Amazon Nova Act Hands-On POC (Part 2)

Amazon Nova Act : Hands-On POC (Part 2)

Objective of This SOP

By following this SOP, you will create a basic action-oriented AI workflow using Amazon Nova Act that:

  • Reads CPU usage
  • Checks a condition (CPU > 80%)
  • Executes an action using AWS Lambda
  • Returns a result

This is the foundation of Nova Act–style AI.

Prerequisites

Before starting, ensure you have:

  • An AWS account
  • Access to AWS Console
  • Region set to us-east-1
  • Create the EC2 instance and use this instance id for lambda
  • Basic familiarity with AWS navigation

STEP 1: Log in to AWS Console

  1. Open browser
  2. Go to: https://console.aws.amazon.com
  3. Log in with your AWS credentials
  4. Set region (top-right) to us-east-1

STEP 2: Enable Amazon Bedrock (AI Intelligence)

Purpose:

This provides the “thinking capability” for the AI.

Steps:

  1. In AWS search bar, type Bedrock
  2. Open Amazon Bedrock
  3. From left menu, click Model access
  4. Auto Enable:

5. Go from left menu, click Model Catalog

6. Click on Nova 2 Lite

Click on the yellow button: Open in Playground

7. Type the message “hello” to test whether Nova 2 Lite is working or not.

It’s Working.

STEP 3: Create IAM Role (Permissions for Actions)

Purpose:

Nova Act and lambda requires permission to interact with AWS services.

1. Search for IAM

2. Click Roles → Create role

3. Select:

  • Trusted entity: AWS service
  • Use case: Lambda

4. Click Next

Add Permissions:

  • AWSLambdaBasicExecutionRole
  • CloudWatchReadOnlyAccess
  • cloudwatch+ec2
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "cloudwatch:GetMetricData",
        "cloudwatch:GetMetricStatistics",
        "cloudwatch:ListMetrics",
        "ec2:DescribeInstances",
        "ec2:RebootInstances"
      ],
      "Resource": "*"
    }
  ]
}

Role name:

lambda-execution

Click Create role

AI now has permission to perform actions.

STEP 4 – Open the Lambda Console

  • In the AWS Console search bar, type Lambda
  • Click on Lambda
  • Click Create function

Function Create Settings

Select exactly the following options:

Function name: get_cpu_utilization

Runtime:

Python 3.12

Execution role:

Use an existing role

Role name: lambda-execution

Now click on Create function.

Wait for a few seconds while the function is being created.

When Lambda Opens – Go to the Code Tab

As soon as the Lambda function opens, you will see the Code tab.

There will be some default code already present.

Delete the default code completely

Copy–paste the full code below:

import json
import boto3
from datetime import datetime, timedelta, timezone

cw = boto3.client("cloudwatch")

def _extract_from_bedrock_event(event):
    """
    Bedrock Agent action-group (OpenAPI schema) sends:
    event["requestBody"]["content"]["application/json"]["properties"] = [{name,type,value},...]
    """

    instance_id = None
    minutes = 5

    rb = (
        event.get("requestBody", {})
             .get("content", {})
             .get("application/json", {})
             .get("properties", [])
    )

    for p in rb:
        if p.get("name") == "instanceId":
            instance_id = p.get("value")

        if p.get("name") == "minutes" and p.get("value") is not None:
            try:
                minutes = int(p.get("value"))
            except:
                pass

    return instance_id, minutes


def lambda_handler(event, context):

    # Handle both Bedrock Agent events and manual Lambda test events

    if event.get("messageVersion") == "1.0":
        instance_id, minutes = _extract_from_bedrock_event(event)
    else:
        instance_id = event.get("instanceId")
        minutes = int(event.get("minutes", 5))

    cpu_avg = None
    datapoints = 0

    if instance_id:

        end = datetime.now(timezone.utc)
        start = end - timedelta(minutes=minutes)

        resp = cw.get_metric_statistics(
            Namespace="AWS/EC2",
            MetricName="CPUUtilization",
            Dimensions=[
                {
                    "Name": "InstanceId",
                    "Value": instance_id
                }
            ],
            StartTime=start,
            EndTime=end,
            Period=60,
            Statistics=["Average"]
        )

        dps = sorted(
            resp.get("Datapoints", []),
            key=lambda x: x["Timestamp"]
        )

        datapoints = len(dps)
        cpu_avg = round(dps[-1]["Average"], 2) if dps else None

    result = {
        "instanceId": instance_id or "",
        "windowMinutes": minutes,
        "cpuAvgPercent": cpu_avg,
        "datapoints": datapoints
    }

    # Return Bedrock Agent response envelope

    if event.get("messageVersion") == "1.0":

        return {
            "messageVersion": "1.0",
            "response": {
                "actionGroup": event.get("actionGroup"),
                "apiPath": event.get("apiPath"),
                "httpMethod": event.get("httpMethod"),
                "httpStatusCode": 200,
                "responseBody": {
                    "application/json": {
                        "body": json.dumps(result)
                    }
                }
            },
            "sessionAttributes": event.get("sessionAttributes", {}),
            "promptSessionAttributes": event.get("promptSessionAttributes", {})
        }

    # Manual Lambda test output

    return result

Do NOT Forget to Click the Deploy Button

After pasting the code:

Click the Deploy button (top right)

If you see “Successfully deployed” → Perfect

Lambda Test

Now we verify whether the Lambda function is working.

  • Click the Test tab on the Lambda page
  • Click Create new test event

Event name:

cpu-test

Paste this exact JSON into the Event JSON:

{
  "instanceId": "i-05b89d8172f7e0d20",
  "minutes": 5
}

Click Save

Now click the Test button

Expected Output (If You See This → PASS )

Your output should look like this:

{
  "instanceId": "i-05b89d8172f7e0d20",
  "windowMinutes": 5,
  "cpuAvgPercent": 0.11,
  "datapoints": 1
}

Congratulations! Your Lambda function is working perfectly.

STEP 5– Now go back to Bedrock.

  • Look at the left sidebar
  • Scroll down a bit
  • You’ll find a section called Build

Inside Build, click on Agents.

Create Agent

The Agents page will open.

On the top right, you will see the Create agent button.

Click on Create agent

Enter Agent Details

A form will appear.

Fill in exactly the following:

Agent name:

cpu-monitor-agent

Description:

This agent checks CPUusageand takes actionif itis high. 

Click Create

The next page will open

You will now be on the Agent Builder page

Select the Foundation Model

Now you need to select a model (this will be the AI’s brain).

Remove the checkmark on the option that says ‘optimized for Bedrock Agents’, then search for the Nova 2 Lite model

Choose Nova 2 lite text model from the list

Do NOT select Nova Sonic (voice model)

Click Apply

Paste Instructions

Here we define the agent’s behavior.

Paste exactly this into the Instructions box

“You are a CPU monitoring assistant. Use the tool GetCpuUtilization to fetch CPU average for last 5 minutes. If CPU average > 85%, recommend restart. Otherwise recommend monitoring. Always ask for instanceId if missing.”

Additional Setting – User Input

In the Agent Builder, find the section:

User Input

Currently, it will be:

Disabled

Enable User Input

Select:

Enabled

Click Save

STEP 6 – Add an Action Group

You will now see the Action Groups screen.

Click Add action group

Fill in:

Action group name:

cpuMetrics

Action type:

Select Define with API schemas

Select the Lambda Function

Now choose your Lambda function.

Select the Lambda you created earlier

(example: get_cpu_utilization)

If the Lambda already exists:

Select it

Action group schema

click Define via in-line schema editor

OpenAPI schema

Copy–paste the full code below:

openapi: 3.0.1

info:
  title: cpuMetrics
  version: 1.0.0

paths:
  /getCpuUtilization:
    post:
      operationId: GetCpuUtilization

      description: Get average CPU utilization for an EC2 instance over a time window.

      requestBody:
        required: true

        content:
          application/json:
            schema:
              type: object

              required:
                - instanceId

              properties:
                instanceId:
                  type: string

                minutes:
                  type: integer
                  default: 5

      responses:
        "200":
          description: OK

          content:
            application/json:
              schema:
                type: object

                properties:
                  instanceId:
                    type: string

                  windowMinutes:
                    type: integer

                  cpuAvgPercent:
                    type: number
                    nullable: true

                  datapoints:
                    type: integer

Click Save
After saving:

The Prepare button will become clickable

Then click Prepare

STEP 7 – Open the Lambda Console

In the AWS search bar:

Lambda

Click it.

Select Your Lambda Function

Choose:

get_cpu_utilization

Open it.

Go to the Configuration Tab

On the Lambda screen, you will see these tabs:

Code

Test

Monitor

Configuration

Click Configuration.

Open the Permissions Section

Scroll down

Permissions

Click it.

Resource-based Policy

You will see a button:

Add permissions

Click it.

Fill in EXACTLY

Statement ID

AllowBedrockInvoke

Principal

bedrock.amazonaws.com

Action

lambda:InvokeFunction

ARN

"arn:aws:bedrock:us-east-1:613602870156:agent/*"

Click Save

STEP 8 – Then next go to Role

And in Bedrock, in the Agent Builder

Select the Agent resource role that is visible by going to the IAM roles section : choose

AmazonBedrockExecutionRoleForAgents_FF41143KBXU

Add Permissions:

  • AmazonBedrockAgentInferenceProfilesCrossRegionPolicy_L9ZHEFYE8FI
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "AmazonBedrockAgentInferenceProfilesCrossRegionPolicyProd",
      "Effect": "Allow",
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:InvokeModelWithResponseStream",
        "bedrock:GetInferenceProfile",
        "bedrock:GetFoundationModel"
      ],
      "Resource": [
        "arn:aws:bedrock:us-east-1:613602870156:inference-profile/global.amazon.nova-2-lite-v1:0",
        "arn:aws:bedrock:*::foundation-model/global.amazon.nova-2-lite-v1:0"
      ]
    }
  ]
}
  • AllowLambdaInvoke
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": "lambda:InvokeFunction",
      "Resource": "arn:aws:lambda:us-east-1:613602870156:function:get_cpu_utilization"
    }
  ]
}
  • BedRock
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "AllowInvokeNova",
      "Effect": "Allow",
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:InvokeModelWithResponseStream"
      ],
      "Resource": "*"
    }
  ]
}

STEP 9 – Test the Agent

Open the agent

Go to the Test tab
In the input box, type:

check cpu for instance i-05b89d8172f7e0d20 for last 60 minutes

Click Run

EXPECTED OUTPUT (If You See This → SUCCESS )

Congratulations!

You are now:

Taking decisions using AI

Triggering Lambda using AI

Successfully implementing a real Nova Act pattern

Related Searches

Related Solutions

Tags

Related Posts