Great video Oskar - very helpful and your suggestion of using Zep is very smart - a great open source Vector storage that can be self-hosted and thus save in fees over time.
Hi, thanks a lot for watching! I'm happy you found the suggestion helpful. Zep is indeed a fantastic open-source solution for vector storage - it has a great potential for saving on fees in the long run. Cheers for your support!
Congratulations! Very good video, produce more content about n8n it has helped a lot! Could you make a video on how to use the "Ollama Chat Model". Sorry for the English translated from Google, I don't speak English :( I'm from Brazil, State of Rondônia (Amazonia).
Hi, thanks for your kind comment! I haven’t played much with Ollama yet and honestly I don’t have enough experience in that field to record a tutorial. But I will give it a try soon and hopefully create a video in the near future. Sending best wishes to Brazil!
Thanks for an amazing and informative video! I noticed that it seems there have been changes to Zep Community Edition, and features like the Web UI and some others are no longer available in that version. Do you have any tips for alternative solutions for a local vector database that could work in a similar way? Thanks in advance, and keep up the fantastic work!
Great video, I was able to follow along and impliment. I have a couple of questions if you would be so kind. How do you add metadata to vector collections and conversations and 2. In conversation bot why does the memory not persist across sessions even if the I'd is the same? Thanks for the awesome videos 👌
Hello, thank you for your comment! Certainly - in this node I pinned URLs of some blog posts from n8n blog. I extracted them from sitemap of n8n blog with a separate workflow (not included in the video, since scraping itself is a bit different topic). I received sitemap from the other workflow as an array, so I had to split out the list into separate items (URLs). The node in this video is an old version of „Split Out” node (right now it’s a completely separate node with purple icon - docs below). docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-base.splitout/ Long story short: I simply used this node to pin mock data (list of URLs) that I got from the separate workflow. Feel free to let me know if you have any further questions!
@@workfloows Thank you for your response, earlier I tried using XML and split out as well, and I changed the chat using a Telegram bot from your other tutorial, and it worked well. Thank you, I hope there will be more great videos from you.
Hello Oskar, excellent video. Very clear ❤. I created these workflows by following the instructions which made it super easy. One limitation I seem to have is the zep memory for sessions. For some reason even with the same session ID I'm only able to retrieve several lasts conversations maybe like 6 or so but it's unable to retrieve earlier messages. Any idea why? Is there a workaround you can think of. Many thanks for your contributions ❤
Hello @worfloows, after watching your videos, and built everything. 1) I want to know the best combinaison for production Q&A chatbot with memory (in your opinion) ? :-) Zep/Supabase for VectorDb + flowise for interface + redis for memory ? You didn't use Redis in this video for memory but you said in the past that it's faster. 2) Is it possible to mix Redis for memory and Zep for VectorDb. Thank you for all your videos, it's a great job. Thanks in advance for your answer.
Hey! Thanks for your comment and sorry for my delayed response. To answer the first question: actually, I don’t have any specific favourite combination of tools in this field. For vector DB I use recently more often Supabase, but it’s only my personal preference 😃 All depends on the use case here. I didn’t use Redis in this video because I wanted to focus purely on Zep this time - but I’m glad you raised that point! About the second question: yes, you can use Zep as vector store and Redis as a memory. In such a case, when you use AI agent in n8n, you need to simply connect Redis as memory, and create a separate workflow to retrieve data from Zep vector store (connect is as a tool). I think such a combination should do the work. Once again, thanks a lot for your comment and for watching my videos!
Can you be so kind and provide the json file with your flows? Im having an error in the question and answer chain node and i do it exactly as you do fx {{$json.query}} and it returns error popup ->Parameter Text is required!!
Thanks for these n8n tutorials, especially on the AI side, great content. I have been looking for a self-hosted Vector DB, and will give Zep a try. Have you used Zep much, how are you finding it? My use-case is for storing technical data for use in customer facing Q&A Chatbot. Any other self-hosted Vector DB's to suggest?
Hello, thank you for your comment and kind feedback on my work! I use Zep mostly for my non-business projects (for crucial use cases I prefer managed cloud solutions). Overall, Zep made a good progress in the past months and working with it became quite easy. From other vector DBs I can recommend Supabase (but I use it only in cloud version). Library of n8n AI includes nodes that support working with Supabase, so configuration is pretty straightforward. Thanks a lot once again for your support!
That cool , let say you have whatapp chatbot so every user have unique number and unique chat … can use phone number as index in zep to retrieve previous chat ?
Hello @oskar, great video. Thank you With the memory session, can the bot save the all questions to improve itself? Or we don't need to improve a Q&A retrieval bot?
Hey, thank you for your kind comment and apologize for my late response. When you use Zep and memory session, bot automatically stores the history of conversation for the specific user (session). If you want to add some data based on conversation to the long-term memory (e.g. so other users can have access to it), it requires a separate flow to insert such data to database.
Interesting to get videos how to use n8n in web environment (wordpress, Webflow, etc) Ps. What do you use for such cool video tutorials for this mouse movement, smart focus, etc?
Hi, thanks for you your comment and kind suggestion - I will add it to my list and try to include this topic in future videos. App for screen recording is Screen Studio. For editing, I use Davinci Resolve.
@@workfloows thank you, I didn't know this software. I hope you will get more views. Can I ask, why only n8n topics? Why not to show other AI and automation related topics which have much more audience?
Thank you for your question. I’m aware that there are tools much more popular than n8n, but I personally believe that in terms of flexibility, n8n is one of the top low-code automation apps available in the market (and it's mostly free). I will most likely introduce other tools to my channel in the near future, but since I work with n8n almost every day (and it’s kind of my field of expertise), I want to establish a solid knowledge base for my viewers in this topic first.
Hi, thanks a lot for your comment and sorry for my delayed feedback. Great suggestion! I have already similar idea on my list and hope to create the tutorial about it in the near future. Thank you for watching my video!
Hey, thank you for your comment! Since Zep is a vector database, we need to generate embeddings for the content that we insert into it. I used embeddings model hosted on HuggingFace in this case, because it’s free to use. But you can use OpenAI or any other embeddings model of your choice.
A question I have is could you use this configuration with embeddings, zep and langchain with a private llm where you can work with private company data that can’t be exposed to public llamas?
Hey, thank you for your comment. Sure - as long as n8n, Zep and language model (connected e.g. through Ollama) are hosted on your own servers, you can use them in totally private environment.
Hi, thanks for your comment! I think the best way would be to use external API through HTTP Request node. There are some free available (e.g. GOQR) and many paid that include more features and customization options. Feel free to let me know in case you have further questions.
Hello, thank you for your comment! As far as I know, Zep requires at least 2 GB memory and supports OpenAI and Anthropic LLM services for summarization, intent analysis tasks etc. (Google Palm API so far unfortunately not supported). Below please find link to Zep docs chapter about LLMs configuration. docs.getzep.com/deployment/llm_config/
Great video Oskar - very helpful and your suggestion of using Zep is very smart - a great open source Vector storage that can be self-hosted and thus save in fees over time.
Hi, thanks a lot for watching! I'm happy you found the suggestion helpful. Zep is indeed a fantastic open-source solution for vector storage - it has a great potential for saving on fees in the long run. Cheers for your support!
Omg. You did a great job to describe everything in this automation flow. Big thanks
Excellent tutorial, Oskar. Thank you!
You’re very welcome and thank you for watching!
Amazing tutorial to understand logic behind.
Thank you very much!
It's a great presentation. The video is very comfortable！
Thank you very much for your comment and positive feedback!
😮 great Oskar, I can't stop watching your videos... what do you use to record the screen with that zoom and cursor?
Congratulations! Very good video, produce more content about n8n it has helped a lot! Could you make a video on how to use the "Ollama Chat Model".
Sorry for the English translated from Google, I don't speak English :( I'm from Brazil, State of Rondônia (Amazonia).
Hi, thanks for your kind comment!
I haven’t played much with Ollama yet and honestly I don’t have enough experience in that field to record a tutorial. But I will give it a try soon and hopefully create a video in the near future.
Sending best wishes to Brazil!
Thanks for an amazing and informative video! I noticed that it seems there have been changes to Zep Community Edition, and features like the Web UI and some others are no longer available in that version. Do you have any tips for alternative solutions for a local vector database that could work in a similar way? Thanks in advance, and keep up the fantastic work!
Hey, thank you for your kind comment! You can try Qdrant - I think it may be a good self-hosted alternative. Cheers!
Hi Oskar, thanks for another great video. could you explain why you decided to use this particular embedding system?
Great video, I was able to follow along and impliment. I have a couple of questions if you would be so kind. How do you add metadata to vector collections and conversations and 2. In conversation bot why does the memory not persist across sessions even if the I'd is the same? Thanks for the awesome videos 👌
Hi Oskar, very nice video thank you for sharing, I'm a bit confused about step 8, can you explain more about the node you're using for listing url?
Hello, thank you for your comment!
Certainly - in this node I pinned URLs of some blog posts from n8n blog. I extracted them from sitemap of n8n blog with a separate workflow (not included in the video, since scraping itself is a bit different topic).
I received sitemap from the other workflow as an array, so I had to split out the list into separate items (URLs). The node in this video is an old version of „Split Out” node (right now it’s a completely separate node with purple icon - docs below).
docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-base.splitout/
Long story short: I simply used this node to pin mock data (list of URLs) that I got from the separate workflow.
Feel free to let me know if you have any further questions!
@@workfloows Thank you for your response, earlier I tried using XML and split out as well, and I changed the chat using a Telegram bot from your other tutorial, and it worked well. Thank you, I hope there will be more great videos from you.
Hello Oskar, excellent video. Very clear ❤. I created these workflows by following the instructions which made it super easy. One limitation I seem to have is the zep memory for sessions. For some reason even with the same session ID I'm only able to retrieve several lasts conversations maybe like 6 or so but it's unable to retrieve earlier messages. Any idea why? Is there a workaround you can think of. Many thanks for your contributions ❤
Awesome video! Keep it up.
Thank you very much! 🙌
Hello @worfloows, after watching your videos, and built everything.
1) I want to know the best combinaison for production Q&A chatbot with memory (in your opinion) ? :-)
Zep/Supabase for VectorDb + flowise for interface + redis for memory ?
You didn't use Redis in this video for memory but you said in the past that it's faster.
2) Is it possible to mix Redis for memory and Zep for VectorDb.
Thank you for all your videos, it's a great job.
Thanks in advance for your answer.
Hey! Thanks for your comment and sorry for my delayed response.
To answer the first question: actually, I don’t have any specific favourite combination of tools in this field. For vector DB I use recently more often Supabase, but it’s only my personal preference 😃 All depends on the use case here. I didn’t use Redis in this video because I wanted to focus purely on Zep this time - but I’m glad you raised that point!
About the second question: yes, you can use Zep as vector store and Redis as a memory. In such a case, when you use AI agent in n8n, you need to simply connect Redis as memory, and create a separate workflow to retrieve data from Zep vector store (connect is as a tool). I think such a combination should do the work.
Once again, thanks a lot for your comment and for watching my videos!
Thank you for the answer
Great video! New sub.
Thank you very much!
more n8n!👍👍
Can you be so kind and provide the json file with your flows? Im having an error in the question and answer chain node and i do it exactly as you do fx {{$json.query}} and it returns error popup ->Parameter Text is required!!
Thanks for these n8n tutorials, especially on the AI side, great content. I have been looking for a self-hosted Vector DB, and will give Zep a try. Have you used Zep much, how are you finding it? My use-case is for storing technical data for use in customer facing Q&A Chatbot. Any other self-hosted Vector DB's to suggest?
Hello, thank you for your comment and kind feedback on my work!
I use Zep mostly for my non-business projects (for crucial use cases I prefer managed cloud solutions). Overall, Zep made a good progress in the past months and working with it became quite easy. From other vector DBs I can recommend Supabase (but I use it only in cloud version). Library of n8n AI includes nodes that support working with Supabase, so configuration is pretty straightforward.
Thanks a lot once again for your support!
That cool , let say you have whatapp chatbot so every user have unique number and unique chat … can use phone number as index in zep to retrieve previous chat ?
Hello, thank you very much for your comment.
Yes, you can use phone number as session ID, which will identify specific chat history in Zep database.
Hello @oskar, great video. Thank you
With the memory session, can the bot save the all questions to improve itself? Or we don't need to improve a Q&A retrieval bot?
Hey, thank you for your kind comment and apologize for my late response.
When you use Zep and memory session, bot automatically stores the history of conversation for the specific user (session). If you want to add some data based on conversation to the long-term memory (e.g. so other users can have access to it), it requires a separate flow to insert such data to database.
@@workfloows. Ok it's clear;-)
Interesting to get videos how to use n8n in web environment (wordpress, Webflow, etc)
Ps. What do you use for such cool video tutorials for this mouse movement, smart focus, etc?
Hi, thanks for you your comment and kind suggestion - I will add it to my list and try to include this topic in future videos.
App for screen recording is Screen Studio. For editing, I use Davinci Resolve.
@@workfloows thank you, I didn't know this software. I hope you will get more views. Can I ask, why only n8n topics? Why not to show other AI and automation related topics which have much more audience?
Thank you for your question.
I’m aware that there are tools much more popular than n8n, but I personally believe that in terms of flexibility, n8n is one of the top low-code automation apps available in the market (and it's mostly free).
I will most likely introduce other tools to my channel in the near future, but since I work with n8n almost every day (and it’s kind of my field of expertise), I want to establish a solid knowledge base for my viewers in this topic first.
Hello, it will be awesome if you could provide with some tutorial to build a appointments chatbot
Hi, thanks a lot for your comment and sorry for my delayed feedback.
Great suggestion! I have already similar idea on my list and hope to create the tutorial about it in the near future.
Thank you for watching my video!
Great video Oskar! I have one question. What's the purpose of hugging face and why use it if Zep is a datastore? Thanks.
Hey, thank you for your comment!
Since Zep is a vector database, we need to generate embeddings for the content that we insert into it. I used embeddings model hosted on HuggingFace in this case, because it’s free to use. But you can use OpenAI or any other embeddings model of your choice.
Gotcha, thanks.
A question I have is could you use this configuration with embeddings, zep and langchain with a private llm where you can work with private company data that can’t be exposed to public llamas?
Hey, thank you for your comment.
Sure - as long as n8n, Zep and language model (connected e.g. through Ollama) are hosted on your own servers, you can use them in totally private environment.
What's the best way to make qrcode in n8n?😊
Hi, thanks for your comment!
I think the best way would be to use external API through HTTP Request node. There are some free available (e.g. GOQR) and many paid that include more features and customization options.
Feel free to let me know in case you have further questions.
hey how did u integrate n8n with zep?
What is minimum ram requirements for Zep AI?
Could i use other llm chat api other than openai, such as google palm api?
Hello, thank you for your comment!
As far as I know, Zep requires at least 2 GB memory and supports OpenAI and Anthropic LLM services for summarization, intent analysis tasks etc. (Google Palm API so far unfortunately not supported). Below please find link to Zep docs chapter about LLMs configuration.
docs.getzep.com/deployment/llm_config/
@@workfloows Oic, thanks a lot for the video and your help