Build a No-Code Chat-with-PDF LangChain app using Flowise and Bubble

[@tursunable]

This is what I was looking for! Thank you for the amazing tutorials that fill in the gaps on Bubble!

[@home1250]

Same here guy

[@sitedev]

Wow. I've been stumbling around with LangChain using Notebooks and building simple things in Bubble while trying to figure out how to pull everything together. Finally, you've created something that joins the dots. Please do continue creating more videos like this one and I also would like to express interest in any courses you produce that dive deeper in to this space as it continues to evolve (assuming we can all keep up with the ridiculous pace ).

[@menloparklab]

Awesome! Thanks.

[@malekaix]

Same here! Thanks a ton for this video@@menloparklab !

[@jalengonel]

I cannot tell you how appreciative I am of you for making this, you were incredibly helpful for this. Cheers to you

[@menloparklab]

🙏

[@joshuadelgado4460]

THANK YOU SO MUCH! I WAS LOOKING FOR SOMETHING LIKE THIS WAS VERY USEFUL

[@Dylan-c5v]

I've tried following this several times and it has not worked. At first, I was stumped until i read about the Flowise API updates and started using the /vector/upsert API instead of the /prediction API. But I still can't attach a .pdf or .txt file without getting error messages. Enjoyed watching your video and getting excited about the possibilties. But completely bummed to waste so much time on a process that failed over and over.

[@nocodenexus]

You are the best man! This is a very detailed video. I know nothing about Langchain, but you made me fall in love. Thank you so much. You made everything so easy

[@olegxander]

Thank you for sharring all your great knowledge! It would be great if in the future you could dive deeper into combining multiple chains and tool chains

[@scottvickrey2743]

Yes, seconded!

[@dhayas8027]

Awesome bro!!! Thanks for detailed video.. awaiting for the continuation of this video in the next part

[@Therealishmatt]

This was absolutely fire! Thanks a million

[@tusharmudgal6345]

which architecture diagram tool you use to make flow diagrams?

[@chandrakalagowda3129]

thanks for giving the link to the bubble app. For beginners it is really helpful. Thank you for this video as well.

[@omountassir]

Yess! The course would be great to have access to! Nice content by the way!

[@menloparklab]

Thanks a lot ;)

[@JuanBernard-z3r]

Hey man this is great, but just had one question can we use the bubble database instead of the Pinecode database in this case we will have more control on the input since it is in the database

[@BT-xm7ti]

To retrieve or do the Search Query, which Pinecone module should you add instead of the Pinecone Load existing index that doesnt exist anymore. Do I use the same workflow for the 2nd part, of Search Query/ Retrieve information from the PDF?

[@Masterhunter325]

Every your video is gold. Please keep up!

[@CoreAspect]

This is great stuff yet again... looking forward to the next instalment.

[@Francotujk]

HI! How can it be done with the current version of Flowise (they have made one node for the vector store). It's quite complex to upsert and then load the document

[@Karnadagan]

Hey, whenever there is question sent to flowise, it is considering every question as new. how to send history also via the same api call

[@TristanvanDoorn]

In the demo it seems that you can only upload one single PDF or file. Would it be possible to upload - let's say - 20 PDF files and still have a concise answer from the bot?

[@RafayRaheel]

Thank you for the amazing tutorial. Just wanted to ask you about one issue im facing. My api returns response with : {"text":" "} and this text header gets stored in the database too and this is what gets gets displayed on the front end. Is there a way to make sure that api responses dont have {"text":" "}?

[@scottvickrey2743]

More Flowise videos, PLEASE!!! Do the autoGPT flow and maybe even mention for others how to avoid cors errors by editing the client.ts in flowise by adding 'Access-Control-Allow-Origin': '*' You already helped me a lot! Thank You!

[@JamesH-v3g]

excellent!! Your sample app is throwing an API error with flowise when u try and upload a file or use chat. do you happen to know why your sample app is causing an error?

[@edouardratiarson5084]

I'm getting the same thing.

[@menloparklab]

I have disabled the sample app, as it had link to my flowise app and some folks started changing flows in it. :(

[@Megamil]

Hello, I am currently using the FlowiseAI APIs for my chatbots. As of now, I am using a single chatflow for all my clients, identified by the ID: 52...a5. However, I am interested in creating or cloning chatflows dynamically for each client via the API, instead of manually creating them through the graphical interface. Could anyone guide me on how to achieve this? Any help would be greatly appreciated. Thank you.

[@RaphiNathan]

Hi, I’m at 9:19. I’m getting an error when testing with a pdf in the chat. “Error: OpenAI or Azure OpenAI API key not found”. I’m using gpt-3.5 and have tried different api keys. Maybe I’ve made a mistake in setting up the Pinecone index? Please help. Thanks

[@Wellbeing525]

Hi, Ive followed your instructions and when I try to load a PDF to test, It doesn't load automatically. I have to click on the upsert to Vector database button to the top right. I'm using the free version of Pinecone and Flowise. ANy ideas?

[@ariwidodo-video]

Thank you very much. I expect more videos about building apps.

[@asdsadsadsadasdads]

Hey! How does the file gets send through the POST api when it is a form type? Using a PDF url did not work for me, but if it is a local file path, it'd work. Is there any magic happening behind the scene?

[@Belmeck]

Awesome video! I'm building a knowledge based writing app and had a question. Is it possible to use my trained model as a data source to write and display copy instead of a chatbot interface?

[@menloparklab]

Yes definitely. You can use a text element that gets updates with the content you would like for it to display.

[@peralser]

Thanks for sharing. Clear and helpful!

[@sertrader514]

Amazing tutorial!, buen When asking something about the document and having this information divided into two pages, the program omits it. How could I fix this? Thanks!

[@menloparklab]

Please try increasing chunk size. With GPT-4 access, larger chunks shouldn’t be an issue.

[@raindrops20008]

Absolutely fabulous video! I finally made it work on Flowise! But I dont get any response when added as a widget. What might be the reason?

[@menloparklab]

Thanks! Wondering if there is a connection issue to Flowise. From outside of the Flowise application?

[@menloparklab]

Please check the shared template. It has API connector available with all details.

[@tonyhouston261]

Thanks for the comprehensive demo. One of the document loader types in Flowise is 'Folder with files', however it is not clear how to establish the document folder path. Using Render, I tried to create a disk, upload files from my computer to disk, and then set the folder path in flowise to the Render disk file. None of this worked. Can you provide any guidance? Thanks.

[@menloparklab]

That's a good question. I haven't tried using folder with files before. But wondering if you tried uploading multiple pdf in the pdf block? Flowise gives option to select multiple files at once in the pdf block. Or if you are using other data type, you could select that block and multiple files of that block. Let me know if this works in your case.

[@88seulkim]

Hi i have a question, for the pdf upload chatflow, do i need to change any settings to the pdf file part? I’m asking because i feel like i did everything the same as tutorial but my files aren’t being uploaded to pinecone

[@88seulkim]

So just to dbl check, i’m using the upsert URL for pdf upload (the upload scenario 3) and then pdfquery (query API call filter), sorry i’m just trying to debug and would love your feedback

[@menloparklab]

No, you don't need to change anything other than the video steps. Hope it worked for you.

[@automatizapedidos6078]

Hi, if we use supabase for vector storage, isn't it possible to filter by the same document name in the pinecone where we can filter by namespace?

[@menloparklab]

Yes, that's also possible.

[@othon.rodrigues]

Incrdible! I was looking for this for a while. Thank you so much! I have a question: What about scaling this feature inside bubble? I will have to upgrade the Render plan and also Pinecone? There is another bottleneck that I need to worry about when scaling?

[@menloparklab]

I believe Flowise have added or will add multi-threading soon, that helps run multiple users ask questions at the same time. So that should solve the issue for multi user app. Render will work fine to an extent and then you might have to bump a level. If you anticipate a lot of documents by users, then using open source vector store instead of pinecone, can save some $$. I would suggest to test it with this setup, most likely it will handle a good bit of your use cases.

[@othon.rodrigues]

@@menloparklab Hummmm, good know that! Do you know where can follow these Flowise updates, like this multi-threading? I already follow them on Twitter, but I don't know if it is enough haha, thank you again!

[@anuragjha6825]

Bro you are a GOAT. ❤ For real, keep making videos like this and surely you’ll be the top creator in No/low core field, no doubt about it. Looking forward to meeting you someday and say thanks personally. ❤

[@JeroenErne]

really cool stuff 😉 very well explained

[@ShaneJayakrishna]

I think the Flowise documentation has been updated so that there's 3 different endpoints for different actions. POST /api/v1/vector/upsert/{your-chatflowid} for Upserting POST /api/v1/prediction/{your-chatflowid} for Predictions GET /api/v1/chatmessage/{your-chatflowid} for Messages Misbah, does using 1 endpoint for all of them still work for you?

[@finleymacdonald5026]

The original predication POST wasn't working for me even with the additional parameters. Switched to vector/upsert and it worked. Thanks

[@MrBenjo000]

Thanks for this great tutorial... for the "Conversational Retrieval QA Chain" why are you choosing text-davinci-003 and openai component instead of the standard chatopenai with gpt-3.5-turbo?

[@menloparklab]

Good question. So I chose LLM block as an example, but you're right, we could've used chat block with 3.5-turbo

[@MrBenjo000]

Thanks for getting back to me, yes that makes sense.@@menloparklab I noticed that all my chatflows disappeared 😞any suggestions for how to handle this kind of error? Is there a trigger or activity that may cause these to drop? I'm using the paid starter plan on render, so don't think it that. I also see I can export my chatflows as a form of backup and recover, but it's not really optimal if things start dropping off!

[@menloparklab]

I agree, there seems to be routine refresh of deployed instances on Render which is a bit annoying tbh. I might switch to Railway.app or something else in future vids. Thanks for informing.

[@simarpreetsingh6604]

Hi. When I embed the flowise script code inside my website header, I see the chat widget, but it says "Powered by Flowise". How can I remove that? Thanks very much.

[@menloparklab]

Hey there! Since it was built in flowise, most likely they have kept their watermark. You might wanna check with the flowise team if there is a way to remove it.

[@DS-ig1bf]

Great content. question! I tried it and it seems answers are restricted to the content of the doc and can’t go out of context. Is there any way around that so it can respond like the normal ChatGPT and also refer to the doc?

[@othon.rodrigues]

I really need this also! Be restricted to the content of the doc is not so good :( I need the answer to be a mix of doc and GPT's data. As would be a normal input.

[@menloparklab]

Yes keep the temperature setting above 0.5 and also change prompt to answer question even if not present in the documents

[@DS-ig1bf]

@@othon.rodrigues any luck since?

[@michaelortiz1444]

Great video! Could you do a video trying to use Replit for persistence?

[@menloparklab]

Will be doing some more replit vids soon

[@JuanMackie]

This is absoluty great, thank you. Are you able to provide advice on why flowise keeps deleting and losing my flows? I have setup a disk during setup to try and offset this issue,but still no luck?

[@menloparklab]

If you are using Render or Railway, they do tend to cause deletion. Suggestion would be to deploy on a platform with a persistent disk.

[@JuanMackie]

@@menloparklab I have been running Railway without any issues. Thank you again. Is there a way to have multi-file questioning?

[@eduardomoscatelli]

Great tutorial . I have already signed up for the course. Is there a possibility to release some classes before the 24th of July or will all classes be released after that date?

[@menloparklab]

Hey thanks a lot for signing up. Working on releasing a few before as well. Stay tuned.

[@diegosandoval7462]

I keep getting Error: Request failed with status code 429 , have everything exactly set up, could use the help from anybody willing to jump on a call

[@moreshk]

any issues with using vercel instead of render?

[@menloparklab]

The setup for vercel is a little different than render, but could be deployed on vercel as well. They have some examples one could follow to achieve the same.

[@topofferswarehouse]

i have found an issue, the created and saved chat flows on flowise will disappear sometimes later. How to resolve this

[@menloparklab]

Free instance on render makes them disappear. You might need to get the starter instance.

[@topofferswarehouse]

@@menloparklab i am using AWS EC2 server

[@onangarodney7746]

That was brilliant. Thank you for the great video. I have 1 question, what ai model was used to make the search into the document, was it an OpenAI's one? Is there an email where we could reach out to you for some contracts?

[@menloparklab]

Thanks! Yes openAI model was used to search the document. Feel free to reach out to [e-mail]. Would be happy to help.

[@ashminjayson3978]

Does anyone know the website he uses for making the diagrams?

[@menloparklab]

It’s Excalidraw

[@nicolasrny2593]

In your repeating group you have :"me: question" but you don't have : "AI: answer". How do you get the AI answer display if it's not in your repeating group ? Many thanks for your content !

[@menloparklab]

I am using a conditional to distinguish between AI and human messages. It’s the right most tab on the repeating group details pop up.

[@Masterhunter325]

What tool do you use for drawing workflows?

[@menloparklab]

I use Excalidraw

[@havendv]

@@menloparklab What component library are you using for Excalidraw? Or is it a paid version?

[@stoicguac9030]

What software are you using to make these diagrams? @5:56

[@menloparklab]

It’s a tool called Excalidraw

[@ChuckWilliamsTechnology]

Quick question...can I use ChromaDB instead of Pinecone vector DB?

[@menloparklab]

yes, definitely. You can swap the vectorstore blocks out.

[@automatizapedidos6078]

Hi! Can you activate the subtitle/closed captions of the video?

[@menloparklab]

Added external subtitles, as youtube ones were still processing.

[@raymondduong2016]

Is there any way that we can make it load/read multiple PDFs at the same time?

[@menloparklab]

From the Flowise UI, you can select multiple documents at once and upsert. From bubble as well, we could send multiple to Flowise. But it might require some changes to the API call configured in Bubble.

[@lokeshkumarpro]

@@menloparklab Can you please show us what changes we need to make to API call for multiple files?

[@Steve-js7bp]

I dont know if you could have made this any clearer but i am still getting an error on the hosted flowise page when i try to talk to my data. I have just used the standard markertplace QA retrieval but keep getting the following error when i try to use the in-built chatbot. TypeError: Cannot read properties of undefined (reading 'replaceAll') Has anyone encountered this issue? I am unsure if it is Flowise, render or any of the other components of the chain

[@menloparklab]

Actually I got that error a few times. I had to delete some of the blocks, re-add them, save and run. Hope that works in your case as well.

[@Steve-js7bp]

@@menloparklab thanks! definitely makes me feel a bit better. it was driving me crazy. flowise is looking great but seems a bit temperamental sometimes

[@jalengonel]

Thanks for responding. I had the same error!

[@10points56]

Quality content.

[@JT-Works]

6 months later and Flowise still does not have hosting option... huge missed opportunity on their part... Excellent video on your part!

[@Vexxter09]

do we need a paid OpenAI account for this ?

[@menloparklab]

you just need the APIs configured with openAI, it is charged on a per usage basis.

[@ko-Daegu]

How to save the chat history and context for later

[@menloparklab]

You might have to save chat responses and user queries in Bubble. Then send it with the user query to Flowise.

[@leveladsmarketing5889]

Could it accept multi-files ?

[@menloparklab]

yes it can

[@leveladsmarketing5889]

@@menloparklabWhat about using a customized model? A fine-tuned model

[@Steve-js7bp]

is there a way to edit the prompts for this?

[@menloparklab]

there is a prompt template in flowise that could be used, it might require for you to rearrange the flow though

[@Steve-js7bp]

@@menloparklab awesome, thankyou. great job by the way!

[@TF-qt3jh]

🎉Incredible content!!! By the way, how would you design a flow if I want to give GPT my feedback after every step. What I’m trying to develop is, like a research app. It does first step research and returns the result, then I want to give it my feedback, e.g. “Maybe you should expand your scope to industry B as well” or “The amount of your research at this stage is too little”. Then, it goes to the next step so that the final result through several steps could more likely to meet my requirement. ❤

[@menloparklab]

I would suggest using an agent for such a use case. They are more interactive and can take feedback.

[@menloparklab]

Please check my other video on agents. ;)

[@TF-qt3jh]

@@menloparklab kzbin.info/www/bejne/oHWtenyMjLmNeqc this one right?

[@scottvickrey2743]

This can easily be handled on the front end using basic flows on the backend until we get a full grasp on Flowise and our knowledge of it grows. looking forward to more flowise detailed videos.

[@SylvanSequester]

Can you make this HIPAA compliant for research?

[@menloparklab]

So for this particular demo, OpenAI and Pinecone were used, which basically means sending data to them. I will be uploading another video soon on how to make privacy-centric apps that could be used for HIPAA compliant research.

[@SylvanSequester]

@@menloparklab thank you!!!! Do you do paid consultations and/or commissions to create these systems if I get a research grant in the future?

[@menloparklab]

@@SylvanSequester yes, we can build the system or do consultations. ;)

[@fraternitas5117]

I can't get it to build on render.

[@menloparklab]

What error do you see? Wanna share on discord?

[@nishantkumar-lw6ce]

Please do a tutorial on bubble as it’s very difficult for a new person. Tried using it but still throws errors.

[@menloparklab]

Building a full course on it, to go in details at builtbyou.com/

[@LuqmanAsif-r5d]

How we build a ChatGPT application with multiple PDFs?

[@menloparklab]

You can call the same workflow multiple times with the PDFs

[@RogueIntellectuality]

Sir, please make a video on flowiseAI and Wix website functionality like playground, chatgpt , chat bot, pdf upload chat etc. It will surely help me in my start up. I have sent an email to you already Thanks

[@menloparklab]

Hi there! I haven’t used wix, so not very familiar with it. Do they offer an ability to call external APIs?

[@muazashraf409]

You already created flows and just describing them! Why. Please create scratch workflows for better understanding

[@chrisbitoy7272]

I got error trying to deploy via railway, same with Vercel. Only render seems to be the service that can deploy this app for some reason...so frustrating with the constraints . Error: Dockerfile:11 ------------------- 9 | # Copy root package.json and lockfile 10 | COPY package.json ./ 11 | >>> COPY yarn.lock ./ 12 | 13 | # Copy components package.json ------------------- ERROR: failed to solve: failed to compute cache key: failed to calculate checksum of ref moby::ihwhj: "/yarn.lock": not found Showing 61 logs

[@menloparklab]

It seems like you are using docker as the deployment method, that gives an error. Try using node based installation.

[@MashZ]

i tried exactly this but replacing openaiembeddings with cohere embeddings. Got 'this.retriever.getRelevantDocuments is not a function' error.

[@menloparklab]

I haven’t tried cohere embeddings with flowise, will have to test it myself

[@MashZ]

@@menloparklab You really should! I believe cohere embeddings has far more potential than openai as they're updating the pricing to tokens system to match openai and it supports 100+ languages

[@Slipknotbira]

Hello, is there any way to delete a specific PDF from Pinecone with Flowise? or delete the index/namespace completely? Thanks.