LangGraph JS Minimal Example

A minimal LangGraph JavaScript agent deployed to AWS Bedrock AgentCore using Serverless Framework.

Features

LangGraph JS: ReAct agent pattern with tool calling
Claude Sonnet 5: Powered by Amazon Bedrock
Simple Tools: Calculator operations and time queries
Docker Deployment: Auto-detected Dockerfile for easy deployment

Quick Start

Prerequisites

Node.js 20+
Docker (for local development)
AWS credentials configured
Serverless Framework CLI (npm install -g serverless)

Install Dependencies

npm install

Local Development

Start the agent locally with hot reload:

sls dev

This will:

Build the Docker image
Start the container with AWS credentials injected
Open an interactive chat interface
Watch for file changes and auto-rebuild

Try these prompts:

"What time is it?"
"What time is it in Tokyo?"
"Calculate 25 multiplied by 4"
"Add 100 and 250, then divide the result by 7"

Deploy to AWS

sls deploy

Invoke Deployed Agent

# Via Serverless Framework CLI
sls invoke --agent assistant --data '{"prompt":"Hello! What can you help me with?"}'

# Plain string is also supported:
# sls invoke --agent assistant -d "Hello! What can you help me with?"

# Or via curl (replace URL with your runtime URL from `sls info`)
curl -X POST https://your-runtime-url/invoke \
  -H "Content-Type: application/json" \
  -d '{"prompt":"Hello! What can you help me with?"}'

Remove

sls remove

Project Structure

langgraph-basic-dockerfile/
├── serverless.yml    # Serverless Framework configuration
├── agent.js          # LangGraph JS agent with tools
├── package.json      # npm dependencies
├── Dockerfile        # Container definition
└── README.md         # This file

How It Works

Agent Architecture

User Input
    │
    ▼
┌─────────────────┐
│  BedrockAgent   │
│    CoreApp      │◄─── HTTP Server (port 8080)
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│   LangGraph     │
│  ReAct Agent    │◄─── Alternates between LLM and tools
└────────┬────────┘
         │
    ┌────┴────┐
    ▼         ▼
┌───────┐ ┌───────┐
│ Tools │ │ Claude│
│       │ │  LLM  │
└───────┘ └───────┘

Tools Available

Tool	Description
`get_current_time`	Get current date/time with optional timezone
`add`	Add two numbers
`multiply`	Multiply two numbers
`divide`	Divide two numbers

Configuration

The serverless.yml is intentionally minimal:

service: langgraph-basic-dockerfile

provider:
  name: aws

ai:
  agents:
    assistant: {}

The Dockerfile is auto-detected, and default settings are applied:

Protocol: HTTP
Network: PUBLIC
Port: 8080

Customization

Adding New Tools

Edit agent.js to add new tools:

const myNewTool = tool(
  async ({ input }) => {
    // Tool implementation
    return `Result: ${input}`
  },
  {
    name: 'my_new_tool',
    description: 'Description for the LLM',
    schema: z.object({
      input: z.string().describe('Input parameter'),
    }),
  },
)

// Add to tools array
const tools = [getCurrentTime, add, multiply, divide, myNewTool]

Changing the Model

The default model is global.anthropic.claude-sonnet-5, read from the MODEL_ID environment variable in agent.js. Override it via the MODEL_ID env var, or edit the fallback in agent.js:

const MODEL_ID = process.env.MODEL_ID ?? 'global.anthropic.claude-sonnet-5'

const model = new ChatBedrockConverse({
  model: MODEL_ID,
  // Or use other models:
  // model: 'us.amazon.nova-2-lite-v1:0',
  // model: 'us.meta.llama3-70b-instruct-v1:0',
  region: process.env.AWS_REGION || 'us-east-1',
})

langgraph-basic-docker - Python version
langgraph-gateway - Lambda functions as tools

LangGraph JS Minimal Example

A minimal LangGraph JavaScript agent deployed to AWS Bedrock AgentCore using Serverless Framework.

Features

LangGraph JS: ReAct agent pattern with tool calling
Claude Sonnet 5: Powered by Amazon Bedrock
Simple Tools: Calculator operations and time queries
Docker Deployment: Auto-detected Dockerfile for easy deployment

Quick Start

Prerequisites

Node.js 20+
Docker (for local development)
AWS credentials configured
Serverless Framework CLI (npm install -g serverless)

Install Dependencies

npm install

Local Development

Start the agent locally with hot reload:

sls dev

This will:

Build the Docker image
Start the container with AWS credentials injected
Open an interactive chat interface
Watch for file changes and auto-rebuild

Try these prompts:

"What time is it?"
"What time is it in Tokyo?"
"Calculate 25 multiplied by 4"
"Add 100 and 250, then divide the result by 7"

Deploy to AWS

sls deploy

Invoke Deployed Agent

# Via Serverless Framework CLI
sls invoke --agent assistant --data '{"prompt":"Hello! What can you help me with?"}'

# Plain string is also supported:
# sls invoke --agent assistant -d "Hello! What can you help me with?"

# Or via curl (replace URL with your runtime URL from `sls info`)
curl -X POST https://your-runtime-url/invoke \
  -H "Content-Type: application/json" \
  -d '{"prompt":"Hello! What can you help me with?"}'

Remove

sls remove

Project Structure

langgraph-basic-dockerfile/
├── serverless.yml    # Serverless Framework configuration
├── agent.js          # LangGraph JS agent with tools
├── package.json      # npm dependencies
├── Dockerfile        # Container definition
└── README.md         # This file

How It Works

Agent Architecture

User Input
    │
    ▼
┌─────────────────┐
│  BedrockAgent   │
│    CoreApp      │◄─── HTTP Server (port 8080)
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│   LangGraph     │
│  ReAct Agent    │◄─── Alternates between LLM and tools
└────────┬────────┘
         │
    ┌────┴────┐
    ▼         ▼
┌───────┐ ┌───────┐
│ Tools │ │ Claude│
│       │ │  LLM  │
└───────┘ └───────┘

Tools Available

Tool	Description
`get_current_time`	Get current date/time with optional timezone
`add`	Add two numbers
`multiply`	Multiply two numbers
`divide`	Divide two numbers

Configuration

The serverless.yml is intentionally minimal:

service: langgraph-basic-dockerfile

provider:
  name: aws

ai:
  agents:
    assistant: {}

The Dockerfile is auto-detected, and default settings are applied:

Protocol: HTTP
Network: PUBLIC
Port: 8080

Customization

Adding New Tools

Edit agent.js to add new tools:

const myNewTool = tool(
  async ({ input }) => {
    // Tool implementation
    return `Result: ${input}`
  },
  {
    name: 'my_new_tool',
    description: 'Description for the LLM',
    schema: z.object({
      input: z.string().describe('Input parameter'),
    }),
  },
)

// Add to tools array
const tools = [getCurrentTime, add, multiply, divide, myNewTool]

Changing the Model

The default model is global.anthropic.claude-sonnet-5, read from the MODEL_ID environment variable in agent.js. Override it via the MODEL_ID env var, or edit the fallback in agent.js:

const MODEL_ID = process.env.MODEL_ID ?? 'global.anthropic.claude-sonnet-5'

const model = new ChatBedrockConverse({
  model: MODEL_ID,
  // Or use other models:
  // model: 'us.amazon.nova-2-lite-v1:0',
  // model: 'us.meta.llama3-70b-instruct-v1:0',
  region: process.env.AWS_REGION || 'us-east-1',
})

langgraph-basic-docker - Python version
langgraph-gateway - Lambda functions as tools