Crew AI Build AI Agents Team With Local LLMs For Content Creation

Summary

The video tutorial demonstrates how to use Crew AI, a framework for building automated workflows using multiple AI agents. It shows how to connect local LLMs like Anthropic's Constitutional AI and LM Studio to Crew AI to define different agents with specific roles. A sample workflow is created with a 'researcher' agent that gathers information on AI advancements online using DuckDuckGo and a 'writer' agent that generates a blog post on the topic. The workflow enables automating research and content creation by harnessing multiple LLMs. Crew AI provides flexible integration of different LLMs to build efficient, multi-agent automation.

Takeaways

• 😀 Crew AI allows creating workflows using multiple AI models as agents
• 👥 You can connect Crew AI to OpenAI, Anthropic, Cohere, or local LLMs like AMA
• 💻 Installation requires Python and some dependencies like DuckDuckGo
• 🤖 Agents are defined with a LLM, work scope, and backstory
• 🔀 Tasks define what each agent must do in the workflow
• ✏️ A sample workflow gathers AI advancements info to generate blog content
• 📝 One agent researches info, another writes content based on that info
• ⚙️ Workflows can be sequential or hierarchical depending on needs
• 🔗 You can connect LM Studio in addition to AMA as agents
• 🎉 Crew AI enables automating workflows using multiple LLMs easily

Q & A

What is Crew AI?

Crew AI is a multiple AI agent framework that allows users to build workflows using multiple AI models as autonomous agents to complete automation tasks.

What types of large language models can be connected to Crew AI?

Crew AI can connect to models like GPT from OpenAI, as well as local large language models hosted on the user's machine, like those from Anthropic, Cohere, or LM Studio.

How does Crew AI allow the AI models to search the internet?

Crew AI uses the DuckDuckGo search library to enable the AI agents to search the internet and retrieve information.

What is the purpose of defining a workscope and backstory?

Defining a workscope and backstory provides context for the AI agents to understand their roles and objectives within the workflow.

What were the two agents defined in the example workflow?

The two agents were a 'Researcher' responsible for gathering information, and a 'Tech Content Strategist' responsible for using that information to write tech content.

How did the two agents interact in the workflow?

The Researcher gathered information on AI advancements from the internet, then passed that information to the Tech Content Strategist to write content based on it.

What tools did the agents use if they needed additional information?

The agents used the DuckDuckGo search library to search the internet for any additional information they needed.

What was the end result of the workflow?

The end result was a set of paragraphs with titles constituting a blog post on AI advancements.

Besides AMA, what other local large language model could be connected?

The script shows how LM Studio could also be connected as one of the agents in the workflow.

What benefits does Crew AI provide for content creation?

Crew AI allows automating content creation by using multiple AI agents in defined roles within a workflow.

Outlines

00:00

😊 Introduction to Crew AI and its capabilities

The paragraph introduces Crew AI, a tool that allows creating workflows using multiple AI models as autonomous agents. It allows connecting local LLMs like AMA or LM Studio to perform various tasks like content creation. The tutorial will demonstrate setting up Crew AI on a PC and connecting it to AMA and LM Studio to generate content.

05:01

👩‍💻 Step-by-step walkthrough on using Crew AI

The paragraph provides a detailed walkthrough on using Crew AI - installing it, connecting local LLMs, creating a script with multiple agents, defining tasks for research and content writing, executing the workflow to gather information and generate content. It shows a sample output demonstrating how the researcher gathers information which the writer uses to generate content.

Keywords

💡crew AI

Crew AI is a multiple AI agent framework that allows users to build workflows using multiple AI models as autonomous agents to complete automation tasks. It is the main focus of the tutorial video, allowing connection to large language models like AMA and LM Studio to define agents and tasks.

💡workflow

A workflow refers to a sequence or chain of multiple AI agents and tasks defined in crew AI to accomplish an overall goal. The video demonstrates building a simple 2-agent workflow to gather information and write blog content.

💡agent

Agents in crew AI refer to the individual AI models like AMA, LM Studio etc that are assigned specific roles and responsibilities in the workflow. The video defines a 'researcher' and 'content writer' agent.

💡task

Tasks define the objectives to be completed by each agent within the workflow. Example tasks are gathering information, writing content paragraphs etc based on the agent's role.

💡DuckDuckGo

The DuckDuckGo search library allows the AI agents to search the internet and retrieve external information when needed to complete their tasks in the workflow.

💡AMA

AMA refers to Anthropic's conversational AI model, which is locally hosted and accessed via API in the tutorial video to serve as one of the multiple agents in crew AI.

💡LM Studio

LM Studio is used to locally host a large language model which can also be connected as an agent in crew AI workflows, demonstrated in the later part of the video.

💡automation

A key benefit of crew AI is enabling automation of tasks by defining workflows for agents. The video automates research and content writing through its 2-agent workflow.

💡API

API refers to the interface for programmatically accessing AI models like AMA and LM Studio. Crew AI connects to them via defined API urls and keys.

💡code

The video walks through sample crew AI code for importing libraries, defining agents and tasks, and executing the automated workflow with multiple AI agents.

Highlights

Crew AI allows us to build a workflow using multiple AI models as autonomous agents

We can connect Crew AI to OpenAI, Anthropic, Cohere, or a local LLM like AMA or LM Studio

Crew AI provides instructions on connecting different LLMs and customizing AI agents

We design workflows using sequential or hierarchical processes with multiple agents

Crew AI shows how to connect AMA and LM Studio locally without an API key

We define work scope, backstory, LLM, and tasks for each agent in the workflow

One agent researches information, the other writes content based on that

Agents analyze if they need more info and can search DuckDuckGo if researcher

We define the crew with agents, tasks, and execute the workflow

Researcher gathers info, writer creates content in 4 paragraphs

Example workflow shows automation for content creation from research

Can connect multiple LLMs like AMA and LM Studio as different agents

Code shows how to define API URL to connect other LLMs as agents

Example workflow automates researcher and writer tasks for content creation

Transcripts

crew AI a multiple AI agent framework

allows us to build a workflow that

autonomously agents to complete

automation tasks for us let's check it

out so in this tutorial we are going

through crew AI it allows us to build a

chain of workflows using multiple AI

models as our autonomous agents to

accomplish certain tasks that you define

for the AI agents here we will have a

step-by-step walkthr on how we can set

this up on our PC and also connect it

with our local llm models to run certain

tests and perform content creation

processes so right here as we can see

the crew AI official page is open-

Source tools and it allows us to use any

large language models for example you

can connect with your open AI API key to

use Chad GPT or if you have a local

large language model you can connect

with Ama or LM Studio to incorp

incorporate those large language models

into your workflow and Define multiple

large language models as individual AI

agents to complete specific tasks in the

documentation they provide clear

instructions on how to connect with any

large language models customize the

agents and utilize sequential or

hierarchical processes to design

efficient workflows importantly you need

to know how to connect with a large

language model hosted locally on your PC

for example we have LM Studio here in

this section it shows you how to define

the API based URL for LM Studio you

don't need an API key so leave that

field empty additionally they provide

examples on how to connect with AMA in

this video we are going to connect with

these two popular options AMA and LM

Studio you can also connect with

Microsoft Windows Azure to access their

AI Studios using authorization keys and

API Keys now let's get started with

creating a workflow and connecting those

large language models in crew AI first

we need to install crew AI by running

this command prompt and downloading the

necessary

files after that it will automatically

install the required

components then we need to install the

duck duck go search Library this library

is essential for enabling our AI to

search the internet and retrieve

information for

us once that is done we are good to go

now let's open Visual Studio code vs

code and start creating our script for

multiple AI agents here I have already

run two AMA large language models one is

Mistral and one is llama 2 so both of

them just keep that command prompt

running in the background and also the

AMA running the AMA server here which

allows API access for AMA so in here we

have vs code and this script you can

download it on the GitHub page of crew

AI I will post a link to it in the video

description below scroll up here and as

you can see I have the Lama 2 and

Mistral using as well to connect here

basically I will use this tool to

generate content for my workflows and to

run two individual agents in different

roles in this workflow and right here

this is the template of the very basic

crew AI multiple agents creation script

and above that we have the installation

command prompt you can just copy this

too and install it and it's pretty easy

actually let's go through each of the

rows that we are going to do so in the

OS versions here we have the open AI API

key now we don't need that so we just

put that into comments and we don't need

to run that script then we scroll down

into here let's go through each one so

running 2 and then Mistral running as

the second agent using

Al and right below that we have the land

chain tools that is we import the Duck

Duck Go search and we are going to use

Duck Duck Go search to gather

information online if the large language

model necessary needs some information

from them and then we have to Define

the llm on each agent here these llms

are using the llm which is the Llama

tool that I defined at the beginning of

this script then we have the work scope

and the backstory you have to define

those as well and then the second agent

here is the tech content strategist

which is the content writer for the tech

content blow poost these agents are

responsible for gathering information

from the first agent and then writing

content based on it

these agents are going to use mistal llm

from my defined Al agent large language

model for this agent lastly we have to

create the tasks the tasks are going to

Define what these two agents have to do

in this workflow there's task one which

is the researcher and task two which is

the writer the writer is going to write

about the AI advancements based on the

content they're getting from the

researcher lastly we have to Define find

the crew the crew includes the agents

and tasks that are included in this crew

workflow lastly it will execute on the

last command code here is my first run

using this code and it gathers

information from the researcher so the

researcher is going to the internet and

then Gathering the information secondly

we have the tech content strategist who

will use that information to write four

paragraphs of content or code for my

blog post this is a very typical iCal

task for lots of bloggers and website

owners who want to create new content

this kind of workflow is simple yet

effective and practical allowing people

to automate this task here as you can

see each AI agent has its own automation

they will analyze if they need tools to

run their tasks and then answer

themselves if the answer is yes they

will use duck ducko if the agent is a

researcher they will go online and

search for re AI advancements and pass

that information to the

writers the content writers will then

write the content based on that

information the content is divided into

different paragraphs as you can see with

title paragraphs and all the content and

additional thoughts about the content

this is how we use crew AI Additionally

you can also connect LM studio with crew

AI not just AMA in this case we have the

first agents using Ama and the second

agents using LM Studio the code is

provided here for the first one as you

can see we use the URL at the top to

connect and create content using other

large language models as agents in this

workflow I hope you find inspiration in

using crew AI this is a simple workflow

using two agents in one automation

workflow I will see you in the next

tutorial have a nice day

bye