Learn how to build and deploy AI agents with LangGraph using watsonx.ai

23 June, 2025/in News/by albamart

Artificial intelligence no longer just responds, it also makes decisions. With frameworks like LangGraph and platforms like watsonx.ai , you can build agents that reason and act autonomously 🤯.

In this article, we will explain how to implement a ReAct (Reasoning + Action) agent locally and deploy it on IBM Cloud, all this with a practical example that includes a weather query tool 🌤️.

A practical guide to using your agents with LangGraph and Watsonx.ai

Project architecture

Machine with local project
- Here you develop and test the agent with Python, LangGraph and dependencies.
ZIP (pip-zip)
- Package with your code and additional tools.
Software Specification
- Environment with libraries necessary to execute the agent.
watsonx.ai
- Platform where you deploy the service as a REST API.
IBM Cloud Object Storage
- Stores deployment assets.

Let’s prepare the environment for our agent

We need:

Python 3.12 installed
Access to IBM Cloud and watsonx.ai
Poetry for dependency management

Have you got everything? Well, first things first, clone the repository that we will use as an example. It is based on the official IBM examples.

git clone https://github.com/thomassuedbroecker/watsonx-agent-langgraph-deployment-example.git 
cd ./agents/langgraph-arxiv-research

First of all, let’s understand the example project.

[Developer Workstation] → [CI/Build Process] → [Deployment] ↓
[IBM Cloud / watsonx.ai]

The main files of the agent are:

`ai_service.py`	Main file that starts the agent service in production.
`agent.py`	Core logic of the AI agent based on LangGraph. Defines the workflow.
`tools.py`	Tools connected to the agent (Weather API).

Let’s configure the environment

python3.12 -m venv .venv

source ./.venv/bin/activate

python3 -m pip install --upgrade pip

python3 -m pip install poetry

We also recommend the use of Anaconda or miniconda. It allows us to manage virtual environments or Python packages in a simple way and is widely used in ML.

In order for Python to find our custom modules (such as agents and tools), we need to include the current directory in the environment variable PYTHONPATH

export PYTHONPATH=$(pwd):${PYTHONPATH}

echo ${PYTHONPATH}

Once we have the environment ready, it is time for the variables. You must create a config.toml file if you don’t already have one and use your IBM Cloud credentials:

[deployment]

watsonx_apikey = "TU_APIKEY"

watsonx_url = "" # Tiene que seguir el siguiente formato: `https://{REGION}.ml.cloud.ibm.com0`

space_id = "SPACE_ID"

deployment_id = "YOUR_DEPLOYMENT_ID"

[deployment.custom]

model_id = "mistralai/mistral-large" # underlying model of WatsonxChat

thread_id = "thread-1" # Más información: https://langchain-ai.github.io/langgraph/how-tos/persistence/

sw_runtime_spec = "runtime-24.1-py3.11"

You will find your variables here:

https://dataplatform.cloud.ibm.com/developer-access

Once there, select your deployment space and copy the necessary data (API Key, Space ID, etc.).

Execution at the agent’s premises

It is time to test the agent:

source ./.venv/bin/activate

poetry run python examples/execute_ai_service_locally.py

Since it’s a weather agent why don’t you try it with something like something like…?

“What is the current weather in Madrid?”

The console should give you the time in Madrid. Congratulations! we only need to do the deploy in watsonx.ai

Agent deployment in watsonx.ai

source ./.venv/bin/activate

poetry run python scripts/deploy.py

This code will deploy the agent in Watsonx.ai

deploy.py does the following:

Read the configuration (config.toml) with your credentials and deployment space.
Package your code in a ZIP file for uploading to IBM Cloud.
Creates a custom software specification based on a base environment (such as runtime-24.1-py3.11).
Deploy the agent as a REST service in watsonx.ai.
Save the deployment_id , needed to interact with the agent later.

✅ In short:
takes your local agent, prepares it and turns it into a cloud-accessible service.

We check that everything is correct in watsonx.ai and go to the “Test” section. There we paste the following json (it is only one question)

{

"messages": [

{

"content": "What is the weather in Malaga?",

"data": {

"endog": [

0

],

"exog": [

0

]
},

"role": "User"

}

]
}

Click on predict and the agent will use the weather_service.

In the response json you will see the agent process -> call tool -> collect city -> process and return temperature.

Your agent is up and running on watsonx.ai!

If you want to test it from the terminal to make sure that it works, just use

source ./.venv/bin/activate
poetry run python examples/query_existing_deployment.py

Conclusions

If you have any doubts, we recommend the following video tutorial where you can follow step by step the development connected with watsonx.ai

🚀 "AI Agents: A Complete Guide to Developing and Deploying on watsonx.ai with IBM Cloud" 🚀

🚀 "AI Agents: A Complete Guide to Developing and Deploying on watsonx.ai with IBM Cloud" 🚀

If you want to continue exploring these types of implementations or learn more about cloud development and artificial intelligence, we invite you to explore our AI courses.👇

Blog!