Using the LLM client¶

The LLM Client library in YARF provides Robot Framework keywords for interacting with Large Language Model servers that support the OpenAI Chat Completions API format. This enables you to integrate AI capabilities into your test automation workflows.

In this guide, we will cover:

Setting up an LLM server
Basic text prompting
Using images in prompts
Configuring the client

Setting up an LLM server¶

The LLM Client is designed to work with any server that implements the OpenAI Chat Completions API. The most common setup is using Ollama:

Installing Ollama¶

Install Ollama on Linux¶

curl -fsSL https://ollama.com/install.sh | sh

Starting a model¶

Download and run a vision-capable model¶

ollama run qwen3-vl:2b-instruct

By default, Ollama serves on http://localhost:11434 with the Chat Completions API available at /v1/chat/completions.

Basic text prompting¶

To use the LLM Client in your Robot Framework tests, first import the library:

Import LLM Client library¶

*** Settings ***
Library    yarf.rf_libraries.libraries.llm_client.LlmClient

Simple text prompt¶

Basic text prompting example¶

*** Test Cases ***
Ask LLM a Question
    ${response}=    Prompt Llm    What is the capital of France?
    Log             ${response}
    Should Contain  ${response}   Paris

Using system prompts¶

System prompts help guide the LLM’s behavior and responses:

Using system prompts to guide behavior¶

*** Test Cases ***
Structured Response
    ${response}=    Prompt Llm    
    ...             prompt=Analyze this test result
    ...             system_prompt=You are a test automation expert. Provide concise, actionable feedback.
    Log             ${response}

Using images in prompts¶

The LLM Client supports multi-modal prompts that include both text and images. This is particularly useful for visual testing scenarios.

Image from file path¶

Analyze a screenshot with the LLM¶

*** Test Cases ***
Analyze Interface Screenshot
    # Take a screenshot first (example using YARF's screenshot capabilities)
    ${image} =              Grab Screenshot
    
    # Prompt the LLM to analyze the image
    ${analysis}=    Prompt Llm    
    ...             prompt=Describe what you see in this user interface
    ...             image=${image}
    
    Log             ${analysis}
    # The screenshot shows an image of a simple calculator
    Should Contain  ${analysis}    calculator

Image validation workflows¶

Use LLM for visual validation¶

*** Test Cases ***
Validate Installation Screen
    ${image} =          Grab Screenshot
    ${validation}=      Prompt Llm
    ...                 prompt=Does this screen show the ubuntu installation on the "choose your language" step? Answer with YES or NO.
    ...                 image=${image}
    ...                 system_prompt=You are a UI testing assistant. Be very precise in your answers.
    
    Should Start With   ${validation}    YES

Configuring the client¶

The LLM Client can be configured to work with different servers, models, and parameters.

Changing the model¶

Switch to a different model¶

*** Test Cases ***
Use Different Model
    Configure Llm Client    model=phi4-mini:3.8b
    ${response}=            Prompt Llm    Hello, what model are you?
    Log                     ${response}

Using a different server¶

Connect to a remote LLM server¶

*** Test Cases ***
Remote Server Setup
    Configure Llm Client    
    ...    server_url=http://192.168.1.100:11434
    ...    model=llama3.2-vision:11b
    
    ${response}=    Prompt Llm    Test connection
    Log             ${response}

Adjusting token limits¶

Configure maximum tokens for responses¶

*** Test Cases ***
Short Response
    Configure Llm Client    max_tokens=100
    ${response}=            Prompt Llm    Write a brief summary of automated testing
    Log                     ${response}

Complete configuration example¶

Full client configuration¶

*** Test Cases ***
Custom Configuration
    Configure Llm Client    
    ...    model=qwen3-vl:7b-instruct
    ...    server_url=http://llm-server:11434
    ...    endpoint=/v1/chat/completions
    ...    max_tokens=2048