Extracting Address with Mistral via Ollama and UiPath - Locally and for Free

Generative AI continues to reshape automation, the ability to run large language models (LLMs) locally opens up new possibilities for cost-efficient, private, and fast automation workflows.

In this post, we’ll walk you through how to:

• Run Ollama locally with the Mistral model
• Use API calls to interact with the model
• Parse addresses and extract structured details 
• Integrate it with UiPath using HTTP requests and JSON deserialization

Let’s get started!

Step 1: Install Ollama Locally

First, you need to install Ollama, a lightweight tool that allows you to run LLMs on your local machine.

• Download Ollama from the official site (Ollama supports Windows, macOS, and Linux):
  https://ollama.com/download

Once installed, test your installation in your cmd:

• Ollama --Version

This will confirm that Ollama is running correctly, you can also check the local host link (http://localhost:11434/):

Add the Mistral Model

Next, we’ll pull the Mistral model, known for its efficiency and strong performance in language tasks.

Run this command in your terminal:

• ollama pull mistral

Once downloaded, you can run it:

• ollama run mistral

Ollama runs a local server (usually on http://localhost:11434) to handle API requests.

Define the Address Parsing Task

Our goal is to send natural language address strings to the Mistral model and receive structured JSON responses with fields like:

{

  "street_name": "Champs-Élysées",

  "street_number": "50",

  "postal_code": "75008",

  "city": "Paris",

  "country": "France"

}

Let’s define a sample prompt:

"Extract the street name, street number, postal code, city, and country from the following address and return it as a JSON object:

50 Champs-Élysées, 75008 Paris, France"

 Integrate with UiPath

Now let’s use UiPath to send the request and process the response.

Tools in UiPath:

• HTTP Request Activity
• Deserialize JSON Activity

Set up your HTTP Request:

• Method: POST
• Endpoint: http://localhost:11434/api/generate
• Body (JSON):

{

  "model": "mistral",

  "prompt": "Extract the street name, street number, postal code, city, and country from the following address and return it as a JSON object:\n50 Champs-Élysées, 75008 Paris, France"

} 

Deserialize JSON:
Once you receive the response, use Deserialize JSON activity to parse the "response" field.
From there, you can access each value using:

• jsonObject("street_name").ToString
• jsonObject("postal_code").ToString