Bedrock AgentCore SDK for TypeScript

Deploy AI agents to AWS with VM-level isolation and zero infrastructure

Documentation   |   Samples   |   Python SDK

• Zero infrastructure — No servers to provision, no containers to manage, no scaling to configure.
• Session isolation — Each agent session runs in its own VM. No shared state, no noisy neighbors.
• Long-lived sessions — Sessions persist across requests. Your agent maintains context without external storage.
• Managed tools — Secure code execution and browser automation, ready to use.
• Credential management — Centralized API keys and OAuth tokens, injected at runtime.

The SDK is designed to work with any agent framework. Here's how:

• Runtime — Deploy agents built with any framework. The BedrockAgentCoreApp accepts any request handler — plug in Strands Agents, Vercel AI SDK, LangChain, or your own custom logic.
• Tools — Code Interpreter and Browser tools include built-in integrations for Strands Agents and Vercel AI SDK, so tools can be passed directly to those frameworks' agents. For other frameworks, use the core clients (CodeInterpreter, PlaywrightBrowser) directly and wire them into your own tool definitions.
• Identity — Credential wrappers (withAccessToken, withApiKey) work with any async function, regardless of framework.

npm install bedrock-agentcore @strands-agents/sdk

import { BedrockAgentCoreApp } from 'bedrock-agentcore/runtime'
import { Agent, BedrockModel } from '@strands-agents/sdk'
import { z } from 'zod'

const agent = new Agent({
  model: new BedrockModel({ modelId: 'global.amazon.nova-2-lite-v1:0' }),
})

const app = new BedrockAgentCoreApp({
  invocationHandler: {
    requestSchema: z.object({ prompt: z.string() }),
    process: async function* (request) {
      for await (const event of agent.stream(request.prompt)) {
        if (event.type === 'modelContentBlockDeltaEvent' && event.delta?.type === 'textDelta') {
          yield { event: 'message', data: { text: event.delta.text } }
        }
      }
    },
  },
})

app.run()

Use requestSchema to validate agent entrypoint input. Keep user prompts typed as strings and pass only prompt text to the agent framework.

BedrockAgentCoreApp creates an AgentCore Runtime-compliant server—handling request parsing, streaming responses, and session management for seamless deployment.

Building multi-agent systems? Host agents on AgentCore's A2A protocol path with serveA2A — see the runtime guide.

Give your agent secure code execution with three lines:

import { CodeInterpreterTools } from 'bedrock-agentcore/experimental/code-interpreter/strands'
import { Agent, BedrockModel } from '@strands-agents/sdk'

const codeInterpreter = new CodeInterpreterTools({ region: 'us-east-1' })

const agent = new Agent({
  model: new BedrockModel({ modelId: 'global.amazon.nova-2-lite-v1:0' }),
  tools: codeInterpreter.tools,
})

// Agent can now execute code in a secure sandboxed environment

• Runtime — Secure, session-isolated compute → Examples
• A2A — Agent-to-agent communication on the A2A protocol path → Guide
• Code Interpreter — Execute Python/JS/TS in a sandbox → Examples
• Browser — Cloud-based web automation → Examples
• Web Search — Search the web through a Gateway connector target → Guide
• Identity — Manage API keys and OAuth tokens → Examples
• Memory — Persistent knowledge across sessions → Guide
• Gateway — Transform APIs into MCP tools (coming soon)
• Observability — OpenTelemetry tracing (coming soon)

npm install bedrock-agentcore

Prerequisites: Node.js 20+, AWS credentials, AgentCore access

• Sample Applications — Working examples with deployment templates
• CloudFormation — Infrastructure as code
• AWS CDK — Infrastructure as code
• Deployment Guide — Step-by-step walkthrough

• AgentCore Documentation
• Python SDK

Apache 2.0 — see LICENSE

Contributing · Security