Think also gets a model/tool loop and chat behavior.

Think uses overrides such as getModel(), getSystemPrompt(), getTools(), configureSession() and maxSteps. Per-instance data can be persisted with configure<T>() and read with getConfig<T>(); methods can consult it to vary behavior.

This illustrative TypeScript sketch exposes one simple tool:

import { Think } from "@cloudflare/think";
import { createWorkersAI } from "workers-ai-provider";
import { tool } from "ai";
import { z } from "zod";

export class ResearchAgent extends Think<Env> {
  maxSteps = 4;

  getModel() {
    return createWorkersAI({ binding: this.env.AI })(
      "@cf/moonshotai/kimi-k2.6"
    );
  }

  getSystemPrompt() {
    return "Explain findings clearly. Use tools when needed.";
  }

  getTools() {
    return {
      utcTime: tool({
        description: "Read the current UTC time",
        inputSchema: z.object({}),
        execute: async () => new Date().toISOString()
      })
    };
  }

  beforeTurn() {
    return { activeTools: ["utcTime"] };
  }
}

Env is generated from Wrangler bindings. getTools() adds custom tools to Think's existing tools. activeTools narrows what the model can call in this example.

Wrangler declares the exported class, SQLite storage and resources such as the AI binding. The current configuration guide uses exports to declare SQLite storage; older examples use class migrations. This sketch is based on documentation and has not been compiled or deployed here.

Sources: Think configuration, custom tools, turn hooks, Wrangler configuration.