Build AI Persona Chatbots using Flowise AI

#flowise #langchain #openai
Get expert advice from celebrities and influencers at your fingerstips by building an AI Persona Chatbot.
☕ Buy me a coffee:
www.buymeacoffee.com/leonvanzyl
📑 Useful Links:
Git: git-scm.com/downloads
Node: nodejs.org/en
Flowise Repo: github.com/FlowiseAI/Flowise
KZbin Transcript: youtubetranscript.com
💬 Chat with Like-Minded Individuals on Discord:
/ discord
🧠 Let me build your chatbot for you!
www.cognaitiv.ai
🕒 TIMESTAMPS:
00:00 - Why AI Persona Chatbots
00:40 - Prerequisites
00:53 - Setting up Flowise
02:29 - Preparing persona chatbot data
03:59 - Create Flowise Chatflow
07:49 - Adding OpenAI API Key Credentials
09:25 - Setting the personality system message

[@leonvanzyl]

Enjoy this video! When creating Persona Chatbots for clients, please approach it in an ethical and respectful manner 😊.

[@AndreasKunze]

In a word: brilliant. The best reason to use Flowise are these videos.

[@leonvanzyl]

Thank you 🙏

[@jeremiahgrove9201]

I was waiting on your next upload, thank you for another great video

[@leonvanzyl]

Thanks 🙏

[@shortthrow434]

Thank you Leon. Much appreciated. I'm working on this to create fun AI tutors for kids to help them learn. Was doing this the hard way (without flowise) which is doable but challenging. Your method really opens up the space to focus more on the content and tweaking versus the setup. Imagine One Piece Luffy or Castlevania's Belmont, Morgan Freeman, and David Bowie teaching you Calculus or Physics or certification materials. The sky is the limit. This is very useful and thanks again.

[@leonvanzyl]

Dude, those are so e hilarious ideas! 🤣 Glad I could help!

[@Jamie_57]

Your story is truly inspiring and I can't wait to see what topics you'll cover next. Subscribed!

[@leonvanzyl]

Thank you very much 🙏

[@senientmedia]

Seker een van die beste AI kanale wat ek nog tee gekom het. Jy verdien baie meer views. Dankie, jou videos help baie!

[@leonvanzyl]

Baie dankie!

[@micbab-vg2mu]

Great tips - thank you.

[@leonvanzyl]

You're welcome 🤗

[@marinodelacruz7216]

This one opens a lot of possibilities! Imagine, now we can clone anybody just downloading books and uploading them to the system, examples that comes to mind: Albert Einstein, Charles Darwin, Isaac Newton, Confucius, Aristotle, Plato, Sigmund Freud, etc... just need to find their books, put them on the system and boom! you have your personal advisor at hand 24/7. Heck, we can just create new ones just by uploading manuals and books. Example, you can have your own Splunk Administrator at hand just uploading the books available for it and then just name it Splunk Administrator... imagine this for any technology and / or programming language. This is Huge!!!! Best part is, NO HALLUCINATIONS! This is amazing! Thanks so much for showing us the door to heaven!

[@leonvanzyl]

Exactly!! You now have instant access to the experts and you can convert with them in a natural way.

[@alvaroclaure3552]

Thank you, Leon, for your generosity in sharing so much valuable knowledge. The first videos I watched were the ones using OpenAI Assistant. After that, I saw your whole Flowsise series and videos like this one that are about a personal assistant. My question is: if I use the OpenAI Assistant, which supports Retrieval, Code Interpreter and Functions, then I no longer need to implement what is shown in this video and others. Right? For me, it is important to understand this question and I would appreciate your answer. Thanks!!

[@leonvanzyl]

Hey there! You are correct - OpenAI Assistant includes all of those features natively, we simplifies the Flowise flow. One thing to take into account though, is that you have vendor lock in. In those other videos you have full control over the vector database, memory store, etc. and can migrate at any time. With OpenAI Assistant, you are locked into the OpenAI ecosystem.

[@nyacumo]

Leon thanks for the excellent video and effort. Can i ask you to explain the procedure to use faiss as a local vector? i would like to ingest files from a folder so i have question, does the embedding ocurr everytime the bot starts up? if so does it makes the embedding fro al the content on the folder or the previous embeddings dont a mean it only makes the embedding on the new files? as you see got a lot of questions on regard local vector, i use embeddings from openai. thanks in advance.

[@leonvanzyl]

Thank you 🙏. Have a look at my video on Vector Injesting. Flowise will load the data AGAIN after a server restart, resulting in duplicates in the vector store. In a production version of this bot I would definitely split the data load from the main chatbot.

[@copaxchannel]

How do you connect a chatbot like this to whatsapp?

[@Muhammad-Hanzala]

Hey Leon! I created a chatflow in flowise (used render free) and linked it with botpress. Everything was working fine. But the day after the whole flow was deleted. Can you tell me that how did it happened? Will render delete it after a dedicated time? Please help

[@leonvanzyl]

Hey there. Render's free account does not provide persistent storage, which means your flows will be lost after some time. I have a video on Deploying to Render that you might want to check it. It explains this and shows you how to attach a disk to your Render account.

[@KHYYUL]

Hey leon, you're a G. I was curious though, instead of using a conversational chain, I decided a conversational agent would be better. What are some example inputs for the Human Message parameter? And what's the difference between the system & human message? Can't seem to find any examples anywhere

[@leonvanzyl]

Thanks 🙏. The system message is used to prime the bot, so you can enter something like "you are an AI assistant, ...". The human message represents the user's side of the conversation.

[@lanc70]

Hey, thanks for the content ! What is the difference with using vector DB like Qdrant or Pinecone ? It seems way easier like that

[@leonvanzyl]

You're welcome 🤗. I used a local vector store to keep the tutorial short and to the point, but if you want to learn how to use a 3rd party service like Pinecone then check out the Document Chatbot tutorial on my channel 👍

[@practical-skills-school]

I desperately need a FAQ session on Flowise to clarify such questions: 1. What if the answer needs many chunks (or all of them) to answer the question like give me summary, or who are 10 main characters of the story? When I asked how many references has a scientific paper, it gave wrong answer because it didn't understand that several chunks contained the data. 2. How to update/add new knowledge? As far as I know, you can only add the WHOLE new stuff to the existing index in Vector stores. So you have first to clear the whole Vector store index, using API somewhere else. And then run with updated data the whole process 3. Is it possible for AI to understand that it is working with a paper, and the paper has the name and authors? When I asked about this data, it got stuck. 4. Can I prompt to provide links which are scattered in the text? Like "see ... for more" 5. Do not answer if there's no information on that in the knowledge base. You've done it here, thank you.

[@leonvanzyl]

Hey there. These are too complex to answer in a comment. We can arrange for a meeting so I can assist you. Go to cognaitiv.ai and complete the contact form. It goes through to me.

[@practical-skills-school]

@@leonvanzyl Thank you for the reply. Is there any chance that some of these topics will be covered in your videos later?

[@leonvanzyl]

@@practical-skills-school some of those have been covered in my series, and you'll have to get creative with some of those topics. I probably won't create videos talking about those specific points, but they'll be covered indirectly.

[@fyctuyfv]

@@leonvanzyl I'm interested in the links question too, could you answer that?

[@KenHammondOnline]

Thanks for this! I went through the entire process, but getting this error message in the chat: The maximum length of context that the model can handle is 4096 tokens. Any ideas?

[@leonvanzyl]

Try changing the Chuck size to a smaller size.

[@BadrLaajali]

Hi, I've been wondering about the optimal approach to utilize the RAG method with Flowise. While Langchain provides us the ability to manipulate vector storage and context, it appears that Flowise may not offer the appropriate tools for achieving this. Any insights on how to make it work effectively? Next video ? :)

[@leonvanzyl]

Interesting point this. I need to give this some thought. Can you maybe elaborate on which features Langchain offers to achieve this? Flowise does is make possible to update the vector store and change the context via system prompts.. like Langchain. It also offers integration with LangSmith and debugging in order to evaluate and improve the responses from the LLM. Really want to hear your thoughts on this.. I could take these gaps to the Flowise Dev team 👍

[@BadrLaajali]

@@leonvanzyl Thanks for your feedback :) To clarify my perspective, I'll break down the process into two main steps: First Step: Creating the Vector Database: I use the Pinecone library which allows me to handle vectors optimally. For instance, Pinecone's upsert function combined with zip is extremely helpful in inserting or updating vectors. It allows me to associate multiple attributes in a single operation. I also have the capability, through other libraries, to transform and utilize datasets as data structures, like DataFrames. To my knowledge, Flowise doesn't seem to offer such flexibility with datasets. Second Step: Utilizing the Database: In Flowise, I've noticed the "VectorStore To Document" function. If I understand correctly, this could be the equivalent of the similarity_search in Langchain. This function is crucial in the RAG method for extracting relevant results based on similarity. The "Prompt Template" feature in Flowise is also an essential step to incorporate the conversation history and enrich the context. Regarding LangSmith, even though I haven't had the chance to access its beta version yet, I am curious to see how it can integrate and enhance the interaction with Flowise in this process step. I hope this clarification gives you a better idea of my approach. I would also like to see how, with your knowledge, you could create an optimized RAG approach (combining datasets and external sources) on Flowise.

[@Felipeaaf11]

Leon thanks so much for the video! Is it possible to to store the conversation of the bot to a CRM? If so, what API tool would I use it? and also, is it possible to use this bot on Whatsapp?

[@leonvanzyl]

You're welcome 🤗. I'm not aware of any way to store the conversation to a CRM system, but if use a memory database like Redis then the conversation will be stored in a database. Yes, it's possible to integrate the bot into WhatsApp via webhooks.

[@suyogkhanal5542]

Hey, great one ❤ If I were to start a chatbot delivering agency, how would I make that out with flowise? Can I do it with flowise on scale?

[@leonvanzyl]

Thank.you.🙏. You can definitely build chatbots for clients using clients. Guess it depends what the client needs.

[@noblerboy2004]

Thank you for your video. I tried this. However, sometimes, AI will answer questions not in the file (could be his own trained data). is there any way to controll that?

[@leonvanzyl]

Yes, you can control that with the system prompt. Tell the bot not to make up answers.

[@noblerboy2004]

I already set system prompt. But the bot still answer question outside the provided documents :) @@leonvanzyl

[@practical-skills-school]

Thank you so much. I'm collecting useful prompts, so pity the prompt is not given as text here. Will OCR the screenshtts )

[@leonvanzyl]

You're welcome 🤗

[@hassanullah1997]

Hey Leon, great content as always. I dropped a message your AI agency, just checking that you got it?

[@leonvanzyl]

Thanks 🙏. Working my way through the emails 😊

[@leonamnagel9999]

Can I connect flowise to a Google Drive, or Google Docs?

[@leonvanzyl]

Yes! Check out my video on custom tools in Flowise where we get Flowise to write entries to Google Sheets 👍 kzbin.info/www/bejne/lXycqX2nesyhqLcsi=D5gLin9JufzMK3N2

[@nachomartinezescoda]

Hey, I just want to create a workflow on Flowise like a librarian with my ebook files. Do you know if it supports .pdf .mobi or .epub files? It interacts will all files at the same time with chunks and vector memory? Thanks in advance!! Very helpful!

[@leonvanzyl]

That is a great use of Flowise 👍. There are many different types of doc loaders, PDF being one, and these are updated all the time.

[@davemaguagroup5198]

Howzit Leon, I am new to Ai . Do you have a video going through this process on a Mac ? Regards Dave

[@leonvanzyl]

Hey there. The setup process is identical. There might be small differences in terms of opening the terminal, but that should be simple enough

[@OfficiallyArtificiallyInformed]

Lol it's the same man... what did you seem in the video that led you to believe otherwise? This is why I don't like apple. Removes any critical thinking

[@nicolasgramnea1324]

Can the chatbot be embedded on a website?

[@leonvanzyl]

Yes! Check out my video on embedding Flowise bots on websites 👍 kzbin.info/www/bejne/joDIdIlnrd6Bapo

[@siddhantgupta1300]

Hey leon just one doubt why didn't you used pine cone for the memory

[@leonvanzyl]

I didn't want to make the video longer than it needs to be, and I already have tutorials on including Pinecone in Flowise flows 😁. You can definitely use Pinecone if you want 👌

[@siddhantgupta1300]

@@leonvanzyl First of all thanks leon for replying, The reason I asked that question was because when ever I am running my flowise setup with pinecone im getting this error "ending Node not found" and whenever Im running it with simple in memory vector store the bot is running smoothly. Please help 😭

[@user-et6ku9bz9t]

Thank you very much for you video! I am really waiting for the video how to make a telegram personal assistant bot connected with Flowise. Is that possible?

[@erickcampos50]

It's a gratidão ideal. I hoje Leon find time to make this video

[@leonvanzyl]

You're welcome 🤗. I'll see what I can do.

[@eslammohamed-rx5jp]

GetHup dont work with me can you give me link flowiseAi in gethuo?

[@leonvanzyl]

github.com/FlowiseAI/Flowise

[@guersomtucumani]

Great, and how can I offer all that as a service?

[@leonvanzyl]

If you'd like your own persona bot, I can build it for you @ $200 and integrate it into Telegram. That's how 😊.

[@italy973]

Can anyone suggest a free alternative to OpenAi and a free alternative to Pinecone?

[@leonvanzyl]

You could use an opensource model (I have a tutorial on it). You could also use a local vector DB like Chroma or Local Vector Store. Cheers.

[@user-yr2nf9cr4v]

Wonderful Leon. The only thing missing would be to add a text-to-speech converter to complete the ai persona into a full fledged AI Avatar🤔

[@leonvanzyl]

Sounds like a great idea! I have videos on building web apps that use text to speech to play audio when chatting to the bot. I also have a video on training the TTS model on any voice. Combine those with Flowise and you'll have a fully functional Persona Avatar with voice.

[@user-yr2nf9cr4v]

@@leonvanzyl Looking forward to a new tutorial from you on this.....Cheers

[@themax2go]

i got my hopes up that it'll be 2 or more chatbotS talking to each other... can you make a "silly tavern" type of "group chatbots" version?

[@leonvanzyl]

Sounds like a fun challenge 😄

[@themax2go]

@@leonvanzyl ikr... but can it be done with flowwise?

[@claudi7009]

Great video. But in my scenario doesn´t work fine. I have flowise running in docker and I am using the component "Folder with files" and Pinecode as VectorStore when I save the proyect and run te Upsert Vector Database I get the following error: 2024-02-15 09:41:52 [ERROR]: ENOENT: no such file or directory, scandir '/home/user/Escritorio/Datos/Files/' Error: ENOENT: no such file or directory, scandir '/home/user/Escritorio/Datos/Files/' 2024-02-15 09:41:52 [ERROR]: [server]: Error: Error: ENOENT: no such file or directory, scandir '/home/user/Escritorio/Datos/Files/' Error: Error: ENOENT: no such file or directory, scandir '/home/user/Escritorio/Datos/Files/' at buildLangchain (/usr/local/lib/node_modules/flowise/dist/utils/index.js:331:19) at async App.upsertVector (/usr/local/lib/node_modules/flowise/dist/index.js:1345:13) at async /usr/local/lib/node_modules/flowise/dist/index.js:1017:13 I have tried with /home/user/Escritorio/Datos/Files/ and /home/user/Escritorio/Datos/Files but nothing. Do you have any idea? Thanks in advance.

[@leonvanzyl]

Try updating Flowise, and remember to save your chatflow before running upsert. I will be releasing the updated RAG video in the next day or two, so maybe that will assist you as well.

[@claudi7009]

@@leonvanzylI have already done and working with the latest version but it still fails. Thanks for your answer and your videos.

[@enriquez.3332]

Same problem here.

[@Agilesensei]

I had the same problem. WIth docker, you need to copy the directory into the docker volume that the docker instance uses for read/write. It's were sqlite is installed. In my case it was /root/.flowise, so I created a directory there. (/root/.flowise/mrbeast). Works perfectly.

[@claudi7009]

@@Agilesensei Not in my case. I deployed flowise on another instance (not default) and when I create the folder on that new volume I have the same issue.