How to Build a Self-Improving AI with Agentic RAG and Flowise

[@leonvanzyl]

Hope you enjoy this video! Please consider subscribing to my channel for more Flowise videos ⭐. One more thing, please use my referral link if you do decide to sign up for Flowise Cloud. I'll earn a commission 🙂 . flowiseai.com/auth/signup?referralCode=LEONVZ

[@ibrahimhalouane8130]

I will. Is there a way to avoid throttling or hitting the rate limit?

[@MaliRasko]

Graet video Leon. Including the “wrong” path and explaining why it should be done diferently is super helpful.

[@leonvanzyl]

Thank you for the support! Yeah, I was thinking that approach might frustrate some people, but I really wanted to show the difference between the agent and tool nodes.

[@sdcharly]

@@MaliRasko you are right. It was really needed. Real time scenarios will need this

[@bernardofontes4530]

Brazilian guys watching u every weak!!! Congrats for the job. U are saving us kkkk Here we dont have many peoples exploring flowise/rag and correlats.... ty

[@leonvanzyl]

Welcome guys! Regards from sunny South Africa 🇿🇦

[@JPBCDE]

@@leonvanzyl dont know where that sun is ... its quite cold down here :D

[@aolowude]

Your guide are always great. Thanks as always.

[@leonvanzyl]

My pleasure!

[@plannedillusion]

Amazing, been watching you and implementing your solutions for almost a year.

[@leonvanzyl]

I'm honoured. Thank you for watching my vids ☺️

[@krishnankuppaswamy7553]

Was able to follow along and get the hang of it. Thanks again!

[@amarah99]

Brilliant as always. Thanks so much!!

[@MicaMica139]

can we have both self improving AI and auto update knowledge bases in one set up?

[@artur50]

Fantastic!

[@Independent_AI]

Another real world great tutorial, thanks! 👏🏻

[@leonvanzyl]

Thank you 🙏

[@misheltal692]

Thank you for the amazing content! I'm really enjoying your tutorials on Flowise. Could you possibly cover how to set up a database agent that can retrieve and update information, as well as use the database as a knowledge base? That would be incredibly helpful. Thanks again!

[@leonvanzyl]

Thank you! Database integration is a very popular topic. I'll create a video ASAP. PS. Which DB do you use? MySQL, Postgres?

[@alessandro.depoli]

@@leonvanzyl It will be a very interesting tutorial! Supabase (postgres) please. Thank you for all your videos

[@kreak100]

@@leonvanzyl mongoDB please as well!

[@JPBCDE]

@@leonvanzyl Ive given up on MS SQL. Im now experimenting with generating summarized data (almost Data Warehousing) and upserting that into MySQL and then querying the MySQL db, with not to bad success.

[@jorellecastillo9026]

@@leonvanzyl Supabase please!

[@memba-digital]

Thank you for your time spent in such a great video

[@leonvanzyl]

You're welcome 🤗

[@mikew2883]

Good stuff as always! 👏

[@leonvanzyl]

Thank you 🙏

[@0xAsharib]

Thank you so much for putting hardwork on creating these amazing tutorials ❤ btw, when n8n with flowise?

[@leonvanzyl]

Thank you! I'm working on a complete n8n tutorial series, so it will short another week or so.

[@bernardofontes4530]

Me and my friend are using N8N, Flowise, Lagfuse and Meta to create comercial agents....its amazing.

[@leonvanzyl]

@@bernardofontes4530 this is great to hear!! I would love to find out more about your Meta integration. Are you integrating into WhatsApp?

[@0xAsharib]

@@leonvanzyl that's amazing!

[@nwokobia]

Brilliant... implemented! Can you demonstrate WooCommerce integration please for order status and cancellation queries?

[@samcon777]

Love what you r doing here

[@leonvanzyl]

Thanks Sam ☺️

[@shaankhan9490]

Thanks for the content en explaination! Just one problem, it’s running extremely slow. How can I troubleshoot the problem? I think that the problem lies in the retriever which takes too long to process the data. Maybe something with Vector settings? Hardware is no issue, it’s more than capable to run even 70B models fast and with ease in Ollama. If you could help I’d appreciate it, thanks in advance.

[@romainharter7395]

Your content is truly fantastic! Could I suggest creating a video on how to maintain session history in embedded chats using Render? I'm having trouble locating where the history is stored and how to access it so that users can revisit previous discussions with my Agentflows. I can see the history in the Flowise "View Messages" menu, so it must be stored somewhere.

[@sdcharly]

Beautiful one again. well still trying to clearly understand how the agent node was not the right one..:( soon will figure out.. Would like to see some SQL based ones which can fetch and save data into SQL tables. maybe Oak& Barrel reservations?

[@leonvanzyl]

Thank you. The Agent node produced it's own response based on the documents that were retrieved. This wasn't useful to the Conditional Agent. The Tool Node on the other hand was able to pass the "raw" documents to the Conditional Agent, which the CA could use to score the relevancy.

[@sdcharly]

@@leonvanzyl thanks. Got it now.

[@LauraSrzic]

Love your videos

[@leonvanzyl]

Thank you!

[@JPBCDE]

brilliant video!

[@leonvanzyl]

Thank you!

[@St_Phoenix]

As usual, your videos are on point, thank you very much. Out of curiosity, do you give private classes or consultations?

[@leonvanzyl]

Thank you!! I wish I had the time 😔

[@servicosmkt]

Parabens pela explicaçao ! Você é sensacional, fala mais sobre agentes sequenciais usando os stage !

[@aibizadvisor]

So amazing video, Leon! I would have two questions: 1. How you prevent the chat from displaying all the flow logic (like what agent was called, what route was selected etc.) and just give the answer. 2. I am wondering if Agent Memory node should be added to the flow?

[@sukuna8547]

Under "Share Chatbot" is an option to turn off "Show Agent Reasoning".

[@josephperkins8766]

Please do a detailed video on the document store and how to query a structured DB using agentic flows

[@akhnt1]

Thanks Leon! How could one iterate through each of the retrieved documents and answer a question using an LLM and then filter from those answers what one needs?

[@leonvanzyl]

I think you might need to combine Flowise with a process automation tool for that.. like Zapier, Make or n8n. Unless someone in the comments has a better idea ☺️.

[@blakeyang7893]

thank you for your Flowise course. It's awsome. btw, the readFile and WriteFile tools are not available right now?

[@leonvanzyl]

I used the writeFile tool in one of my other Agentflow videos. Are you looking at the correct menu?

[@blakeyang7893]

@@leonvanzyl Hi Leon, thank you for your reply, I watched all 24 flowise courses in your playlist. but didn't see about the Writefile tool. could you share me the link or number you mentioned course? thanks again. have a great day.

[@Done4YouPrompts4Sheets]

Thanks Leon. How does this relate to structured outputs from open AI? Is this a no code alternative?

[@Chill3dKr0ete]

Great Stuff Leon! But how do you implement this into a real chat? When I try to share my Chatflow it will give always the Agent/LLM Nodes before the final answer... Is it possible to just show the final answer to the user, without having all the system notes before?

[@Gwin-AI]

It all makes a lot of sense, but I am still struggling with some of the concepts used. How does the Retriever actually retrieve information? Is it always only one piece of data? Or several? How does it decide where to cut off? I mean for any question more complex than a simple Q&A document I can't resolve it by only getting one piece of data? How do I assess in a loop all retrieved chunks for relevance? Do I loop it? Isn't that extremely costly?

[@colibrideplaya]

Hey Leon! I would like to pick your brain about Flowise capabilities. We're looking to create a highly interactive and human-like conversational bot, that can handle multiple user intents and different data inputs like buttons, single-select, multi-select, loops, conversation branches, etc. How would you approach such project? I haven't found a way inside Flowise to capture different type of inputs and manage different flows. For example, if I need to capture specific dietary options like "Keto, Paleo, Vegetarian, Vegan" to also save it on a database, I would like to present a single-select input or buttons, but I don't find a way to render such inputs in Flowise. The only way I can think of is doing some LLM prompt engineering to perform an entity extraction, but I'm concerned that might rack up token costs, since I have many discrete questions. The other option I'm considering is using a more conversational tool to capture the inputs and pass them to Flowise. Any guidance would be tremendously appreciated. Thanks!

[@leonvanzyl]

I would agree on the second option. I've found that it's best to build an application around Flowise. That way you have way more control over the user interface itself, and you can also deal with user authentication, etc. You can invoke the Flowise API for any AI related logic.

[@pavanj2238]

@leonvanzyl Great videos.. loving them... 1. I have two questions... can we connect to corporate LLM gateway using flowise ui? 2. tried the Ollama tutorial with flowise, I configured everytghing as shown in the video.. but when the chat is started.. it throws an error as "fetch failed".. checked logs in dev console... sometimes it shows an error as unauthorized .. my flowise ui has no credentials.. and ollama is running fine and doubled checked that with the url configured in flowui.. any idea why am i getting "fetch failed" in chat window

[@undisclosedlocations]

The Ollamas aren't very powerful LLM models and a lot of my flows that I'm sure are working can be debugged when I choose a beefier LLM. Even if I go with OpenAI gpt-3.5-turbo vs gpt-4o-mini some of my bugs and errors go away. Ollama I can almost always guarantee my flows won't work, I use it sparingly As far as question 1 goes, I'm pretty sure you can. There's a lot of customization, including POST and GET requests

[@adityachoudhary151]

Thank you, I am not a programmer but your tutorial helped me in installing flowise and building the agents. Can you please help me with one question - I am trying to build a workflow that prepares a report in a standard format using the data that is available in SQL database(the data is very large). can you guide me what should be the best floe architecture for me?

[@leonvanzyl]

You're welcome 🤗. I've actually received a lot of requests to cover a SQL flow. Better create it ASAP 😜

[@samcon777]

Love to see a video how to use the python interpreter

[@leonvanzyl]

Will create a video on it at some point 👍

[@conneyk]

Thank you for this video! I‘m wondering if some ollama powered llm like llama3.1:8b is capable of doing the agent jobs. Is this possible or is the model not „strong“ enough?

[@leonvanzyl]

You would need to use at least Llama 3.1 70b 👍

[@jootjemootje]

Thnx for this one 👌 If you use this chat in a website is it possible to hide the agent messages so only the final outcome will be shown in the chat window? And what do you think of the response time? Because (as far as I know) there is no streaming option yet for the sequential agents it takes some time before the user gets the answer.

[@leonvanzyl]

You're welcome 🤗. I'm also not aware of a streaming option yet. Would be awesome if they could add it soon.

[@jannufamily]

@@leonvanzyl Is there no way to hide the agent messages?

[@achmadathoillah8152]

Great video ❤. What if i dont want the user to see what is the agent doing? I think it is inappropriate if i have maybe something like restaurant customer to see all of the background message. Is there any feature to hide it?

[@sukuna8547]

Under "Share Chatbot" is an option to turn off "Show Agent Reasoning".

[@GilbertMizrahi]

Great!

[@prosedox]

could we delete or hide the steps of nodes which we step on chat history ?

[@ocraminator7247]

Hey Leon, I run into an infinite loop when asking something like "what is the price of beer?". On the other hand a question like "Do you have beer?" will only be rewritten once and then an response is being generated. What's now?

[@seanolivas9148]

What are your thoughts on n8n workflow agents?

[@leonvanzyl]

They're awesome. I'll be releasing an n8n agent video soon.

[@seanolivas9148]

@@leonvanzyl Cool, would be good to see if you can make conditional agents like this in n8n since I kind of bought with the rest of my workflows

[@iokinpardoitxaso8836]

Amazin tutorial as always Leon. Quick question or doubt about this exact case. What would happen if the user ask with common conversational questions like "Hello", "How are you?", "I would like to start the conversation".... Thanks in advance

[@leonvanzyl]

There are plenty of ways to deal with that case as well. Could simply be another condition. The end condition could simply link to an LLM node.

[@bambanx]

Self host can be used for comercial use for free? Thanks you

[@leonvanzyl]

Yes, you can use it commercially and self host.

[@servicosmkt]

Você tem alguma comunidade para assinantes, exemplo Discord?

[@leonvanzyl]

I do, but it's been frustrating keeping it clean. Too many spam bots 😬.

[@servicosmkt]

@@leonvanzyl Aqui no Brasil voce esta se tornando referencia nas comunidades de flowise, nao tem ninguem aqui que cria conteudos sobre a ferramenta.

[@SebKrogh]

So generally - do you recommend flowise or Vectorshift for someone starting out?

[@leonvanzyl]

Flowise is a really good platform to start with. It's very stable and I've used it for my own projects. VS offers automations though, and easier integration with 3rd party tools (almost like Make or n8n), which Flowise doesn't.

[@SebKrogh]

Righto - i am looking for something able to replace make but also work as like an ai first platform. So I guess VS is the way to go then.

[@wesleymogaka]

Hello. I'm learning lang graph in parallel. In your opinion, is flowise as powerful and configurable as lang graph?

[@leonvanzyl]

It's very close. Flowise Sequential agents use LangGraph under the hood, so it's also a great tool for visually learning LangGraph.

[@aibizadvisor]

Can I build the same agent but with Llama and Groq? Any limitations in this case?

[@leonvanzyl]

The Llama 3.1 70b model seems to do a good job with these Agentflows. Definitely not at OpenAIs level, so you might experience a few funnies.

[@Blooper1980]

Can you use vision with flow wise? As in image input?

[@leonvanzyl]

Absolutely, if the chat model supports vision, you can simply enable to image upload toggle on the chat node. You'll then be able to upload an image in the chat.

[@micbab-vg2mu]

thanks

[@mingyukang6592]

Is there any difference using between agent node and LLM node?

[@leonvanzyl]

Agents tend to produce better results, since they're able to self reason. LLM nodes are great for quick, once off executions, like sentiment analysis, etc. For chatbots or more advanced apps, just use agents.

[@mingyukang6592]

@@leonvanzyl Wow, Thanks's very much. I really appreacite your works. ^^

[@HermesMacedo]

How to make Flowise send Media (image, audio, video, PDF and other files) during the conversation and not just links? $$ for example: Get information(media) from a Google Drive, and sends the media, not the link or tumb. I pay you to teach us😰

[@leonvanzyl]

Perhaps you need to look at platforms like Zapier, Make or n8n for your use case. I'm actually working on an n8n series by the way. AI Apps are not really intended for passing files / data along in the way you described it.

[@HermesMacedo]

@@leonvanzyl yes, I realized that I have to use n8n. Maybe this could be your class? how$$

[@oknoobcom]

Hey Leon. Amazing content as always. A quick question on Agentflows. How you prevent the chat from displaying all the flow logic (like what agent was called, what route was selected etc.) and just give the answer. In a chat environment what doesnt look good.

[@asrulmunir]

I got this when using Groq: Error buildAgentGraph - connectedToolNode.data.instance.node.seekPermissionMessage is not a function

[@asrulmunir]

When using Claude: Error buildAgentGraph - Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"tools.0.name: String should match pattern '^[a-zA-Z0-9_-]{1,64}$'"}}

[@asrulmunir]

SOLVED. Google Embeddings need some adjusting 😁