Great work Mervin! But there is an issue in this setup - While setting up vector database you are using postgres vector store. It should be Qdrant vector instead.
Awesome work! I wonder though, it seems perfect for prototyping and DIY purposes, but can it be scaled to handle multiple users? I've not heard about n8n before, so it might be a dumb question :) ...thinking about it, probably the ideal answer to this would be something like 1. create workflow in n8n, 2. extract code that does all the work behind scenes, 3. add it as a module to whatever project you are working on, with all the credentials and variables configured with simple yaml file. One can dream :)
This is really great. Thank you so much for this information. Could you please create a video that will tell us how to create an entire book using this. Thank you.
This detailed implementation of an n8n is a great video. Not big on no code since want to learn actual p[rocesses, but interested in these for their triggerr functions. Maybe you could do a video on how to do those outside of these no-code flows.
The video was complete and impressive but because the llm requirements locally, i think it’s hard for now for most of people. I would love to see if you create a video on free agent framework with api like gemini or Deepseek and QwQ, which are cheap alternatives
If I'm not mistaken, when he creates the AI agent at about 5:50, just select another option. OpenAI is one option (4o Mini is pretty cheap), but others can be added instead.
Initially it was PGVector (ua-cam.com/video/TIMOfVhnjQA/v-deo.html), later he switched changed to Qdrant (ua-cam.com/video/TIMOfVhnjQA/v-deo.html). I reckon it's the same Qdrant vector store has been re used for Indexing. Should 've mentioned that.
@@MervinPraison so i should extract to postgres or get qdrant? im super noobish sorry if you explained it already but never used postgres and it just finished downloading should i dl qdrant too?
@@justrobiscool4473 Yes please use Qdrant on both indexing and querying. I have fixed it in this mer.vin/2024/12/self-hosted-ai-starter-kit/ You can just copy the JSON and paste it in the n8n workflow page. Everything should work as expected
@Techonsapevole 's question. Yes you could even use pgvector on both indexing and querying, if you wish. Just compare the performance and use whatever you prefer.
I want to get some insights on a specific problem in RAG. suppose I want to provide domain specific knowledge to llm so that the LLM can analyse specific type of document based on the knowledge I provided. What will be the suitable way to address this kind of problems? I do not want to fine-tune the LLM .
I wish the 100% private dream was feasible with a sub-second inferecne. It's not, without hefty investments in really expensive GPUs (even with 8B models) particularly when considering agentic RAG against big data sources. That's why the only way to go is to use Groq. Maybe we'll get powerfull enough GPUs in the future for AI inferecne that compete with Groq.
When I'm trying setting up Qdrant Vector Sore, the qdrant collection is empty and there is an error "Parameter Quadrant Collection is required", Anybody happen the same?
I wonder, The last time i tried to Create client id and secret it was not allowing to take localhost as host name in gcp ! how did it didn't show any error
so i have the workflow hooked up and all my creds locked in but for some reason i still cant pull simple info from the my docs the whole workflow runs but it doesn't seem to get the info if anyone can help that would be awesome i know mervin says even a beginner can do it so obviously i aint that experienced....thanks in advanced
This video looks fantastic. I am new programming AI, so I have some questions. The first question is, can I integrate crewai to work with this n8n AI solution? I have gotten deep into using crewai and I like how they do a lot of things, and putting the two together seems like it would be interesting. Second question is how do I figure out what should I look for in a laptop? I am thinking in the next year to buy a laptop that is good for running AI, but in looking around I am confused at what laptop would be best. I know chips like nvidia are required, but there are so many of them and reviews I have seen are confusing. They are more like sales pitches than a critical review of the hardware. Maybe I just haven't found the right website. Anyway, your video looks interesting and I can't wait to try it.
Dont purchase yet. Best performance will be in deploying agentic swarms like as is given in this video. This can only be done using cloud compute. Are you deploying agents in yoir work?
@@paulmuriithi9195 Thanks for the feedback. I am retired. I have been learning AI just for fun and to keep active and to keep my mind alert. There are so many fascinating AI projects, but I settled on CrewAI because it makes sense to me. Other projects look great, but you have to start somewhere.
Thanks for the vid. how to solve error - Something went wrong Sorry, something went wrong there. Please try again. When trying to link Google Drive gmail account. Have setup GCP OAuth client creds. enter into n8n and accept scopes, then error/denied.
I'm trying to chat with the agent and i keep getting this error on the postgres node under the agent. i did the ingestion as you did, but it doesn't work. cant figure why
You forget to add the error you made in the querying part using Postgress instead of Qdrant as vectordatabase.. Your final solution works as you change it but noy in the video part 3.
how to put this bot on my website. so it can actually pop up like a chat window and talk to my clients...., must be a way... please add that , thanks in adv
I think you're dealing with vector database in a bit sloppy way - whenever some file is changed on a source you should be first deleting the old relevant info from the vector database, and only then to upload the new one. Otherwise you're just adding new stuff side to side with the old stuff which is already there.
Marvin you dont Speak German, so why the frik you use German Title? Can you not Butcher your Own Language. This Bad Translation is your channel not worth it. With a german Title, I Expect german Speaking content! if you cant deliver that, then do not trick me!
And the physical requirements to run all of this without watching paint dry as tokens trickling?
Great work Mervin! But there is an issue in this setup - While setting up vector database you are using postgres vector store. It should be Qdrant vector instead.
You are awesome! I am at the beginning of a large project and this just might be my gateway. Thank you.
I'm completely frozen. You're a boss!!! Increíble video.
I haven’t made notification ON for UA-cam so far but you made me to do it man… lovely keep it up and do like this.
Love and support from Ireland ❤
This is priceless. Here in kenya, buisnesses are deploying agentic workflows as we speak, and we r using these tools
To do what specifically?
yet another great video Mervin, well done!!
Seeing the 0$ in the thumbnail, I clicked so fast that I injured my index finger. I have to pay 100$ minimum to the doctor 😄
Awesome work! I wonder though, it seems perfect for prototyping and DIY purposes, but can it be scaled to handle multiple users? I've not heard about n8n before, so it might be a dumb question :) ...thinking about it, probably the ideal answer to this would be something like 1. create workflow in n8n, 2. extract code that does all the work behind scenes, 3. add it as a module to whatever project you are working on, with all the credentials and variables configured with simple yaml file. One can dream :)
great demo
This is really great. Thank you so much for this information.
Could you please create a video that will tell us how to create an entire book using this.
Thank you.
This detailed implementation of an n8n is a great video. Not big on no code since want to learn actual p[rocesses, but interested in these for their triggerr functions. Maybe you could do a video on how to do those outside of these no-code flows.
Great work!
agora sim, em Portugues, top top top 👏👏👏👏👏👏👏
The video was complete and impressive but because the llm requirements locally, i think it’s hard for now for most of people. I would love to see if you create a video on free agent framework with api like gemini or Deepseek and QwQ, which are cheap alternatives
If I'm not mistaken, when he creates the AI agent at about 5:50, just select another option. OpenAI is one option (4o Mini is pretty cheap), but others can be added instead.
u can use mistral free api
hardware requirements?
Thank for audio😊
But i think docker its no good because in linux its more vulnerability, Microsoft azure create this app for him server, its not good worked 😒😔
great, why using qdrant when you can add pgvector support to postgres?
Initially it was PGVector (ua-cam.com/video/TIMOfVhnjQA/v-deo.html), later he switched changed to Qdrant (ua-cam.com/video/TIMOfVhnjQA/v-deo.html).
I reckon it's the same Qdrant vector store has been re used for Indexing.
Should 've mentioned that.
Sorry it’s my mistake , it should be the same on both indexing and querying
@@MervinPraison so i should extract to postgres or get qdrant? im super noobish sorry if you explained it already but never used postgres and it just finished downloading should i dl qdrant too?
@@justrobiscool4473 Yes please use Qdrant on both indexing and querying.
I have fixed it in this mer.vin/2024/12/self-hosted-ai-starter-kit/
You can just copy the JSON and paste it in the n8n workflow page. Everything should work as expected
@Techonsapevole 's question. Yes you could even use pgvector on both indexing and querying, if you wish. Just compare the performance and use whatever you prefer.
Cool system! Thanks!!!💛💙
amazing! small typo on the blog for the docker-compose file name
does anyone know any free chatrooms for chatting about things like this...i havent used n8n yet and was just wondering how to set it up
I want to get some insights on a specific problem in RAG. suppose I want to provide domain specific knowledge to llm so that the LLM can analyse specific type of document based on the knowledge I provided. What will be the suitable way to address this kind of problems? I do not want to fine-tune the LLM .
Wonderful, but I have a problem with qdrant config, my qdrant collection is empty :-(
I am using office laptop is it possible without using docker
Nice but is n8n a scalable solution in production environment?
thank you are the best
I wish the 100% private dream was feasible with a sub-second inferecne. It's not, without hefty investments in really expensive GPUs (even with 8B models) particularly when considering agentic RAG against big data sources. That's why the only way to go is to use Groq. Maybe we'll get powerfull enough GPUs in the future for AI inferecne that compete with Groq.
may i know if it's possible to migrate from qdrantDB to chromaDB (OLLAMA EMBEDDING)?
When I'm trying setting up Qdrant Vector Sore, the qdrant collection is empty and there is an error "Parameter Quadrant Collection is required", Anybody happen the same?
I wonder, The last time i tried to Create client id and secret it was not allowing to take localhost as host name in gcp ! how did it didn't show any error
Vera level
good jobe🙂
so i have the workflow hooked up and all my creds locked in but for some reason i still cant pull simple info from the my docs the whole workflow runs but it doesn't seem to get the info if anyone can help that would be awesome i know mervin says even a beginner can do it so obviously i aint that experienced....thanks in advanced
Impressive
This video looks fantastic. I am new programming AI, so I have some questions. The first question is, can I integrate crewai to work with this n8n AI solution? I have gotten deep into using crewai and I like how they do a lot of things, and putting the two together seems like it would be interesting. Second question is how do I figure out what should I look for in a laptop? I am thinking in the next year to buy a laptop that is good for running AI, but in looking around I am confused at what laptop would be best. I know chips like nvidia are required, but there are so many of them and reviews I have seen are confusing. They are more like sales pitches than a critical review of the hardware. Maybe I just haven't found the right website. Anyway, your video looks interesting and I can't wait to try it.
Dont purchase yet. Best performance will be in deploying agentic swarms like as is given in this video. This can only be done using cloud compute. Are you deploying agents in yoir work?
@@paulmuriithi9195 Thanks for the feedback. I am retired. I have been learning AI just for fun and to keep active and to keep my mind alert. There are so many fascinating AI projects, but I settled on CrewAI because it makes sense to me. Other projects look great, but you have to start somewhere.
can i use GPU instead of CPU
Plz make a video on how to use GPT 4o realtime audio and voice using Azure AI Foundry with Python.
Thanks for the vid.
how to solve error - Something went wrong
Sorry, something went wrong there. Please try again.
When trying to link Google Drive gmail account. Have setup GCP OAuth client creds. enter into n8n and accept scopes, then error/denied.
Using a different browser helped and I was shown the consent prompt which was not being shown. approved the app, and OAuth connection is now working.
I'm trying to chat with the agent and i keep getting this error on the postgres node under the agent. i did the ingestion as you did, but it doesn't work. cant figure why
nice!
You forget to add the error you made in the querying part using Postgress instead of Qdrant as vectordatabase..
Your final solution works as you change it but noy in the video part 3.
how to put this bot on my website. so it can actually pop up like a chat window and talk to my clients...., must be a way... please add that , thanks in adv
Are there any workflow tools that are cheaper or free?
All these tools are open source
@MervinPraison thanks, do I need any subscription or tokens for n8n. I am a newbie, so the basic questions
Hey, Google Drive is not local!
True, that is an example . Instead you could get the file from your local computer
Is this all free?
Yes
@ and do you feel - for production ready app - llama is genuinely on par with frontier models now?
I’m in search of Workflow building steps similar to AutogenAI
I think you're dealing with vector database in a bit sloppy way - whenever some file is changed on a source you should be first deleting the old relevant info from the vector database, and only then to upload the new one. Otherwise you're just adding new stuff side to side with the old stuff which is already there.
Marvin you dont Speak German, so why the frik you use German Title? Can you not Butcher your Own Language. This Bad Translation is your channel not worth it. With a german Title, I Expect german Speaking content! if you cant deliver that, then do not trick me!
Sorry, that’s not me. UA-cam automatically doing it