CrewAI Tutorial for Beginners: Learn How To Use Latest CrewAI Features
HTML-код
- Опубликовано: 15 май 2024
- 🤖 Download the CrewAI Source Code Here:
brandonhancock.io/crewai-upda...
Don't forget to Like and Subscribe if you're a fan of free source code 😉
📆 Need help with CrewAI, join our free Skool Community:
skool.com/ai-developer-accele...
Ready to lead an AI revolution? Watch and learn how to build your own CrewAI from the ground up using the latest CrewAI features, and get set to deploy an army of AI agents at your command. This video is your ultimate guide to creating a powerful digital workforce, enhancing your projects with intelligent automation and streamlined workflows. Discover the secrets to customizing AI agents, setting them on tasks, and managing a smooth operation with CrewAI. It’s time to amplify your tech capabilities, and after this tutorial, you'll be equipped to engineer an AI crew that transforms any complex challenge into a simple task. Start your journey to AI mastery with CrewAI today!
📰 Stay updated with my latest projects and insights:
LinkedIn: / brandon-hancock-ai
Twitter: / bhancock_ai
Resources:
- github.com/joaomdmoura/crewAI...
- www.crewai.io/
- / 1756428892045496608
- serper.dev/
Music used in the video:
Retro Electro EDM by Aries Beats: / @ariesbeats
MONIN - Binary Stars: • MONIN - Binary Stars [...
Heavy Synthwave Party by Aries Beats: / @ariesbeats
Timestamps:
00:00 Introduction
00:59 4 Major CrewAI Updates
06:28 Setup Project & Dependencies
09:53 Setup Agents in Crew
12:59 Setup Tasks in Crew
17:51 Setup Main.py with Agents and Tasks
24:00 Setup Tools
28:47 Setup Callback to Save Newsletter to Markdown File
30:43 Run Crew
33:21 Showcase Crew Results - Final Newsletter
34:27 Outro 
Наука
![How To Connect Local LLMs to CrewAI [Ollama, Llama2, Mistral]](http://i.ytimg.com/vi/0ai-L50VCYU/mqdefault.jpg)
![How To Connect Local LLMs to CrewAI [Ollama, Llama2, Mistral]](/img/tr.png)
I really enjoyed the way you rolled through that tutorial Brandon! For me, you hit all the points at just the right level of detail. Super easy to follow and left with just what I needed. Thank you for this!
By far the best tutorial I've seen on any of these AI agent frameworks. Thank you!
Thanks Jim! I really appreciate you saying that! What's crazy is CrewAI is updating so fast that they just released another major version. If you want to see how to use the latest features, I just dropped a new video for you guys that you can check out here:
ruclips.net/video/OumQe3zotGU/видео.html
Solid update. I hope you'll continue with Crewai videos. Thank you for your hard work.
Thank you! 😁 I have a lot more CrewAI videos planned for you guys!
Incase anyone is wondering why its not working, try checking your OPEN AI credit balance, they recently updated the way they charge switching to a refill balance model. Without credit, it disable all api keys.
You are genius. All comments confirm that you make things very easy to grasp CrewAI perfectly.You really produced a very professional video, where you cascaded the steps of coding, and make things easy for us.👏👏
One more thing to add - this is the first example that I downloaded, followed all the instructions, and it worked incredibly well 👏👏👏 and it's also super useful - thank you again for the video and the code!
Hey Ali! I'm glad you liked it and that everything was super easy to setup on your end!
I have a lot more CrewAI tutorials planned for you guys so you'll definitely want to check those out as well when the time comes!
Thanks man, it was my first working complex project in my local. Appreciated.
You have the best tuts on the net, my man.
3 minutes in and I can already tell this is gonna be pure gold. Thanks!
Thanks man! Also, I can see you're putting in the work on your RUclips channel! I just subscribe!
Oh thanks so much! I appreciate that alot!
Also, the video was SUPER useful I learned alot from it. 10/10@@bhancock_ai
@@bhancock_ai@bhancock_ai 41 seconds in, and I realised I'd stumbled upon the feature I was looking for in no-code SaaS platforms-Ker Ching!. However, I hope they make it more user-friendly soon, for us technophobes who have never coded.
As someone who started in the space recently, i really appreciate you going step by step and actually explaining what you're doing in each section. I've watched about 5 times already and finally I was able to get my first crew up and running. I also asked chat GPT how many times you said fantastic in this video and it said 19.... to which I say, thank you, this video was Fantastic.
That’s hilarious! Or… should I say Fantastic! Lol
@@bhancock_ai20
I can't unhear this now🤣
You have the best CrewAI video tutorial! Thank you so much!
Thanks Jeff! I put a lot of work into these videos so that really means a lot to me!
Great tutorial! Thanks. Now I need to go back & do the full tutorial. 👍🙏😀
😂 I'm keeping you guys busy lol!
Awesome, thank you so much! I downloaded the code and got it working in an hour or so. This is a great guide.
Great to hear! You’ll have to keep us posted on what you build!
This is awesome, I have learned so much, thank you.
clean, clear i love your way to explain steps, your way of writing your code, and organization thank you
Thanks yassine! I really appreciate you saying all of that!
Honestly, the hardest part of making these RUclips videos is trying to figure out how to present all the info in an easy to understand fashion so thank you for saying that!
Well done. You have set the gold standard, my friend. Good luck to other CrewAI/AutoGen/"Flavour of the Month" content creators trying to match this level of excellence.
Thanks wade! I put a lot of work into the videos so I really appreciate you saying that!
Amazing tutorial! Thank you!
Thanks man! 😁
Thanks! Excellent videos. Super useful.
great content! thanks a lot for this useful tutorial
Thank you for the videos! I am going to try this and see if I can have a web development crew.
That would be awesome! Please let me know how it goes!
Great video per usual. Would love to know the color theme you are using (if you customized it) and what extensions you are running in VS Code. Thanks for your hard work, I echo all of the funny and good comments below!
Great video! I love it! 😊
Great video, thanks 👍
Subscribed! Great content and overall presentation of development. Autonomous agents will be important as society builds off these.
I don't think the majority of people realize how impactful these agents are going to become as time goes on!
@@bhancock_ai Absolutely! The next new industry generated by AI.
Excellent video.... thank you.
Thank you for the tutorial!
I was only able to get it to work with a GET request. My POSTs returned 403. Also, it seems there are supposed to be guardrails to ensure the context length cap of the model isn't exceeded, but there's aren't working currently. And I had to change "organic" to "organic_results" since that what my SerpApi responses had.
thank you!
Great video - subscribed - great channel!
You the man Ali!
Hey Brandon, thanks for the detailed video. It's really good content, I like it! Could you make a video on how to use RAG when the downloaded content for the articles is too long. For example, if I want to write a post about the latest scientific research from arxiv, the agent should download (+ embed + retrieve) one document, read it, and if it's not enough, download the next one.
thank you for this content and we need a new video for the new update
Very helpful video, thanks! I'm a little confused about the manager vs editor agent. The editor agent you create sounds like a manager, but it doesn't have an associated task, and when you set the process to be hierarchical with the GPT-4, it looks like the code generates its own agent to be the manager, so what happens to the editor agent? Does the manager delegate to it without an associated task?
AMAZING VIDEO!!
Thank you 🙏 for awesome videos
Thanks man! I appreciate it!
Great video. How about an update showing how to use this with a locally running LLM in LLMStudio for example.
You read my mind! I just dropped a video 5 minutes ago that shows how to use local LLMs with CrewAI: ruclips.net/video/0ai-L50VCYU/видео.html
As it happens, I did some research and figured it out for myself , but will checkout yournew video too to compare notes. Turns out is was not too difficult! Thanks and keep up the great work@@bhancock_ai
hey Brandon thanks your vids are amazing!
One question , what was the cost of tone run of the newsletter generator? mine was 1.2 dollar with GPT4
Tx!
Thanks Hendrik! I'm glad you're liking these tutorials!
And, that's pretty similar to what I've seen on my end. Here are a few things that I've done to reduce cost:
- Set max_iters on the crew leader.
- Provide more constraints in my tasks. For example, if you tell the crew to look up information, it sometimes doesn't know when it has enough info so it will keep running. This means it takes longer for your Crew to finish and it cost you more money.
- I use local LLMs to run things for free. I actually just dropped a tutorial a day ago if you want to learn how to set it up on your own:
ruclips.net/video/0ai-L50VCYU/видео.htmlfeature=shared
@@bhancock_ai thanks for the reply! That's why open source models become so relevant !
Hey Brandon,
Just wanted to drop a quick note to say how much I enjoyed your tutorial - it's awesome stuff! 😊 I've been thinking about the future of AI and had a question I wanted to bounce off you. Do you think there's potential for agents to act more like assistants, you know, collaborating with users instead of doing everything solo? I imagine a future where AI and humans work together, with AI checking in with us before taking action. It's like humans augmenting their abilities with AI, don't you think? So, is this possible with CrewAI?
Looking forward to hearing your thoughts!
Amazing!
I got it to run and it found results in the intermediate results with title and url but it only saved the last article it found with no title or url. Found titles and url in intermediate results but not in the final saved MD.
A second run was more varied but still no url in the final document.
It didn't seem to happen in the video, is there a glitch in the code or did I make an error?
Brandon thanks for tutorial.
I have a question: "How to make an agent that can search Internet as a human?"
Awesome video. I have a doubt.
Agent 1 - Research: Research on the internet, pick out information from input notes provided by teacher, pick out the right info, and weave it into a good story.
Agent 2 - Writing: use research context, learn the patterns of good writing, and produce engaging and interesting content
Agent 3 - Editor: Agents 1 and 2 should engage in arguments and then reach final decisions to conclude.
How to implement third agent using crewAI?
Will you build fullstack SaaS project using Nextjs and CrewAI as backend? Thank you a lot for what you're doing. Hope to see that from you
I think we can use streamlit or chainlit to make some projects for ourselves. Actually I liked those frameworks.
I definitely plan on making a full stack CrewAI video very soon!
My goal is to cover all the basics first and the compile everything into one video and connect it up to NextJS
Hey Brandon, need your help if you havw a solution in which i can display the terminal output on the frontend on flask how can i achieve that, i have tried many different ways but still not able to get the real time output
Thank you it was very clear! However, I have a question. Why did we create an agent "editor" without any task given to him? Aren't we in that case not using "editor" agent?
These are great tutorials. However, considering that this was a follow up CrewAI video to the original, it would have been super helpful to update or re-build the original example project (Travel Agent) instead of starting a new project. As a beginner, its harder for me to understand the updates and features when I'm also having to correlate them back and forth between the projects.
Subbed for crewai
Hi brandon. I joined your skool website and i tried to create a post but it says I either cancel membership or wait approval from admins to be accepted. I'm confused as I dont remember doing some membership and when I looked into settings, I saw some payment tabs. How does it actually work?
Cool stuff, do you have any tips to get this to work with Azure OpenAI deployed models?
I haven't worked with Azure OpenAI deployed models yet! In one of my upcoming tutorials, I plan on using CrewAI with more cloud technologies so I can try to squeeze this in!
how does it know that context that you mentioned is in reference from fetch_news_task()? okay, i see now. it's explained later in the video.
Fantastic! Congrats! Uouuuu
Great content, thank you! How much this newsletter crew ai costs you?
Hey! It cost about $1.50 to run with chatGPT 4. I would use a local LLM for testing purposes when you’re starting out because they’re free! I have another tutorial that explains how to do that if you want to check that out!
I wonder if additional instructions can be made to make it work with local models like Llama instead of ChatGPT, unless hierarchical agents are not supported outside ChatOpenAI
Hey! I actually just released a video covering this very topic:
ruclips.net/video/0ai-L50VCYU/видео.html
I hope it helps!
4.5k views is utterly disrespectful. Brandon, you put the B in BAMF. Love you dawg.
Thanks man! You're awesome and I really appreciate it!
@@bhancock_ai Hey man! You're face is awesome 😂
Question for you. Can this be used to debate a topic and come to a general consensus about the query? Then write a summary about the discussion?
You sure could! Also, if you want more help as you go off and build the agents to debate a topic, I just launched a free Skool community.
In the community, you could post about what you're trying to accomplish and get a lot more support from the members in the group!
www.skool.com/ai-developer-accelerator/about
crewai falls apart when you try to use local llms with tools. Any tutorials on using local llms with search tools would be much appreciated.
Is there a community somewhere besides these comments where we can talk about this app? Discord, reddit, Dev, etc? ther's a lot to talk about and I think we can evolve this demo into a more refined applications.
I'm so glad you asked! I actually just launched a free Skool community for you guys!
Feel free to join here:
www.skool.com/ai-developer-accelerator/about
Just out of curiosity: What was the cost incurred when you ran this code? Also: perhaps you can embed code that calculates cumulative costs incurred as the program is running, as well as providing a maximum cumulative cost stopping point. One of the dangerous things about using AI agents is the lack of cost-control that results from the open-ended nature that running ChatGPT can impose.
Opensource is king for this.
$1,39
Gross@@amorim33_
Running ChatGPT 4 agents can be very costly! It varies for me between $1.00 to $1.40.
I just released a new video that shows how to run local LLMs on your computer for free so you don't have to pay an arm and a leg to OpenAI.
ruclips.net/video/0ai-L50VCYU/видео.html
I hope that video helps!
where are you deploying this on the cloud? local dev is great but its about using this in prod...bedrock? somewhere else? how are the costs?
Hey! Great question! Right now, the way of running CrewAI apps in the cloud is really involved.
The good news is CrewAI+ is coming out very very soon which makes the entire process super easy.
My next 2 to 3 videos are going to be focused on showing you guys how to use this new technology!
Fantastic video! Going through the build in this way is SO MUCH BETTER than all the other channels that think spending 50% of their time in a video explaining "pip install yada yada" is helping the people that actually want to code this stuff. Any creator that will spend the time doing these actual builds, showing their mistakes, will be the gold accounts!
Thanks!
I really appreciate you saying all of that! I put a lot of work into these videos so it's awesome to know that you are getting a lot of value out of them!
If you have any other CrewAI ideas that you'd like to see more videos on, I'd love to hear your suggestions!
I changed it to use local llm, and the task_output going to write the file doesn't appear to contain task_output.result. Although I could see the data being gathered previously. Any thoughts?
Hmmm. That's weird. Which LLM are you using? Also, if you log the task_output, is it completely empty or is just the result empty?
Hey, this is pretty cool! I'm not a programmer, but I had a question. Instead of using the terminal, is it possible to create a user interface or an input form? You know, like an app or something?
Great question! You definitely can do that and I plan to make a video doing exactly what you're talking about after I finish the current video I'm working on now!
@@bhancock_ai that would be great. Thank you
hey brandon awesome build... here is a fan request for you build a habit tracker site with AI giving suggestions
Hey Mugilan! I actually made a video a while ago that shows how to make an AI agent that generates workouts for you in a David Goggins fashion:
ruclips.net/video/b1S04PFjIOY/видео.html
But, that's a little different than what you're suggesting.
First class tutorial
Great video.Can you make a Stock analyzer that utilizes all free llms or use the Google API free key.That would be awesome.
Hey! I just released a video that teaches you how to run local LLMs on your computer for free and how to connect these LLMs to CrewAI:
ruclips.net/video/0ai-L50VCYU/видео.html
I think this is the exact video you're looking for!
thx cool of you loved
i am having the same problem from the last video. I keep getting stuck at the poetry install ---no-root step. everytime i do it says "[tool.poetry] section not found in [folder name]" i even showed a screenshot to gpt and it confirmed that file is in the correct folder. I have 0 coding experience and cant find an answer anywhere. I installed poetry properly. please help :(
how do i create an agent which searches the documents (PDF) provided instead of doing a web search
Supercool does this work with Mistral 7B or similar ?
So certain parts of CrewAI work with local LLMs right now. For example, you can't do async tasks.
I'm thinking in the next version or 2 of CrewAI & Ollama, they will hopefully support more features.
@@bhancock_ai when you're using lmstudio and say Mistral-7B-Instruct-v0.2, what are you supposed to set manager_llm= to? what code do you use to initialize? you rock, thanks!
would duckduckgo work as well, in place of serper?
You could use DuckDuckGo! The only problem I've had with using that tool is that it gives back way too much information and this can't really drive up the price of the crew because we have to pay for tokens that get processed. Serper does a great job of returning decor pieces of information that we need instead of 5000 characters where most of them are just gibberish
Are you not getting validation error from pydantic when making Tasks?
if you are getting one with v 0.19 use expected_output parameter in the Task
You're spot on! expected_output used to be an optional parameter but now it's required! Great find!
I don't have access to gpt-4 so i try it using mistral locally unfortunate doesn't work i don't know if this will not be working either for gpt-3.5
Hey! Certain parts of the new CrewAI update don't work with Local LLMs. I'd check out this tutorial where I talk more about connecting CrewAI to Local LLMs
ruclips.net/video/0ai-L50VCYU/видео.html
I hope that helps!
@@bhancock_ai Thanks this was helpful
Hello Brandon: Your Calendly link is no longer working. Kindly let me know. Many Thx, D
Hey! Thank you for pointing that out!
I just launched a free Skool community because my calendly was getting too busy. Feel free to join over there! I do weekly group calls with all of you guys.
Can't wait to see you in the community!
www.skool.com/ai-developer-accelerator/about
Did not find openai_api_key, please add an environment variable `OPENAI_API_KEY` which contains it, or pass `openai_api_key` as a named parameter. (type=value_error)
Can I help myself to fix this error?
do you already have a .env file?
Can i get this code.I cant find your code in the github repo.And moreover the serper api is not working.I cant even login into the website
Mine dosent show any squggly lines for Agent or crewai.. and i dont see a Python version.. just says Python.. is my VS code configured weird??
OOP at it's best 🎉
Boas, por favor crie um curso didatico para iniciante utilizado está ferramenta maravilhosa. O Alan deu uma dica para você valiosa e poderia divulgar isso melhor no Brasil. Claro se houver seu interesse e melhores cumprimentos.
Loved the video but I signed up twice for the code and I did not recive anything via email. I did check my spam
Good morning Bennie! I checked my email list and it says that you should have received an email with all the code at 6:35AM EST today.
If you didn't get it, please shoot me an email at brandon@brandonhancock.io
For some reason i keep gettin an error telling me about delegate to coworker doesn't exist... any advice?
For questions like this, I definitely recommend hopping over to the Skool Community and recording a quick video of what's going on so that we can help!
Can i create this project and share it to my resume?
Go for it man! I would definitely recommend building your own Crew to accomplish your own task so that you'll get practice using these tools on your own!
@@bhancock_ai I have a question what kind of projects I should include in my resume for a genAI role. Would it be traditional deep learning projects or projects with fancy frameworks like llamaindex and langchain?
thanks for video. music is distracting btw
can this work with chat gpt 3.5?
In theory, yes. The only problem is that GTP 3.5 isn't as "smart" and has a much smaller context window so you will need to make your tasks much simpler.
@@bhancock_ai any chance of you doing ac video with a chat gpt 3.5 project? ❤️
The url is saying it was not existed
I'm not very far into the video, but just wanted to drop a quick note as I'm watching to offer some feedback. The glow around the dots and letters (e.g. 1:45) really messes with my eyes and brain, like everything is slightly out of focus. Okay, back to learning!
oh! Email sent 1/4 hour ago and still not received the source code, no it is not in my SPAM folder !!?
Hey! Feel free to shoot me an email at brandon@brandonhancock.io and I'll make sure you get the source code!
Thanks for sharing, without music would be better
Thanks for the feedback! I'll try to tone things down next time!
Hope CrewAI hires professional user interface experts.
funny thing is i just finished watching the first one.
Perfect timing 😂 what other CrewAI videos would be helpful for you?
I didn't get the code 😔
Good morning! If you didn't get the code, please shoot me an email at brandon@brandonhancock.io and I'll make sure to get it over to you.
Anyone else can't get this to run with gpt3?
I wish this was about something else than a news crew. Seen that 15 times already in the past week. So little innovation in the tech youtube scene.
Hey Mikael! Thanks for the feedback! The next tutorial I'm going to release will show how I've used CrewAI to automate most of my RUclips tasks. That video will be completely different from everything else you've seen on CrewAI so I'm sure you'll like it!
@@bhancock_ai Thanks for taking my whining in stride! :)
I concur! Would really be interested in an osint-ish demo where I could provide the crew with a social media troll account (Twitter profile link) and have them investigate and track down their network of troll accounts on Twitter and from other social channels 😬
wtf, people are teaching you basics and building blocks, so you can build your stuff and have fun
Have fun figuring stuff out, dude
I constantly having the message of “Sorry, I couldn’t find anything about that, there would be an error with sou server api key” can you help? I checked the key and its’ in the env file
COol
WTF is that flickering transition? Are you trying to give me a epilepsy?