Build AI Chatbots (with RAG) for FREE using Langflow and Ollama (Run Models Locally)

[@user-uv3nv2bc6v]

Langflow becomes more and more professional. Please make some more Langflow videos. Thanks Leon! Great Job!

[@leonvanzyl]

Will do

[@JanielJ-i1x]

You are such a Good Teacher, Subscribed, Liked., Keep enlightening us Friend

[@leonvanzyl]

Thank you for the feedback and sub 🙏

[@bubelevakalisa7313]

Great stuff broer👌🏾

[@mt5]

@Leon Very good video, thank you 🙂

[@waynenunn7319]

I always enjoy your videos. They are straight to the point and do not contain sales pitches. One question: In the file input node, can you input multiple files of different or the same type, or must you have a node for each file inputted? Thanks for the great video.

[@leonvanzyl]

Thank you! The node accepts different file types. If you want to upload multiple files then you can either add multiple file nodes to the canvas or you can use the folder upload node.

[@art_Aj]

hello Leon, in your opinion which platform is net better flowise or langflow?

[@tecnopadre]

If the architecture is the same, and model too, wouldn't it be the same results for a RAG? Said that, wouldn't it be the best option the best that you understand better? As I'm seeing Leon on every video, I think VectorShift is pointing towards the best option. Then, because of the price, Flowise.

[@MohsenaAshraf]

I might be wrong, but in the 'Recursive Character Text Splitter', the 'data' output node does not get connected ti the 'Ingest data' of the 'Chroma DB', because of the newest version maybe. Do you have a suggestion on how to pass the data to Chroma db? The 'Split text' component works though.

[@rubbercable]

Can you use conversatinal (speech2speech features to langflow?)

[@Aliexpress71]

Por favor coloque a tela no modo dark para que possamos visualizar o que está fazendo.

[@mello1016]

What are your thoughts on flowise vs langflow? Which would you recommend using?

[@crippsverse]

What PC spec do you recommend to run this setup?

[@BadBite]

Leon! awesome content as always. Is Flowise capable of this too?

[@leonvanzyl]

Thank you. Yes, here's the FW video: kzbin.info/www/bejne/bmbKi2p9Yt-YacU

[@maniecronje]

Dankie Leon, looking fwd watching any of your new postings... PLEASE can you do some more examples on FW , Agentic flows any use case love learning tips and tricks you may have, dankie vir al jou moeite!

[@leonvanzyl]

Absolutely! I'm waiting for some exciting FW features to roll out.

[@JPBCDE]

Dankie Leon. Could you perhaps try and do a video on how to use SQL agents/Databases to retrieve data? Since the V1 some of the previous components/tools is missing and now matter how hard I try, i cant seem to get the SQL Agent to work. Dankie!

[@Torpila123]

Please shed some light on this because I'm going crazy! Spent 3 hours on this. The Ollama node initially didn't work. It gave this gem: "Error building Component Ollama: Ollama call failed with status code 404. Maybe your model is not found and you should pull the model with 'ollama pull llama2". When attempting to use it this way I was using llama3. Then I decided to follow the error's advice and installed llama 2. Now the node works and the A.I. responds in the langflow playground, but I still can't choose from the available models on the Ollama node, so it's useless because I have no way of specifying the embedding model. 14 minutes of tutorial turned into 3 hours of frustration with rage at the end.

[@leonvanzyl]

Did you follow the steps to pull the models first?

[@Torpila123]

@@leonvanzyl Hi Leon. Thank you for replying. Yes. I've pulled a llama3(7b) model and the identical embedding model that you mention in your video. I can talk to it within the Ollama terminal. All good on that end, however in Langflow, the Ollama node is still missing the dropdown for selecting the models. At this point if I attempt to run the node I would also get that error that is recommending to download llama2 -> did just that, still no dropdown. However this time when I run the node it works and I can interact with the LLM in the LF playground. Obviously the replies come from llama2 even if I also have llama3 available. Claude says that most likely the LF node is defaulting to the llama2 model as long as it's present. So it kind of works, but not like it's supposed to. After posting my comment I did discover that I have some errors in the LF terminal so from this point forward who knows? Maybe it's something related to my system. Also tried reinstalling both Ollama and Langflow and it behaves the same.

[@Torpila123]

@@leonvanzyl Managed to solve the issue by modifying the Ollama node code inside of Langflow. Changed this part: " DropdownInput( name="model_name", display_name="Model Name", value="llama2", info="", refresh_button=True, ) " To this: " DropdownInput( name="model_name", display_name="Model Name", value="llama2:latest", info="", refresh_button=True, options=[ "llama3.1:8b", "llama3:latest", "llama2:latest", "nomic-embed-text:latest", "llama3:8b" ] ), ...where I basically hardcoded the names of the models I have on my system.

[@shadhukumar985]

input text is not working , can you suggest?

[@HumanDefenseForce]

Can you show how to use Mem0 with agents by turning its functions into custom tools. To give our agent context memory, please

[@Techonaroll]

perfect timing (mrbeast alleged)

[@leonvanzyl]

😁

[@AllAboutTechnologies]

Hey Leon thank you so much for these videos. I followed along and made a bot but the context and history part is either not working correctly or its behaving weirdly. In one reply it will say 2$ for an item and in next it will say 3$ and it forgets about our conversation too. How can we improve the accuracy? For serious matter what would you prefer? Langflow or building from scratch with LangChain or something?

[@leonvanzyl]

What model are you using?

[@AllAboutTechnologies]

@@leonvanzyl OpenAi. I did change the temperature to 0 and now its performing better. However whats your take on using this for a commercial purpose vs something built from scratch with lang chain?

[@NS_Miata]

Can you show on how to use chatbots on sql databases. Specifically limit the chatbot on some sql commands?

[@paidoclock]

Badass man! Is there a way to make a voice agent for appointment booking or other business related tasks?

[@leonvanzyl]

For sure! Langflow provides AP endpoints (which I'll cover soon) which you can call from your applications. You could use STT in your application to convert the audio to text, call the Langflow API and the convert the response into audio using TTS.

[@乾淨核能]

what's harware requirment? thank you

[@sinanalmac2395]

Hello I thank you for your great video . I have an issue about ollama model . But when I press to refresh button on the right side of model name. I got the message below. Error while updating the Component invalid literal for int() with base 10: '' any suggestion ?

[@e-Course.]

Thanks for nice video , can you make Vectorshift video tutorial for creating chatbot with authentication please 🙏

[@ZeitCode]

why is it important to create a virtual environment for installation? The official channel says nothing about it. Thanks great tutorial.

[@leonvanzyl]

It's good practice in Python to set up an environment for each project, instead of just installing everything globally on your machine.

[@tecnopadre]

I'm missing every time a benchmark, so we know how many users can use this architecture at the same time with a let's say your computer architecture (could it be a RTX 3080,...). Is this something that could be launch into production or do we need Azure AI or similar? Render has been not enough for me for real projects and it's expensive. Oh! And thank you for your videos!

[@leonvanzyl]

Hey there. The point of the video is to run models locally, hence you only have one user at a time. This is not a cloud solution. For that you might want to consider using Flowise or Langflow cloud instead.

[@wEBMedPL]

What do you think is better for low code no code people like me - langflow or flowise?

[@leonvanzyl]

Both platforms are trying to compete in this space. In my opinion, Flowise is slightly more stable, but Langflow does seem to be catching up. You can access Langflow in the cloud (for free), which is a service that Flowise does not offer (yet).

[@wEBMedPL]

@@leonvanzyl I mean, no offence but everything AI related costs a bit (implementing multitude of AI models, like from google, Open AI and Anthropic) costs a little too. Hosting my instance of flowise on Render (as you recommended)+pinecone cost me

[@sehmusakcakale5340]

How can I fix the error "Failed to establish a new connection: [WinError 10061] No connection could be made because the target machine actively refused it" that I keep getting?

[@rayman2474]

Dankie Leon, I did set up everything, and I did pull my Ollama modules. And if I want to use them in langflow. Ollama doesn't pick it up for some reason. Do you know why? I Only says "No parameters available for display"

[@paganaye]

I had the same. Access the code of the Ollama component (first button at the top). change DropdownInput to StrInput and llama2 to llama3

[@marceloguimaraes796]

no show models for me, only work llama2

[@leonvanzyl]

Sounds like you didn't download the models? Be sure to follow the steps to.pull the models.

[@ffblanch]

Same issue here and I'm seeing my model under this API localhost:11434/v1/models

[@ffblanch]

It's working only with llama2 for me and I still doesn't see models list in langflow

[@marceloguimaraes796]

@@leonvanzyl bug

[@JPBCDE]

yeah. I had the same issue. Does not matter what you do, it keeps looking for llama2. What I ended up doing, was change the code for the Ollama component, to change the "dropdown", to be a text input (copied the code from the previous field). Then I could type in the model I wanted to use and that worked.

[@HyperUpscale]

So you can reuse the embeddings node, but you can't reuse or extend the ChromaDB node 🤨?

[@leonvanzyl]

I actually had an issue when reusuing the same Vector Store node. The messages were duplicated in the chat. I guess it technically works 😂