How to Build AI Agent Teams (NO-CODE!) | Flowise Agentflows Tutorial

[@JohnMasse]

Tight video, straight to the point, and deep enough examples to help me break in.

[@leonvanzyl]

Thank you very much for the support ❤️

[@almirbolduan]

Thank you Leon for such helpful videos! Your English is also easy to understand for non-natives (I'm from Brazil)! 👏👏👏

[@leonvanzyl]

Thank you 🙏

[@kirilkirchev285]

Another quality content! As always it gives strong starter base to build on top. Great fan of Leon's channel for the last 12 months

[@spacedreams6566]

@leonvanzyl Awesome video! Thank you so much for your content. What I found particularly interesting is what you mentioned about creating workflows with Make. From this video, I understood that you can create Make workflows for agents to use as a tool. That would be fantastic! For now, I plan to do it the other way around by controlling our chatbot with a Make workflow.

[@ward_jl]

Very interesting. It would be even more valuable with when one can specify the order and a human in the loop, this will give more control over the outcome. I believe Flowise will add this in their next Agentflow release; another good tutorial topic IMO

[@lingzhang2061]

Couldn't agree more. I'm thinking about the same thing "human in the loop." I'm in the field of education, so very much focused on exploring how to ensure information generated by AI-human collaboration is 100% accurate so students receive high-quality instruction and learning supports.

[@ward_jl]

@@lingzhang2061 Nice. You could look at self-correcting agent flows; agent flows that check the quality and legitimateness of an answer before sending it to the user

[@ward_jl]

@@lingzhang2061 cool. You could use self-reflective agent workflows, where agents critique themselves to monitor and enhance quality of responses

[@ward_jl]

Cool. You could use self-reflective agent workflows, where agents critique themselves to monitor and enhance quality of responses

[@lingzhang2061]

@@ward_jl Thank you for the suggestion. I wonder if you could combine self-reflective agent workflow with multi-agent workflow?

[@rikitzp]

Concise, comprehensive, with great examples and easy to follow. Encore!

[@practical-skills-school]

Thank you Leon! I'd love to see how a "client" is chatting with the "sales AI" and is moving through the sales funnel to the sale.

[@jeremybristol4374]

Glad Flowise finally added agents! Does this work with multi-modal models or opensource for something like a SWOT analysis?

[@sarahroark3356]

This is extremely promising. I'm looking forward to playing with this after all the wrestling with other multi-agent/team frameworks (CrewAI, so far the closest to successful, Agency Swarm, AutoGroq, even SillyTavern). Since you invited us to submit use cases: mine may be a bit unusual, but I'm basically trying to create a civ design team that can assist a player in a 4X/grand strategy space empire sim game to craft a truly unique culture and polity. There's a Gamemaster, Counselor/Liaison, Artisan, Game Balance Guru, Scientist, and Lorekeeper/doc writer. The main tools they need are to be able to web search, incorporate a knowledge base, and write to a markdown file. Oh, and it'd be awesome if the Artisan could prompt an image gen app to make art. :-) Thanks for the video!

[@waynenunn7319]

Very nice video, and my compliments on the Google setup. It is one of the simplest, most straightforward presentations I have seen. Most are overly complex and difficult to follow, which usually results in failure. However, in the code generation example, I would have liked to see an example in which the code is executed to get some idea of the quality of the generated code. Thanks for another great video. Wayne

[@leonvanzyl]

Thank you Wayne, I appreciate your constructive feedback 🙏

[@guspoyatos4165]

Best video of the FLOWISE series so far. Love it.

[@MaliRasko]

Great tutorial Leon. Thank you.

[@leonvanzyl]

This is insane! Wow!! Thank you for the generous support ❤️

[@EcomSnif]

Hey Leon, Thanks so much for sharing this valuable information! I have a quick question: I noticed you hard-coded the values for variables like lead_company and lead_person. Could you show us how to allow users to input this information each time they interact with the chatbot? Thanks a lot!

[@xavierf2229]

What a bomb mate,i have finally developed my agency website and getting ready to create the ai agents to work for my agency . Yesterday i have implemented the chatbot made with flowise on my website successfully thanks to you boss ! Would be very appreciated if you make a video on how to make this agents to post the content like blog posts directly on my website blog through vscode. big hug

[@jorgemarques2585]

Fantastic as usual. How would one though would create a decision of a similar tool? For example, I would prefer the agent first to search in a document knowledge base before doing a Google Search (To save API calls) if the answer was not in documents! Should a simple instruction in the prompt be defined or should a sort of IF condition built on top? Would be great a tutorial to show how to choose the right tool to use when they are similar.

[@Alvaro-cs7zs]

Man, you're a legend. Another amazing video!! I would love to see more practical examples done with these agents. Love it!

[@leonvanzyl]

Wow, thanks!

[@jeffmulligan6032]

Leon, I love your videos, I love your tempo and I love your generosity… grateful for this portal into Ai possibility from a non-coding background. You are and continue to be the best in Flowise tuition.

[@leonvanzyl]

Thank you kindly!

[@Organizational_Psychology_PhD]

Thank you Leon. You always make it look much easier than I find it to be when I try it myself but I got it to work today thanks to you. Now it's time to iterate and make it great!

[@jhett1746]

This is very cool, though I really hope to see AI grow less into making companies money and more into making everyones life better, this was a super handy tutorial and i hope to see ways this could be adapted to delegate reminders, appointments, send emails etc just from a personal standpoint in the near future

[@HistorIAsImposibles776AC]

Thats great i already did some with crewai and directly with openai but this is pretty fast and practical, thanks !!!

[@leonvanzyl]

Exactly! I've also done plenty of projects using CrewAI and this blew my mind

[@professions_]

Thank you for this amazing video tutorial Leon! 😍Your videos are the very best because they are not only well explained but also based on real user (useful) cases. I'm so happy you are covering both Flowise and Lanflow.

[@leonvanzyl]

Glad you like them!

[@Ahmed-Sabrie]

Can't express my gratitude for your video .. much appreciated and looking forward to see more great stuff .. keep it up!

[@alfideklerk8790]

Excellent video Leon! Can you do a video on a regulated environment accounting?

[@YitzhakErezKarat]

Thanks, as always clear, detailed and professional. I would love videos with Llamalindex like RAG Chatboot.

[@Francotujk]

Leon your videos are always the best ones! They are extremely helpful!

[@leonvanzyl]

Happy to hear that!

[@WellPiresAI]

thank you so much, Leon! I was hungry for this one hahah

[@joebradford6024]

Leon, these videos are incredible. Thank you for doing this work!

[@TorstenHeissler]

Thank you Leon! Exactly for this I was waiting! Well done!

[@abcdBrugge]

Amazing! Thank you Leon. Would it also be possible for the agents to perform tasks on your computer? Like running terminal scripts, write and create files, ...?

[@rude_people_die_young]

Great ideas

[@subhamagrawal4740]

yes

[@Chasingaxl]

Great video👍. I would love to see using the flowise api connecting to gpt4o or vision to accept images. I can get it to work with the bot, but when I try to connect with the built in api for that workflow, I have issues.

[@stevewarby12]

In the ChatOpenAI node there is an 'Allow Image Uploads' does this help????

[@Chasingaxl]

@@stevewarby12 that works for the embed feature. I am having issues with the Api connection to my own bot.

[@stevewarby12]

I have mine working embedded and will be testing the API. I’ll post if I get any success.

[@Chasingaxl]

@@stevewarby12 me as well. 😂🤞

[@alistairtennant3299]

Hi Leon - great video. Please incorporate upserting a PDF document into Flowise and doing OCR on the PDF (one that’s been scanned or been locked via DocuSign) Thanks!

[@OzgurOzsen]

First of all, I really congrats you bro, your ideas are very useful thanks again. However, I am very curious about your opinion on the following as a custom agent tool structure: “For example, how can you plan a code evaluation agent structure?” For example, .java code will be entered as input, the execution of the code will be evaluated with a certain weight, the proper writing of the code will be added as an evaluation parameter, and the use of commands will also be added as an evaluation parameter, and ultimately a custom agent that returns a score. How do you think we can do it? You usually make web search tool examples, for example, wouldn't this be different? Wouldn't it be great if you prepared such a content video? I'm curious about your ideas on this subject!

[@sv_c6788]

This is amazing. Great video Leon 👌

[@leonvanzyl]

Thank you!

[@Bangs_Theory]

I've been waiting for this.

[@leonvanzyl]

You're welcome 😊

[@Bangs_Theory]

@@leonvanzyl Question: I created a flow for trading, however when I ask a question, the lead trader (supervisor) will repeat the question, but then I get a message stating "message stopped." . Would you happen to know why that is happening?

[@WisienPol]

You are doing history here. Thanks mate!

[@leonvanzyl]

You're welcome 🤗

[@kyunglee1924]

The supervisor agent used here is not great; and these auto flows depend on the supervisor to manage and direct. You can't tell the supervisor to follow directions, you can tell it what agent to start with not much beyond that. Hope we can get GUI Langraph for quick mockups.

[@jamesyoungerdds7901]

Hi Leon - huge fan, you are the best at these detailed tutorials, just consuming everything you're putting out, so please keep it up! The thing I run into is actually deploying things to production for our company/service. I love tinkering and spinning up services locally, and am a huge fan of this - but how do we actually deploy these things so that our users and customers can interact with them? Maybe embeds/functions on our websites, or in our Android and iOS apps. Can you dive into some kind of a persistent production cloud deployment to see how this might work? Specifically for me, I continually hit roadblocks when trying to deploy a system the works great locally on my PC, but craps out when trying to deploy to AWS. Thanks again!

[@christofschuermann3566]

This is a great video to understand the use of supervisor and worker nodes. Thanks to Leon for his excellent work. If have question regarding the context window of the underlying LLM, here GPT-4. Are the three worker nodes and the supervisor node working on their own LLM instances in terms of context windows? Or in other words: Do the different nodes (supervisor and worker nodes) have their own context window and can remember their own context even when doing iterations and are passing their results back to the supervisor that passes the result to the next suitable worker node? That would mean that the agent model is much more capable even if we have “only” a sequential execution of the worker nodes as in the example. Could that also result in less overall token usage because, e.g. in case of loops the worker nodes work on much less amount of tokens compared to executing this manually in a chat window where the context window hold the complete context for each task (design, develop and document)?

[@fredrik-ekelund]

Thanks Leon! Excellent as always!

[@leonvanzyl]

Thank you!

[@mikew2883]

Good stuff! 👍

[@TytanBot]

Thanks Leon for another fantastic tutorial.

[@leonvanzyl]

You're welcome

[@TomDavenport]

The king is back

[@leonvanzyl]

🙏

[@VincentVella-cv6jl]

Thank you, Leon! Once again, you did an amazing job with this tutorial! Regarding example 2, I have a question: how can we ensure that it is the user who defines the subject that will be handled by the Lead Researcher?

[@WayneBruton]

Awesome video Leon, this is way easier than the traditional methods. Is it as good as ay crewAI or Autogen?

[@leonvanzyl]

Well, it's using LangGraph under the hood, so the potential here is pretty much limitless. Definitely huge competitor to CrewAI and Autogen.

[@ward_jl]

@@leonvanzyl I have a preference for LangGraph as it neatly integrates with LangChain and LangSmith

[@leonvanzyl]

@@ward_jl Exactly how I feel as well. That ecosystem is incredible.

[@hosst]

Excellent video Leon, thank you! Could you please tell us how to retrieve data from documents using this multi-agent flow? It will be super useful to see how multi-agents interact with external data. Thank you!

[@leonvanzyl]

Will do

[@exlfactory]

Great stuff a usual. So I was thinking If I wanted to mix custom tools and conversational chain, would it be best to use agents or a mix of IF conditions?

[@traxtibidox5775]

Awesome 👍❤ Leon thank you for this tutorial it's unbelievable 🎉

[@leonvanzyl]

Thank you 😊

[@sathyanarayananramamurthy5681]

Hey thanks for the wonderful video and flowise is awesome, just thinking to see if this agentflow can be used as virtual technical agents (l1/l2) and do some troubleshoting when an system alert is generated.

[@MarufYunus]

Great video, I would love to see if you can show how to use perplexity ai search as a tool or even use as llm inside flowise. Thanks

[@IBBOR-RB]

i would love if you could provide an explanation about how to use the curl api with the overrideconfig option

[@leonvanzyl]

Already have you covered! kzbin.info/www/bejne/b4OYq6Jpfruep7c

[@IBBOR-RB]

@@leonvanzyl ok i will check it out. another thing is when i copy paste the html code to my website the ui is not the same like inside the flowise canvas. why is that? like in my website i dont see the supervisor speaking and telling what to do to say...

[@Introdigitalarts]

we would appreciate if you could do tutorials the integration between Zapier and Flowise

[@celsodeccache6268]

Thank you Leon! Is it possible to draw a chart with data or generate an image of it?

[@turabmehdi1632]

Leon, thanks for creating this video, it is awesome. I do have a question, why does the supervisor only request the developer to create part of the functionality?

[@4pass115]

Amazing!!

[@christofschuermann3566]

How would the flow be enabled to work with any company, lead_company and lead_person? So it would have to extract the values for the variables from the user’s request.

[@jagusiff]

Fascinating! sparking so many ideas/concepts!

[@TomPooleMS]

Question - Can you use Vertex/Gemini for the LLM in Flowise?

[@therodrigofaerman]

Great video Leon... New to your channel and already a fan... There are so many solutions nowadays that it could become confusing on what tools and approachs to follow... So, I am wondering why would we use Agent Teams vs a Sequencial LLM (assuming that in both cases, there is an expected flow of actions)? What is your feelling about this?

[@leonvanzyl]

It's hard to answer in a comment, but we've already found many use cases for this in my profession. I hope to provide enough examples on my channel to show the pros and cons of the different approaches.

[@GilbertMizrahi]

Fantastic! Thank you.

[@mtin79]

Awesome, would be great to see how to add custom tools for the agents using e.g. amazon aws bedrock RAG as a retrieval tool. The other thing is how to extend or customize flowise to add langchain functionality that is not yet in the flowise UI e.g. again amazon bedrock RAG ...

[@leonvanzyl]

Great suggestion. I am planning on doing a RAG video with Agentflows, but it's most likely going to use the new Uptash Vector solution (which is incredible). I'll also do a custom tool example where we call a Make.com endpoint to send email / update a database or something. I do want to spend more time looking into AWS Bedrock though.

[@mtin79]

@@leonvanzyl Sweet! I guess the AWS Bedrock RAG is great as it automatically embeds all documents in an S3 bucket. But I liked that VectorShift does recurring URL crawls for generating the knowledge base. Just prefer having the LangChain (JS) ecosystem in the background why I currently prefer Flowise for creating PoCs and eventually transferring those to a custom (web)app.

[@javierlahoz5462]

Hi good morning, great video. I have one question, when I create a conversational RAG QA chain uploading a PDF file, if I ask something and this info is out of this PDF, the box is always answering me: " I am not sure", because he didnt find the answer in the queestion. Could be possible to conect RAG and normal ChatGPT and only use the RAG in case is needed?

[@bambanx]

What is the benefit of use múltiples agents? Thanks you Leon

[@bismamajeed7805]

Super helpful video @leon.. I want to give the worker some custom data via text file. How can I do that? I did try connecting another chatflow but it stops the main agentflow. Awaiting your response.

[@sunnynie4290]

Nice, Can you allow worker to pick up value entered dynamically from user? for example : put prompt some questions and use the answer from the user

[@TrinaThompson-dl7zw]

Awesome video! Thanks for this great content! Wondering if it is possible to format the output like we can with Chains? Also, there is a node selector on the Worker right below Output. I haven't been able to figure out what you can hook that to. Any ideas?

[@leonvanzyl]

Thank you! I've reached out to the Flowise devs and the worker output doesn't do anything at the moment. It will in the future. Hmm, I'm sure there must be some way to format the output.. lemme look into it and create a video. So basically you want to use an output parser on the agentflow results?

[@TrinaThompson-dl7zw]

@@leonvanzyl Thanks for the info! And you got it, I want to be able to format the output of the end (agenflow) result. In my use case, I'm wanting a JSON object.

[@makiprodan]

Hi Leon! Congratulations and thank you for more this video. Is it possible to use memory in this flow? I used the in memory cache of the chatopenai node, but it didn't seem to work. Thanks

[@leonvanzyl]

Thank you! I think the memory functionality is coming soon. Think I saw the Flowise devs mentioning it somewhere. I'll definitely create a video on it.

[@FlippieCoetser]

Is it also possible to produce code, or JSON or markdown as output and copy that to a speccific folder on your local machine? BTW Great Video!!!

[@leonvanzyl]

Yes, there is a tool that can write to the filesystem

[@gerhardvolkl9736]

Thanks. Did you make any tutorial using flowiseai and Dall-E?

[@nkofr]

It would be awesome if you could show a way to iterate on result in a way like "Artifacts" of Claude3.5. (if it's possible)

[@SteveHodges]

Cool! Instead of search engine could we use perplexity?

[@HarshavardhanGurramYT]

Super cool. As always. Just a quick query... Why is the temperature 0?

[@leonvanzyl]

I just found a temp of 0 to work best in the moment. Ie. I wanted the models to stick to the prompts. I've been using 0.4 lately as well.

[@LLMFacile]

Love your work, Can you enhance the lead researcher, with a lead prospecting agent, and other agent that send /draft emails or DMs and add them to google sheet for tracking?

[@leonvanzyl]

Absolutely!! You can use a custom tool, and in the tool you call a Make or Zapier endpoint.

[@creatiiveart341]

What's the difference between Flowise, LangFlow, and VectorShift? Are there any other similar WYSIWYIG AI dev tools?

[@sitedev]

Not working with local models yet (at least not for me). I wonder if you can create a chatflow node from an Agent chatflow and use it inside another Agent chatflow to build out a team of agents approach?

[@exlfactory]

Was just making the tutorial, and noticed there is no chat memory. Is this a featured in need, or something that needs to be done differently than a conversational retrieval agent? Thanks.

[@brylie]

Does the model generated app run? Does the documentation describe the actual functionality of the generated app code? Please show how you tested the model output 😊

[@unrealnoobs656]

Hi can you pls try some usecases with API Chain?

[@pedrobarros2554]

Can i fine tuning or its just RAG yet? Thank you for your service!

[@cdyao28]

How can I use flowise to access pages of google doc or google sheets that required google account login?

[@NickTugarin]

Nice Video, very helpful ! How can we use RAG for the workers?

[@leonvanzyl]

Thanks! Simply add a retriever tool and document store to the worker 👍.

[@NickTugarin]

@@leonvanzyl thanks, your content is pure gold 😃

[@leonvanzyl]

@@NickTugarin thank you Nick 🙏

[@Amr-Ibrahim-AI]

Thank you for the great tutorials, Leon. I tried to upgrade Flowise (I am using docker compose on a Linux VM). It was upgraded but I still cannot find the menu item for Agentflows. Any idea how I can enforce this? Thank you so much.

[@adanpalma9160]

thanks Leon... I am trying to create worker for search on diferents pdf, every pdf has its own topic and i want supervisor call the right worker based on user question and topic. Can worker searchs or chats with pdfs I tryed using tools but it didnt work?

[@JulianHarris]

Thank you! Question: these agents are interesting but I still struggle to understand what problem it solves. What do you do with the output of this flow? I get the concept of automation but to me it’s only useful for a codified sequence of steps that happen a lot and are stable.

[@unrealnoobs656]

Great Video! How to add a memory retriever for all my previous chats in Agentflows?

[@leonvanzyl]

Thank you! I don't think memory is available in Agentflows yet. It should release in the coming weeks

[@KshitijShah89]

What happens if the responses are too long or get cut off? Does the Supervisor do something to circumvent that?

[@bekakakachia147]

Hello Leon, how can we control sessionid, when exporting flowise workflow via API, in Telegram bot, so it can have unique session id with each user id? Thanks in advance ❤ Because every time user sends message sessionid fluctuates.

[@sahuntjr]

Good work leon. As always above and beyond. Got a quick question. how do I update my instance of flowise using Github. Can't find that anywhere. Thanks.

[@leonvanzyl]

There is a sync fork button on the GitHub repo, usually near the fork button.

[@sahuntjr]

thanks leon for responding. this the thing, i made the repo private then made it public again and i dont have the sync button. I may have to start over.@@leonvanzyl

[@juliovac2113]

only thing that makes me not switch from crewai is that flowise is built on js it makes it so much harder to have flowise function call a custom python tool

[@WayneBruton]

Hi Leon, quick question, is there a way to inject the prompt values (lead_company & lead_person) from the users side? I am trying but cannot seem to fathom it out.

[@RoniBliss]

+1 on this!

[@Francotujk]

Is it possible to generate the outputs from the agents in a format like JSON (that it's more helpful to power external apps)? Or currently it just supports this chat type app? thanks!!

[@RoniBliss]

There is a tool for parsing Json

[@sivakumarm3569]

Is it possible to create these agents with multiple API? Like allowing others to use their own API keys?

[@micbab-vg2mu]

thank you:)

[@Jonzybeatz]

hi leon, is it possible to trigger workflows (make or n8n) from a worker and use the result as input? For example, if I create a workflow in make whose purpose is to scrape the yellow pages. Can I create a 'yellow page scraper' worker in flowise and give it the prompt: use this worker when you need to scrape yellow pages, then retrieve the responses and send them to “worker 2”? thanks !!

[@leonvanzyl]

Absolutely! You can create a custom tool in Flowise which calls the make or n8n webhook and returns the response to the LLM. I've actually done just that in some of my videos 😁

[@Jonzybeatz]

@@leonvanzyl It sounds incredible, so that means Flowise is much more powerful than N8n for all topics related to artificial intelligence. Because I started looking into the LangChain integration within N8n. But with what you're telling me, I think I'll switch to Flowise instead.

[@Jonzybeatz]

@@leonvanzyl btw what are the videos that you are talking about? So curious to watch that :)

[@toi500]

Leon, a question: Why hardcode the variables into the WORKERS as you did in the EXAMPLE 2, not allowing the end user to let the Supervisor know what the research is about, by providing the leap company and leap name?

[@leonvanzyl]

I just wanted to show two different ways of passing instructions to the flows. One major benefit of using variables is that you can dynamically pass the values using the Flowise APIs.

[@RoniBliss]

@@leonvanzyl This! Trying to work out how to go about passing the variables through the API so they can be dynamically input in our SAAS Bubble app...If I leave them as is wehn testing inside flowsie it comes back as error "Worker variables not set"

[@TheNexusDirectory]

Thanks!

[@leonvanzyl]

You're welcome 🤗

[@udnvincent]

謝謝！

[@leonvanzyl]

Thank you very much for the generous super 🙏

[@chibong]

How do i pass the agent the pdf ? I am using a text splitter and in memory vector store, and conversation retrieval QA chain but can't figure out how to link it to worker or supervisor

[@leonvanzyl]

I'm working on a RAG and Multi agent video 👍

[@tecnopadre]

Did you find out what is the link called "worker" in the bottom down right side of the worker? Where can a worker be connected to?

[@leonvanzyl]

Good question!!

[@tecnopadre]

@@leonvanzyl hahaha Let's find out why. Maybe I'm moving to Langflow or other. Thank you for your support.

[@TytanBot]

Hi Leon, I have followed everything down to the T, but all of a sudden, I get a message in Flowise "message stopped" I have updated Flowise to the newest version in Elestio, but that message just keeps comming back whenever I try and test. I have done the agent flow several times just to make sure, I didn't forget anything. Thanks a mill for your help.

[@leonvanzyl]

I believed they added a fix as part of v1.8.2. Email me the chatflow and I'll pass it on to the FW team 👍

[@TytanBot]

@@leonvanzyl Thanks so much. What is your email address?

[@TytanBot]

@@leonvanzyl thanks so much, all sorted.

[@brians2364]

Can you attach a tool that can retrieve data from RAG database to an agent worker and then pass the results to another agent worker?

[@leonvanzyl]

Yes! Working on that video. Just trying to include some useful tips to improve supervisor as well

[@Gesssy]

@@leonvanzyl Thanx. Audience is waiting.