Browser Agent

Overview

This tutorial demonstrates how to run a Strands Agent entirely in the browser using the TypeScript SDK with Vite as the build tool. The example creates a simple chat interface where users can interact with an agent directly from their web browser.

Browser Agent Architecture

Feature	Description
Agent Structure	Single agent architecture
Architecture	Client-side only (browser)
Build Tool	Vite
Model Providers	Amazon Bedrock (Claude Sonnet 4.6), OpenAI (GPT-4o)

Prerequisites

Node.js 22 or later (the version @strands-agents/sdk supports)
One of the following:
- AWS credentials with Amazon Bedrock access (Access Key ID, Secret Access Key, and optionally Session Token for temporary credentials)
- OpenAI API key
Basic TypeScript and web development knowledge

Project Structure

03-browser-agent/
├── images/
│   └── browser_agent.png  # Architecture diagram
├── index.html             # Entry HTML with chat UI
├── src/
│   ├── main.ts           # Agent setup and DOM interaction
│   ├── style.css         # Chat interface styling
│   └── vite-env.d.ts     # Vite type declarations
├── package.json          # Dependencies
├── tsconfig.json         # TypeScript configuration
├── vite.config.ts        # Vite configuration
└── README.md             # This file

Running the Example

cd typescript/01-learn/06-browser-agent
npm install
npm run dev

This will start the Vite dev server and open http://localhost:5173 in your browser.

Credentials Form

Key Concepts

Browser-Based Agent

The agent runs entirely in the browser, which means:

No backend required - API calls go directly from the browser to the model provider (development only - see Security Considerations)
Simple deployment - Can be hosted on any static file server (do not bundle credentials)

Agent Configuration

In browser environments, credentials must be passed explicitly. The application supports two model providers:

Amazon Bedrock

import { Agent, BedrockModel } from "@strands-agents/sdk";

const agent = new Agent({
  model: new BedrockModel({
    region: "us-east-1",
    clientConfig: {
      credentials: {
        accessKeyId: "...",
        secretAccessKey: "...",
        sessionToken: "...",  // Optional, for temporary credentials
      },
    },
  }),
  systemPrompt: "You are a helpful assistant running in the browser."
});

OpenAI

import { Agent } from "@strands-agents/sdk";
import { OpenAIModel } from "@strands-agents/sdk/models/openai";

const agent = new Agent({
  model: new OpenAIModel({
    apiKey: "sk-...",
    modelId: "gpt-4o",
    // Required to allow OpenAI SDK to run in browser environments
    clientConfig: {
      dangerouslyAllowBrowser: true,
    },
  }),
  systemPrompt: "You are a helpful assistant running in the browser."
});

Streaming Responses

Streaming displays the response progressively as it’s generated, providing immediate feedback:

for await (const event of agent.stream(userMessage)) {
  if (
    event.type === "modelStreamUpdateEvent" &&
    event.event.type === "modelContentBlockDeltaEvent" &&
    event.event.delta.type === "textDelta"
  ) {
    responseText += event.event.delta.text;
    // Update UI with each chunk
  }
}

Chat Interface

Vite

Vite is a build tool that enables browser-based development:

TypeScript compilation - Compiles .ts files to JavaScript since browsers can’t run TypeScript directly
Module bundling - Resolves and bundles dependencies (like @strands-agents/sdk) so browsers can load them
Dev server - Serves files locally with automatic reload on save
Production builds - Creates optimized, minified files for deployment

Security Considerations

When running agents in the browser, be aware of:

Credentials Exposure - AWS credentials and OpenAI API keys used in browser code are visible in the network tab and browser developer tools
For Production - Use a backend proxy or secure token-based authentication
Development Only - This tutorial is intended for development and demonstration purposes

Building for Production

npm run build

This creates an optimized build in the dist/ directory that can be deployed to any static hosting service.