I’m going to experiment with the following: - Langchain/Langsmith/Langserve - CrewAI - Autogen & Taskweaver - Openinterpreter - Agency Swarm (with Instructor) …and build a few things.
@maverick1901
8 ай бұрын
@____2080____: can share some insights on Autogen & Taskweaver?
@vijaytilak5988
8 ай бұрын
Interesting! Which combination works the best?
@zaesarius
8 ай бұрын
Awesome. My research is by combining agents with Autogen, CrewAI,... but none seems to work with Open Source LLMs and RAG. The one that seems to solve that is LLamaIndex making custom models and CrewAI (joao) is aware of this.
@mrd6869
8 ай бұрын
im workikg with swarm agents....u can do some nasty stuff if you know how to co ordinate how they move
@florianworfel4131
7 ай бұрын
@@zaesarius I got Autogen (Studio) and LM Studio to work. Implementing RAG as a skill is my next to do. Did you get LLamaIndex and CrewAI going, or are you expecting it to work?
@sevilnatas
8 ай бұрын
Incredible, if only someone actually showed it working completely and not failing in the middle, but quickly clicking away. I love watching these AI framework demos and what seems possible, but I fear that a lot of this suffers from the typical Hello World tech demo issues. Not once have I seen a demo that shows the functionality and then actually analyzes the results, to see if they were what is being promised. Often, the fact that it came back with a response is used as proof that it is working, without actually evaluating if the response is appropriate and useful.
@MervinPraison
8 ай бұрын
Thanks for your suggestion. I will try to show and analyse more of the results if I could squeeze it in, without making the viewers bored. For LLM Response Evaluation there are various other tools which we can integrate with this code. I am trying to make one in the near future.
@ABAdams-gl1gy
7 ай бұрын
Seems like full project demos are to be found on courses ie udemy etc. All one sees on most similar clips on KZitem are, as you said, mere proof of concept which's what the documentation is for. Personally, I go through them using chatbots as a learner.
@TelB
8 ай бұрын
Tremendous videos and instruction - Thank you Mervin. Your channel is one of the most important in AI on KZitem. Please keep these videos and such clear instructions coming. I found this particular video so insightful. Thank you for dropping into the detail and also for providing the code. I ran the code successfully a couple of times but then hit rate limits on gpt-4 (with API Key).
@MervinPraison
8 ай бұрын
Thank you . Will do. Limit will be lifted once you pay enough for OpenAI API usage.
@TelB
7 ай бұрын
Hi @MervinPraison , Thank you. Yes, I do already. I have an annual Team-GPT. It turned out that I changed the model from 'gpt-4-turbo-preview' to 'gpt-4' to see if there were any quality changes. 'gpt-4' is limited to 10000 TPM and 'gpt-4-turbo-preview' 15000 TPM respectively. The query was running at 11241 TPM so exceed the gpt-4 TPM rate limit. Got a langchain API key too and it is now working great - Thank you again. Oh, btw - I have now joined you as a member :-) Thank you.
@MervinPraison
6 ай бұрын
@@TelB Great! Thanks for joining :)
@karthikeyakuncham6929
6 ай бұрын
Clean and clear explanation.... we expect more and more videos from you on langgraph..
@crippsuniverse
8 ай бұрын
I love it when I find a new channel that deserves a subscribe! Great content.
@MervinPraison
8 ай бұрын
Thank you :)
@alessandrosalvaggio6772
7 ай бұрын
a few months ago i made an api that does that in python, by combining the first 5 google search results. nice that now there is another way to make it in a more precise and functional way
@marekjkos
8 ай бұрын
Good channel with all the latest multiagent systems. It would be much better if comparative analysis between these disparate systems is performed.
@MervinPraison
7 ай бұрын
Thanks. Sure will try to incorporate all together when I get sometime
@theaiffice
8 ай бұрын
Really interesting. Amazing how succinctly you were able to build out an example use case.
@gold-junge91
8 ай бұрын
no it is not. what he is not telling is that langchain is useless in production use.
@maverick1901
8 ай бұрын
@@gold-junge91 : can you elaborate on why langchain is not usable in production?
@TelB
8 ай бұрын
@@gold-junge91why do you keep on repeating this? I am sure that he is not deliberately '...telling is that langchain is useless in production use'. Your comment is not helpful in the least. We can see the limitations - Thank you.
@AntonBj3
8 ай бұрын
@@gold-junge91 Explain or point me to a video that supports your claim
@42svb58
7 ай бұрын
Great video! How can we get an agent to reference a vector store database as the first default tool and then have another agent perform an internet search or use another tool for supplemental data?
@micbab-vg2mu
8 ай бұрын
It looks more complicated than Autogen and CrewAI but more powerful - I have to try it - Thank you:)
@gold-junge91
8 ай бұрын
no it is not. what he is not telling is that langchain is useless in production use.
@TelB
8 ай бұрын
@gold-junge91 why do you keep on repeating this? I am sure that he is not deliberately '...telling is that langchain is useless in production use'. Your comment is not helpful in the least. We can see the limitations - Thank you.
@augmentos
7 ай бұрын
Can y'all elaborate? @@TelB
@clt7640
4 ай бұрын
excellent video. can you create a video on Langraph but using OpenSource model and instead of going to the Internet, do a search on a folder of documents before passing it to another agent to summarize the content. thanks.
@mayorc
8 ай бұрын
Question, is that chat ai guided code completion suggesting all the code you wrote during the video, cause you basically typed nothing. Was it trained on a code file you already completed or was a sort of code replayer?
@greatsarmad5147
8 ай бұрын
Brilliant! Can't wait to see implementation of new tools, also which open source model would be best in LM studio in your opinion? Gpt4 still the goat but i am curious.
@MervinPraison
8 ай бұрын
Thanks. Open Source Models via LM Studio are still yet to catchup with GPT4. Models via LM Studio can be used for basic tasks, installed locally on your computer. For now you could try Mistral model. But this might change next month, as AI is a fast growing space.
@rmasineni
7 ай бұрын
Thank you Mervin. Nice and informative video!!! 👏👏👏
@shuntera
8 ай бұрын
Got the following error, I wonder if this function has been deprecated: Traceback (most recent call last): File "c:\Users\shunt\python\langgraph\app_gradio.py", line 124, in inputs = gr.inputs.Textbox(lines=2, placeholder="Enter your query here...") AttributeError: module 'gradio' has no attribute 'inputs'
@MervinPraison
8 ай бұрын
Please try a different version of gradio. some version have a slight variation.
@jeromenicolaes3243
7 ай бұрын
This is awesome, I can think of a lot more use cases then just search. Would this also run with mixtral? So we can test locally?
@MervinPraison
7 ай бұрын
Yes you can run locally using Mixtral if you have a good spec computer
@BradleyKieser
7 ай бұрын
Another great tutorial. Excellent work.
@MervinPraison
7 ай бұрын
Thank you
@antiquesordo
7 ай бұрын
Can you use an open source API instead of OpenAI?
@MervinPraison
7 ай бұрын
Yes you can integrate open source models
@MehrnoushB.Shirzad
2 ай бұрын
I am trying to run the code on my computer, I get error: Anaconda3\lib\asyncio\events.py", line 642, in get_event_loop raise RuntimeError('There is no current event loop in thread %r.' RuntimeError: There is no current event loop in thread 'ThreadPoolExecutor-0_0'. is there any idea of handling this error?
@NetZeroEarth
8 ай бұрын
Thank you Merv!!!! This was very insightful!!! What do you think of Chainlit interface?
@MervinPraison
8 ай бұрын
Thank you. Chainlit is also good if you want to convert this code to something like a chatbot
@herpityderp
7 ай бұрын
Anything like this that does not require OpenAI?
@MervinPraison
7 ай бұрын
You can modify this code to use open source LLM
@israeabdelbar8994
7 ай бұрын
Very helpful video, thank you a bunch! you saved my day. However, I'm wondering if you plan to create a video to create an IA agent doing actions. I mean a chatbot which can talk to the external application using the APIs and do the actions providing by the user!
@JacoduPlooy12134
7 ай бұрын
Interesting, but I’ve found that using langchain alone results in too much manual boilerplate over using it in conjunction with autogen (since the two are interoperable). Maybe it might be a good idea to make a video showing what can be done when using autogen together with langchain in the case of, for example, tool use in multi-agent applications?
@joser100
7 ай бұрын
A table comparing the different agent frameworks would be quite interesting to see, what is CrewAI excelling at over this one for example, I know it is difficult since CrewAI is still very active improving it, but maybe we can help them all by pointing at what aspects are better than others for each framework. I understand that CrewAI was good (better than Autogen anyway) with the workflow definitions, from listening to the video those workflows are done here with Graphs and looks quite flexible, but I assume that Autogen (or swarms or others) may also have some edge that would be of interest to the others. This afternoon I watched the video from Jason about his $5000 spent in an single afternoon with GPT-4 due to a misbehaving agent swarm, I think having access to local LLMs would also be important, and yes, Mervin, your previous video about CrewAI using Ollama with several local LLMs was excellent also...
@joser100
7 ай бұрын
LOL, you gave a short comparison in the last minute... thanks... I still love CrewAI, will use this one too to try to compare
@loknathbasak1533
7 ай бұрын
Thank you much!, Can you please suggest how can I request for LangSmith API_key please? Appreciate it!
@codeplaywatch
8 ай бұрын
Thank you for great video day by day !
@MervinPraison
8 ай бұрын
Thank you
@avg_ape
7 ай бұрын
Thank you for sharing this tutorial & insight. It is very helpful.
@MervinPraison
7 ай бұрын
Thanks
@user-wr4yl7tx3w
7 ай бұрын
it seems to be far more complicated than CrewAI. do you think it could be designed better?
@manuterui1136
7 ай бұрын
Can I swap ChatGPT with an open source model? If yes, how would I do that?
@virgilbarnard4343
4 ай бұрын
Is there a graphical UI for LangGraph yet?
@asiddiqi123
8 ай бұрын
Brilliantly explained!!
@MervinPraison
7 ай бұрын
Thank you
@planplay5921
8 ай бұрын
I spent a whole afternoon studying the project, but finally found that the LangSmith API seemed to be unavailable to me.😭
@MervinPraison
8 ай бұрын
LangSmith is optional. You need to request access.
@shuntera
8 ай бұрын
Just change this line to false instead of true and it will ignore it: os.environ["LANGCHAIN_TRACING_V2"]
@GeandersonLenz
8 ай бұрын
How to build a RAG layer with langgraph? To Make calls and retriever.
@MervinPraison
7 ай бұрын
RAG possible via Langchain ecosystem. You can extend this code to integrate that
@ReubenJohn0
7 ай бұрын
I switched to gpt3 and get an error that it exceeded the context length. Does anyone know a way to add memory?
@haigiri1
7 ай бұрын
Very good Video - How can I do this for Gemini? Getting error on llm.bind_functions
@renierdelacruz4652
7 ай бұрын
Great video, thanks a lot, it´s great LangGraph but there is anyway to rut on Ollama ?
@MervinPraison
7 ай бұрын
Thanks, yes you can include ollama
@renierdelacruz4652
7 ай бұрын
@@MervinPraison can you create a Video?
@VermontStrolls
8 ай бұрын
Bravo, more of this great job, please !
@MervinPraison
8 ай бұрын
Thank you
@HyperUpscale
8 ай бұрын
🤯 This is too advanced for me, even if it has a Web UI 🤠 But I am amazed how can you recreate such a big project and go through it in a record breaking time 🤓
@MervinPraison
7 ай бұрын
Thank you :)
@helix8847
8 ай бұрын
How could we use Local Agents with this flow?
@MervinPraison
8 ай бұрын
You can use local Agents. Just replace llm variable with whatever open source LLM you want to use.
@kenchang3456
8 ай бұрын
Great walk through, thanks!
@MervinPraison
8 ай бұрын
Thanks
7 ай бұрын
Thanks!
@MervinPraison
7 ай бұрын
Thank you 😊
@thefutureisbright
8 ай бұрын
Nice Video. Could you do another one where it works with ollama. I couldnt get it to work. Thanks
@MervinPraison
8 ай бұрын
Here is the code, if you want to try: python.langchain.com/docs/integrations/llms/ollama
@thefutureisbright
8 ай бұрын
@@MervinPraison thanks. The issue I had was the llm.bind is not supported in ollama.
@user-wr4yl7tx3w
7 ай бұрын
Is LangGraph compatible with Ollama?
@changjunlee2759
8 ай бұрын
Really, Really EXCellent!!
@MervinPraison
8 ай бұрын
Thank you :)
@adamchan4403
8 ай бұрын
It’s amazing ! 😄
@MervinPraison
8 ай бұрын
Thank you
@TotallyFriedChannel
8 ай бұрын
Did you automate your video production or are these short form info packed videos that easy to produce in comparison to your longer form content? xD
@MervinPraison
8 ай бұрын
All my videos are generally like this
@alpineai
7 ай бұрын
Solid content
@MervinPraison
7 ай бұрын
Thank you
@alpineai
7 ай бұрын
do you have a discord channel?@@MervinPraison
@ctrade8837
7 ай бұрын
What a Man. Lots of love 😍
@MervinPraison
7 ай бұрын
Thank you
@karankapur3275
7 ай бұрын
Thanks !! Nice and crisp content Can this be integrated with chainlit?
@MervinPraison
7 ай бұрын
Yes, absolutely
@karankapur3275
7 ай бұрын
@@MervinPraison It would be great if you can share a video / code snippet for same As of now i keep getting error No running event loop
@andreatamburrini407
7 ай бұрын
Hi man! Thnks for sharing ;) Do you know if its possible run langgraph with local models?
@MervinPraison
7 ай бұрын
Yes It can.
@iham1313
7 ай бұрын
where did the references go?
@marcosoliveira_accounts
7 ай бұрын
Thank you! It was a very good class
@MervinPraison
7 ай бұрын
Thank you
@kenchang3456
8 ай бұрын
Do you have any use cases that involve human intervention?
@MervinPraison
8 ай бұрын
I need to look in to that.
@rajchinagundi7498
7 ай бұрын
Basically they opensourced Perplexity AI Under the hood code!
@AJ-lg4zr
8 ай бұрын
What do you use for autocomplete code?
@helix8847
8 ай бұрын
I believe he had the code above, so it was completing a lot of it to exactly what he had previously. Could be Co-Pilot
@justsabry
8 ай бұрын
Do you think that IS possible to 777 chmod to let the Ai autoself update the initial agents prompt ? And connect and automate multi agent between them?
@squiddymute
6 ай бұрын
an internet search that charges you for every search you make isn't going to go very far , we need this with open source LLMs
@RaviRanjan1989
7 ай бұрын
Can we use it with local llama2?
@MervinPraison
7 ай бұрын
Yes
@Airatgaliev
8 ай бұрын
RIP CrewAI
@sahil-shegokar
7 ай бұрын
Please create more video
@borislit
6 ай бұрын
Github link?
@phily8020-u8x
4 ай бұрын
doesn't perplexity ai do this already?
@neoglacius
8 ай бұрын
why do you claim it beats autogen?
@MervinPraison
7 ай бұрын
It beats based on usecase. If your application need more customisation on how agents should interact with each other, how agents should behave and extending it with Langchain tools, then LangGraph is preferred solution. For simple non customised solution , autogen or CrewAI is preferred
@thedoctor5478
8 ай бұрын
Autogen can do this as well
@maverick1901
8 ай бұрын
@thedoctor5478 In general, yes - however Autogen does not provide the pipeline definition flexibility. The GroupChat is pretty much predefined. I would be interested in a combination of Autogen & Taskweaver, which I have asked @Mervin Praison already a couple of times - I don't know why he doesn't want to have a showcase on that and compare that with langgraph?
@MervinPraison
8 ай бұрын
@thedoctor5478 maverick1901 is correct. @maverick1901 Too much updates in the world of AI each week. So consuming and producing is taking time. Will try to do Autogen & Taskweaver for you
@thedoctor5478
8 ай бұрын
it's all good. I am one of the contributors of Autogen. Many different configurations of GroupChat (including sub-groups, hierarchies, graphs, etc.) are possible. Ease of use and documentation could use improvement though. At lot is being refactored at the moment.@@MervinPraison
@vijaytilak5988
8 ай бұрын
@@MervinPraison, would be really interesting to see how autogen and taskviewer will work together and how it compares to langgraph and crew ai. All of these frameworks are powerful. But would like your opinion on what combination will work the best and will solve most use cases!
@vijaytilak5988
8 ай бұрын
Maybe if you do a video comparing all 4 and suggesting what works for production would be nice. Thanks much 😊
@76rkhanna
8 ай бұрын
Thanks Mervin, this is an excellent video. I am new to Langsmith & python in general so not sure if my issue is related to your code or my Python environment. When I run the code on my local machine (non-ui version) I get the error "RuntimeError: There is no current event loop in thread 'ThreadPoolExecutor-0_0'." Can you suggest what might be the issue. I am using Python version 3.9
@MervinPraison
8 ай бұрын
Thank you. Probably related to python environment, considering you are running the code as suggested in the video.
@76rkhanna
8 ай бұрын
@@MervinPraison what python version are you using
@cathywan6434
8 ай бұрын
Hey, I have the same issue (python 3.11). Did you solve it?
@dr.walidsoula
7 ай бұрын
@@cathywan6434 install this version from duckduckgo-search "pip install duckduckgo-search==4.2"
@jeromenicolaes3243
7 ай бұрын
This is might be because the async functions(tools) are not ran in threads. Modify the tools to use the event loop: I.e. @tool("internet_search", return_direct=False) def internet_search(query: str) -> str: """Searches the internet using DuckDuckGo.""" loop = asyncio.new_event_loop() asyncio.set_event_loop(loop) try: with DDGS() as ddgs: results = [r for r in ddgs.text(query, max_results=5)] return results if results else "No results found." finally: loop.close() @tool("process_content", return_direct=False) def process_content(url: str) -> str: """Processes content from a webpage.""" loop = asyncio.new_event_loop() asyncio.set_event_loop(loop) try: response = requests.get(url) soup = BeautifulSoup(response.content, 'html.parser') return soup.get_text() finally: loop.close()
@r4v1m5
6 ай бұрын
don ko pakadna?
@b326yr
7 ай бұрын
still not for us, the average enthusiastic users. so no..thanks we will hold on for now.
@8eck
8 ай бұрын
Now add billions of search requests per day and you'll get a bankruptcy very very fast. 😅
@MervinPraison
8 ай бұрын
haha
@shuntera
8 ай бұрын
Yes we need to see it use open source local LLMs
@khalidseid6112
7 ай бұрын
I stopped watching cause you didn’t show us how to install it!!!!
@Slimshady68356
7 ай бұрын
Just git clone dude
@ew3995
8 ай бұрын
too bad that i absolutely hate langchain
@gold-junge91
8 ай бұрын
but what you not telling is that langchain is useless in production use.
@seyisowunmi9091
8 ай бұрын
What production have you used it for that failed? Info?
@TelB
8 ай бұрын
@gold-junge91 why do you keep on repeating this? I am sure that he is not deliberately '...telling is that langchain is useless in production use'. Your comment is not helpful in the least. We can see the limitations - Thank you.
Пікірлер: 152