Building Cloudflare Agents

Your knowledge of the Agents SDK may be outdated. Prefer retrieval over pre-training for any agent-building task.

Retrieval Sources

When to Use

• User wants to build an AI agent or chatbot
• User needs stateful, real-time AI interactions
• User asks about the Cloudflare Agents SDK
• User wants scheduled tasks or background AI work
• User needs WebSocket-based AI communication

Prerequisites

• Cloudflare account with Workers enabled
• Node.js 18+ and npm/pnpm/yarn
• Wrangler CLI (npm install -g wrangler)

Quick Start

bash
1npm create cloudflare@latest -- my-agent --template=cloudflare/agents-starter
2cd my-agent
3npm start

Agent runs at http://localhost:8787

Core Concepts

What is an Agent?

An Agent is a stateful, persistent AI service that:

• Maintains state across requests and reconnections
• Communicates via WebSockets or HTTP
• Runs on Cloudflare's edge via Durable Objects
• Can schedule tasks and call tools
• Scales horizontally (each user/session gets own instance)

Agent Lifecycle

Client connects → Agent.onConnect() → Agent processes messages
                                    → Agent.onMessage()
                                    → Agent.setState() (persists + syncs)
Client disconnects → State persists → Client reconnects → State restored

Basic Agent Structure

typescript
1import { Agent, Connection } from "agents";
2
3interface Env {
4  AI: Ai;  // Workers AI binding
5}
6
7interface State {
8  messages: Array<{ role: string; content: string }>;
9  preferences: Record<string, string>;
10}
11
12export class MyAgent extends Agent<Env, State> {
13  // Initial state for new instances
14  initialState: State = {
15    messages: [],
16    preferences: {},
17  };
18
19  // Called when agent starts or resumes
20  async onStart() {
21    console.log("Agent started with state:", this.state);
22  }
23
24  // Handle WebSocket connections
25  async onConnect(connection: Connection) {
26    connection.send(JSON.stringify({
27      type: "welcome",
28      history: this.state.messages,
29    }));
30  }
31
32  // Handle incoming messages
33  async onMessage(connection: Connection, message: string) {
34    const data = JSON.parse(message);
35
36    if (data.type === "chat") {
37      await this.handleChat(connection, data.content);
38    }
39  }
40
41  // Handle disconnections
42  async onClose(connection: Connection) {
43    console.log("Client disconnected");
44  }
45
46  // React to state changes
47  onStateUpdate(state: State, source: string) {
48    console.log("State updated by:", source);
49  }
50
51  private async handleChat(connection: Connection, userMessage: string) {
52    // Add user message to history
53    const messages = [
54      ...this.state.messages,
55      { role: "user", content: userMessage },
56    ];
57
58    // Call AI
59    const response = await this.env.AI.run("@cf/meta/llama-3-8b-instruct", {
60      messages,
61    });
62
63    // Update state (persists and syncs to all clients)
64    this.setState({
65      ...this.state,
66      messages: [
67        ...messages,
68        { role: "assistant", content: response.response },
69      ],
70    });
71
72    // Send response
73    connection.send(JSON.stringify({
74      type: "response",
75      content: response.response,
76    }));
77  }
78}

Entry Point Configuration

typescript
1// src/index.ts
2import { routeAgentRequest } from "agents";
3import { MyAgent } from "./agent";
4
5export default {
6  async fetch(request: Request, env: Env) {
7    // routeAgentRequest handles routing to /agents/:class/:name
8    return (
9      (await routeAgentRequest(request, env)) ||
10      new Response("Not found", { status: 404 })
11    );
12  },
13};
14
15export { MyAgent };

Clients connect via: wss://my-agent.workers.dev/agents/MyAgent/session-id

Wrangler Configuration

toml
1name = "my-agent"
2main = "src/index.ts"
3compatibility_date = "2024-12-01"
4
5[ai]
6binding = "AI"
7
8[durable_objects]
9bindings = [{ name = "AGENT", class_name = "MyAgent" }]
10
11[[migrations]]
12tag = "v1"
13new_classes = ["MyAgent"]

State Management

Reading State

typescript
1// Current state is always available
2const currentMessages = this.state.messages;
3const userPrefs = this.state.preferences;

Updating State

typescript
1// setState persists AND syncs to all connected clients
2this.setState({
3  ...this.state,
4  messages: [...this.state.messages, newMessage],
5});
6
7// Partial updates work too
8this.setState({
9  preferences: { ...this.state.preferences, theme: "dark" },
10});

SQL Storage

For complex queries, use the embedded SQLite database:

typescript
1// Create tables
2await this.sql`
3  CREATE TABLE IF NOT EXISTS documents (
4    id INTEGER PRIMARY KEY AUTOINCREMENT,
5    title TEXT NOT NULL,
6    content TEXT,
7    created_at DATETIME DEFAULT CURRENT_TIMESTAMP
8  )
9`;
10
11// Insert
12await this.sql`
13  INSERT INTO documents (title, content)
14  VALUES (${title}, ${content})
15`;
16
17// Query
18const docs = await this.sql`
19  SELECT * FROM documents WHERE title LIKE ${`%${search}%`}
20`;

Scheduled Tasks

Agents can schedule future work:

typescript
1async onMessage(connection: Connection, message: string) {
2  const data = JSON.parse(message);
3
4  if (data.type === "schedule_reminder") {
5    // Schedule task for 1 hour from now
6    const { id } = await this.schedule(3600, "sendReminder", {
7      message: data.reminderText,
8      userId: data.userId,
9    });
10
11    connection.send(JSON.stringify({ type: "scheduled", taskId: id }));
12  }
13}
14
15// Called when scheduled task fires
16async sendReminder(data: { message: string; userId: string }) {
17  // Send notification, email, etc.
18  console.log(`Reminder for ${data.userId}: ${data.message}`);
19
20  // Can also update state
21  this.setState({
22    ...this.state,
23    lastReminder: new Date().toISOString(),
24  });
25}

Schedule Options

typescript
1// Delay in seconds
2await this.schedule(60, "taskMethod", { data });
3
4// Specific date
5await this.schedule(new Date("2025-01-01T00:00:00Z"), "taskMethod", { data });
6
7// Cron expression (recurring)
8await this.schedule("0 9 * * *", "dailyTask", {});  // 9 AM daily
9await this.schedule("*/5 * * * *", "everyFiveMinutes", {});  // Every 5 min
10
11// Manage schedules
12const schedules = await this.getSchedules();
13await this.cancelSchedule(taskId);

Chat Agent (AI-Powered)

For chat-focused agents, extend AIChatAgent:

typescript
1import { AIChatAgent } from "agents/ai-chat-agent";
2
3export class ChatBot extends AIChatAgent<Env> {
4  // Called for each user message
5  async onChatMessage(message: string) {
6    const response = await this.env.AI.run("@cf/meta/llama-3-8b-instruct", {
7      messages: [
8        { role: "system", content: "You are a helpful assistant." },
9        ...this.messages,  // Automatic history management
10        { role: "user", content: message },
11      ],
12      stream: true,
13    });
14
15    // Stream response back to client
16    return response;
17  }
18}

Features included:

• Automatic message history
• Resumable streaming (survives disconnects)
• Built-in saveMessages() for persistence

Client Integration

React Hook

tsx
1import { useAgent } from "agents/react";
2
3function Chat() {
4  const { state, send, connected } = useAgent({
5    agent: "my-agent",
6    name: userId,  // Agent instance ID
7  });
8
9  const sendMessage = (text: string) => {
10    send(JSON.stringify({ type: "chat", content: text }));
11  };
12
13  return (
14    <div>
15      {state.messages.map((msg, i) => (
16        <div key={i}>{msg.role}: {msg.content}</div>
17      ))}
18      <input onKeyDown={(e) => e.key === "Enter" && sendMessage(e.target.value)} />
19    </div>
20  );
21}

Vanilla JavaScript

javascript
1const ws = new WebSocket("wss://my-agent.workers.dev/agents/MyAgent/user123");
2
3ws.onopen = () => {
4  console.log("Connected to agent");
5};
6
7ws.onmessage = (event) => {
8  const data = JSON.parse(event.data);
9  console.log("Received:", data);
10};
11
12ws.send(JSON.stringify({ type: "chat", content: "Hello!" }));

Common Patterns

See references/agent-patterns.md for:

• Tool calling and function execution
• Multi-agent orchestration
• RAG (Retrieval Augmented Generation)
• Human-in-the-loop workflows

Deployment

bash
1# Deploy
2npx wrangler deploy
3
4# View logs
5wrangler tail
6
7# Test endpoint
8curl https://my-agent.workers.dev/agents/MyAgent/test-user

Troubleshooting

See references/troubleshooting.md for common issues.

References

• references/examples.md — Official templates and production examples
• references/agent-patterns.md — Advanced patterns
• references/state-patterns.md — State management strategies
• references/troubleshooting.md — Error solutions