This article explains how to build a basic Vertex AI agent that provides city weather information and shows how to call it from a Pega case using A2A. The agent uses the ‘Google Search’ tool, offering an alternative way to integrate Google Search into your Pega flows.
Step1: Build External Agent
Prerequisites
1. In console.google.com, create Google Cloud Project. Make sure billing is enabled for this cloud project
2. Install Google CLI on your laptop. Follow instructions on below page. https://docs.cloud.google.com/sdk/docs/install-sdk. For initialising Google Console, please use the account and project details from Step 1.
3. Enable Vertex AI APIs using below command.
gcloud services enable cloudresourcemanager.googleapis.com \
servicenetworking.googleapis.com \
run.googleapis.com \
cloudbuild.googleapis.com \
artifactregistry.googleapis.com \
aiplatform.googleapis.com \
compute.googleapis.com
Create agent in VS Code
1. Create a virtual environment
python3 -m venv .venv
2. Activate virtual environment and install adk packages
source .venv/bin/activate
pip install google-adk
3. Easy steps to create agent using Google’s Agent Development Kit can be followed here
https://adk.dev/get-started/python/#create-an-agent-project
4. Below are the instructions that you can follow to create an agent hosted on Google Cloud to fetch weather of a given city.
adk create simple_weather_agent
5. Above command creates the folder structure and the python files needed for this agent e.g. init, agent
6. Navigate to ‘Agent Designer’ on google console to create the code for the weather agent.
7. Create an agent and click on ‘Get Code’ to view the code for this agent.
8. Copy the python code and paste it in the agent.py file in the VS code project.
Test your agent
1. Run below command to test your code. (Note: Make sure you in a parent folder of the agent folder)
adk web
2. A localhost link will be crated for this agent as seen in below screenshot
3. Click/Open this link browser to test your agent and you can chat with this agent and test if it is working.
Make this agent A2A Compatible
1. Create agent skill as show below for this agent. You could also use github copilot/vibe coding to add the agent skill. However, adk libraries are changing frequently hence some of the vibe coding tools may not give you the latest code that can run on vertex AI, hence refer to the Google documentation here. https://docs.cloud.google.com/agent-builder/agent-engine/quickstart-adk
# A2A Agent Skill definition
skill = AgentSkill(
id='get_weather_info',
name='Weather Information Tool',
description='Helps with fetching weather information for a given location',
tags=['weather', 'forecast', 'temperature'],
examples=['What is the weather in New York?'],
)
2. Create agent card referring to agent skill here
# A2A Agent Card definition
agent_card = AgentCard(
name='Weather Agent',
description='Helps with fetching weather information for a given location',
url=SERVICE_URL, # Replace with your service URL
version='1.0.0',
default_input_modes=["text"],
default_output_modes=["text"],
capabilities=AgentCapabilities(streaming=True),
skills=[skill],
)
3. Add below line to agent.py
# Make the agent A2A-compatible
a2a_app = to_a2a(root_agent, agent_card=agent_card)
4. You will also need to install ADK a2a packages
pip install a2a-sdk
5. Also, you need to add Dockerfile and _main_.py file to the folder structure to deploy this agent on CloudRun. Refer to this github project for dockerfile sample
https://github.com/google/adk-python/blob/main/tests/remote/triggers/test_agent/Dockerfile
6. Create requirements.txt file for the project by following below command. (Note: you could also use uv instead of pip)
pip freeze > requirements.txt
Deploy to CloudRun
1. Sample command to deploy this agent to the CloudRun is as below. (Please note we are creating a simple agent without any authentication. Please do not use unauthenticated agents in Production. I will be sharing another article to add security to the agents. Google agents need ID tokens as opposed to Access Tokens hence you need to create a service account and keystore to create JWT bearer tokens)
gcloud run deploy weather-agent \
--port=8080 \
--source=. \
--allow-unauthenticated \
--clear-base-image \
--memory "1Gi" \
--region="us-central1" \
--project= <your_project_name> \
--set-env-vars=GOOGLE_GENAI_USE_VERTEXAI=true,\
GOOGLE_CLOUD_PROJECT="your_project_name ",\
GOOGLE_CLOUD_LOCATION="us-central1",\
APP_URL="TEMPORARY_URL"
2. Above command gives the APP URL in this format
https://weather-agent-xxxxxxxxxxxx.us-central1.run.app
Run below commands to update the APP_URL and the SERVICE_URL
3. Run below commands to update the APP_URL and the SERVICE_URL
gcloud run services update weather-agent \
--project="<your_project_name>" \
--region="us-central1" \
--update-env-vars=APP_URL="https://weather-agent-xxxxxxxxxxxx.us-central1.run.app"
gcloud run services update weather-agent \
--project="<your_project_name>" \
--region="us-central1" \
--update-env-vars=SERVICE_URL="https://weather-agent-xxxxxxxxxxxx.us-central1.run.app"
4. You could access agent card using below URL
https://weather-agent-xxxxxxxxxxxx.us-central1.run.app/.well-known/agent-card.json
5. Below is the json format for the agent card that can be viewed in the browser.
Step 2: Call this agent from Pega Case
1. Add a Step ‘CheckWeather’ to the Case as we can see in below screenshot
2. You could specify instructions as shown here
3. In external Tools tab for this agent add the Google Agent instance
4. Create GoogleWeatherReportAgent with details as shown here
Test Workflow
1. Run the workflow locally and enter the Location of your travel
2. View the weather forecast
