Build AI Agents and Multi-Agent Systems with CrewAI/

...

/

Efficient Usage of Different LLMs and Tools in CrewAI

Efficient Usage of Different LLMs and Tools in CrewAI

Learn how to leverage different models for different tasks and how to create and use custom tools in CrewAI.

We'll cover the following...

In the world of AI, models are like the brains behind our operation. But here’s the thing: Not all brains are built the same, and that’s a good thing! Just like how a carpenter wouldn’t use their hammer to paint a picture, different AI agents can use different models depending on the task at hand.

Imagine a crew of AI agents working on an article generation project. First, we have the "Researcher" agent. This agent's role is to sift through vast amounts of data on any given topic and pull out the most relevant and valuable insights. For this task, we can leverage Gemini, a model known for its ability to understand and parse extensive text efficiently. Gemini can dive into reports, articles, and datasets, identifying key information that will form the foundation of our research. Next, we have the "Writer" Agent. This agent's job is to take the research findings and transform them into an engaging and informative article. What if we don't like Gemini's storytelling abilities for this agent? No worries, for this, we can use a model like GPT-4o, a model renowned for its creativity and mastery of language. GPT-4o can craft a narrative that not only conveys the facts but does so in a way that captures the reader's attention, making complex information accessible and compelling.

Press + to interact

Here's how CrewAI makes this possible: We bring these agents together, each powered by their specialized model, to work in unison. The Gemini-powered "Researcher" agent uncovers the data, while the GPT-4o-powered "Writer" agent weaves these findings into a narrative. They operate seamlessly, each playing to their strengths, resulting in a well-researched and well-written article on any given topic.

Initializing the models

Let's dive into how we would actually set up these models within CrewAI, along with a clear walkthrough for each step:

Press + to interact
from crewai import Agent
# Initialize the Gemini model using ChatGoogleGenerativeAI
from langchain_google_genai import ChatGoogleGenerativeAI
gemini=ChatGoogleGenerativeAI(model="gemini-1.5-flash",
                           verbose=True,
                           temperature=0.5,
                           google_api_key=os.getenv("GOOGLE_API_KEY"))
# Initialize the GPT-4 model using ChatOpenAI
from langchain_openai import ChatOpenAI
gpt=ChatOpenAI(model="gpt-4o-2024-08-06",
               verbose=True,
               temperature=0.5,
               openai_api_key=os.getenv("OPENAI_API_KEY"))

We can see that we are using the langchain packages. These packages provides an easy way to integrate the different model into our AI workflows with the option to set the temprature. The temperature controls the randomness of predictions, with ...