The ULTIMATE Local AI Setup: LLMs, Qdrant, n8n (NO CODE!!)
Вставка
- Опубліковано 5 лют 2025
- 🚨 Join my Skool community: www.skool.com/...
Template for this workflow: www.skool.com/...
Sign Up for n8n: n8n.partnerlin...
In this tutorial, you'll learn how to set up a complete AI environment locally in just minutes! We'll cover everything you need to run powerful AI tools directly on your machine, including Large Language Models (LLMs), Qdrant Vector Database, and n8n for automation. No cloud, no external dependencies - just your computer! By the end of this guide, you'll have a fully functional AI stack that runs smoothly on your local setup, giving you total control and privacy. Perfect for developers, AI enthusiasts, and anyone looking to harness the power of AI without relying on the cloud.
In this video, we'll cover:
Installing and running LLMs locally through the self-hosted AI Starter Kit by n8n
Setting up Qdrant for vector databases
Complete Docker setup
Automating workflows with n8n
And much more!
Get ready to dive into the world of local AI with this step-by-step guide!
#ai #aiautomation #n8n #make #aiagents #tutorial #openai #chatgpt #aitools #generativeai #agi #n8n #docker #dockerdesktop #free #ollama #llama #metaai #opensource #qdrant #vector #database
📚 Join my Skool community (access all n8n templates & more): www.skool.com/aiworkshop
🤝 Need Al solutions built for your business? Work with me: aiworkshop.me/contact/
My install failed because I already had ollama running outside docker. It took me several hours to put all the pieces back together manually using chatgpt and google before I could finally follow along with the video and complete this demo. Thanks for the experience.
I have been working with local AIs to augment my security output. I appreciate the walkthrough of how to use n8n. This one is worthy of a bookmark and sharing with colleagues. Thanks for the thoughtful walkthrough and don’t apologize for your computer being too slow. You aren’t here tonight be fast but to help us who are unfamiliar with the n8n tool. Great work. I’ll be subscribing and seeing what else I can learn about these workflow tools.
I watched the entire video. Great content. Thanks for the tutorial, hope to see more content and tools on this topic
thank you for the video, as usual, very helpful.
Happy to help!
Thank you man this is awesome. Great tutorial
Hey @Zubair, awesome content, keep sharing the knowledge. In the spirit of keeping everything local, how would you approach integrating STT (voice2json, ovoid, go-whisper etc ) and TTS (whisper, Coqui).
Great walkthrough! I have problems when I ask a second question in the same chat session. "[No response. Make sure the last executed node outputs the content to display here]"
same issue. but since the first question. did you manage this?
Great video! Thank you!
Could you tell me how to create a completely local AI agent, but with a knowledge base created in advance, which is presented in vector form. I would like the neural network to look for the answer in these documents, and not attach them in the chat
This is phenomenal. What are the chances you would be able to show how this process would need to be tweaked if it were to be built for internal use in a small company that needed cloud server hosting?
Thanks for your amazing videos, I have already watched some of them getting great value out of them and I also subscribed to your great channel 🙂
I have already coded a couple of very simple RAG systems in python using streamlit and I have an idea of how RAG works,
but unfortunately I don't have clear what exactly are the last 5x nodes for (the ones connected to the AI Agent) , how do they contribute to a standard RAG workflow?
Would you kindly explain what each of them (last 5x) is doing and how do they interact with each other to provide the final answer?
Sorry if the answer is kind of obvious on your video but I am new to n8n
Thanks in advance 🙏
absolutely boss and introduction to this, thanks man!
I'm glad you liked it!
This is so interesting Zubair Sir You could put your knowledge into Ai app how to /auto setup,instructions and questions with so much more possibility's. I would download it quick, your show is informative and educational. Nice
I am planning to releasing something soon. Stay tuned :)
Hello, first of all, thank you for your education, I love AI and I am looking forward to learning from you. I would like to build a virtual instructor for my students. What build would you recommend for this?
Great video. I have one question regarding using a similar Pipeline to generate slides presentation from a small document.
Excellent video!!!
Thanks, glad you liked it!
Thanks! Subscribe :)
Is the question of importing workflows and credentials addressed anywhere? Community nodes?
I learned how to import workflows and credentials after the installation. Due to the backup directory, I suspect that my import process differs from how it was originally intended. Also, I haven’t figured out how to transfer community nodes, so I have to add them manually each time.
Thank you for the quality of this lecture. I'm using an UTM virtual machine to run macOS. Do you think there a way to bypass Docker, which is not supported on a virtualization environment ?
Not that I know of
Thank you for this tutorial !
What is the green icone in your chat prompt ? I think it's an extension in your browser. What is it ?
Is it possible to schedule an automated daily documents upload to vectorize, that get added to the knowledge base?
Amazing video (like all the others, great content!). I've setup n8n etc per this video, I'm having problems with Telegram trigger node not working (keep getting Bad Request error) - Can you please help? I want to work on some of your other workflows.
seems like the telegram node on local host is having issues.
Enjoying your n8n series 😊. Thank you
Thanks @tonyblack2141. Appreciate it. I am just getting started :)
At the end you show how to expose the chat window for use in a 3rd party app/site. Is it also possible to expose the chat as an api using the openai format in order to style it yourself?
Exec in Docker requires a paid subscription and the YML manipulation doesn't seem to work as the new models are not showing up. Any idea?
How does this compare to the supabase /pgvector based solution you covered in another video?
Thanks a lot this is great. But if I ask a second question after uploading the file I get this response: [No response. Make sure the last executed node outputs the content to display here] I just started to practice and learn so I hope you can help me solve this. BR
This is amazing!!!! I was working in something similar thanks for the video!!! Quick question: how can I implement OpenWebIO as frontend on this? Not sure if it's possible
Question. What should you do if you already have ollama installed? Do you have 2 ollama instances, or what? Thank you!
I also had ollama already installed in windows, so all I had to do was modify the port numbers in the docker-compose.yml file to get the docker version working. For example, in the x-ollama section, change the ports to '- 11435:11434', and then in the x-init-ollama section change it to 'OLLAMA_HOST=ollama:11435'. You are basically changing the port number that the docker version uses to 11435 (you can pick any free port number you want) , which means that it won't clash with your previously installed version that uses 11434. You can verify afterwards by opening your browser and going to both localhost:11434 and localhost:11435 and they should both say 'Ollama is running'. If you then shut down the docker version, only the localhost:11434 will say that 'Ollama is running', which is your locally installed version.
I would love to see the setup of this on a remote VPS. I have seen only local setups so far. What about external with traefik included ?
yup planning on doing a video on this soon
great video Following your series Amazing content
Thanks brother. Glad you found it helpful. I got some great things in the pipeline so stay tuned :)
Thanks!!
with this workflow we cannot ask followup questions, right? how to make sure we can?
This is a great tutorial. I have one question around how this chat interaction is happening with PDF file. It responds to user question. How is this different from a RAG @AI-GPTWorkshop
How much GPU power is needed to run the flow at a decent speed? ATM I only have a 1gb nvidia running my screen: at most I could get a RTX 3060 12gb in January on my desktop system.
Thank you for this amazing tutorial.
hello, this is the first video of your channel that I see, congratulations..
now I will go back to see the previous ones too.
is it possible to set the response in stream given the slow pc?
i watch every step rewind and rewind but if I ask a second question after uploading the file I get this response: [No response. Make sure the last executed node outputs the content to display here thanks mr.
bro, quick qustion. I see you put local in you title. meaning this should only run on a chunky computer versus an online service?
Many Thank you for the video. This is very good knowledge for me. If you don't mind, I have a Question about the system.
After upload the PDF file once, I can't ask more questions about the other contents of the PDF file. Every time I have to upload PDF file again. It would be great to talk with your all documents after upload files in one time?
this was just to showcase that the user can upload files them selves. You can always add a google drive node too
Hi any idea why I updated my n8n but dont see the new nodes and new agent tool node ?
1.66 version
please show us more about tools and use cases if possible use as many or all tools available also is there a way to use the same code in a streamlit app or similar app?
yup planning to. stay tuned
thanks
Additional information, for the new users like me, if you get stuck with the following msg.
[No response. Make sure the last executed node outputs the content to display here]
Make sure you go on the qdrant node after the chat trigger, go on settings and set always output data to green, active.
Otherwise, the workflow will get stuck on the qdrant block.
For some reason in the video it does not show him doing it or is already set by default to be active, for me that setting was off and it was the reason I was stuck in the loop.
Thanks for pointing that out. I had my workflow active. I must have forgotten to point that out
@@AI-GPTWorkshop No problem happy to help people same way you are helping me, and thanks for your content ;)
I do not know if that comes like that by default or if I did something to work that way, but just in case someone comes to the same situation ;)
Suggestion that you give details on container versions your tutorial works on (not rely on "latest" from n8n). This video is already out of sync with available containers 1 month in. Problems persist within couple minutes of video. No one can follow along without debugging themselves. Which should drive you to another author.
how to lunch n8n start --tunnel?
Hello In the minute 3:35 you have 6 containers installed... I did the same steps and I have only 4. Why?
the fix is: use docker-compose --profile cpu up
Can you share the workflow template exported from n8n?
all templates are available inside my skool community
Awesome video,
If you ask it a second question, it doesn't work. I have to upload the document for each question, is there a way to fix that?
NVM, figured it out :)
@@xedrick I get the same issue. What was the solution?
I did it step by step and the result is an error: Problem in node ‘Qdrant Vector Store‘ fetch failed. TypeError: fetch failed at node:internal/deps/undici/undici:13185:13 at processTicksAndRejections. I did some Google research about it, but no luck so far.
Not sure why but the Qdrant Vector won't work, I keep getting the same error message everytime I try: Problem running workflow
Please execute the whole workflow, rather than just the node. (Existing execution data is too large.)
Why do i have to attach the file again to ask a second question ? for a second quest from the same file it asks for to attache the file or no response message is given . How to Solve ? Overall this ia a wonderful tutorial
I have the same problem, did you find a solution?
You might need to add chat memory.
Sir, i am new to all of ai. I only know how to play minesweeper and write in Office. Please could you explain what the programs you use actually do? What role so they play in thisbwhole thing? Like, what is docker? What is n8n? Etc. Also, does this work on mac only? How to make it work in windows? In linux mint? Thank you, i have only a few braincells so it's dofficult to understand for the many people like me. Thanks for your future videos!
I’m not all the way thru the video so sorry if you answer this but how much space does this take on my local machine
depends on how big of a model you download. it shows the size next to the model on the dropdown list on Ollama
How do you deploy it with a front end to vercel or replit or digital ocean droplet ?
there is instructions on n8n's website on how to deploy this on digital ocean droplet. I am thinking of doing a video on this
is there a way to use openwebui for the user input that connect to the n8n workflow?
yah you should be able to do that
Why are we adding the ollama chat model two times to the agent?
one is for victor store retriever and one is for the ai agent
Hey, will you share github repo with bundle?
hi anyone got this error when installing: context canceled
Great !
I tried with Ollama at my laptop for learning, it is nice but somehow slow compared with langflow. But n8n offered more tools and extensions.
yah it means your default memory is low. Try downloading an smaller model
Can we host it in cloud ? Or what u recommend for more advance performance @@AI-GPTWorkshop
This doesn't seem to work unless you are attaching a document. If you just give it a text input: [No response. Make sure the last executed node outputs the content to display here]
Would love to see your tutorial on how to implement this with DigitalOcean.
There are many existing self-hosted n8n users, and we don’t want to start from scratch. Hopefully, this idea can inspire you to create a tutorial on how to onboard the AI starter kit with an existing DigitalOcean self-hosted setup 🙂
Yah I might do a tutorial on that since a lot of people have been asking
great tuto
are the chat and embedding models of open-AI free to use in n8n?
of course not, open-ai stuff is paid anywhere, there is no way to have gpt free as a service, only in the chatgpt website as a chat, not as api
definitely not free lol
How to repeat about the same thing, but working in an offline environment with non-public NDA data.
n8n is not user-friendly, I installed it, but it seems it is barely usable, like other similar apps.
What editor is that please at 11:03 ? Thanks
vscode
@@vhater2006 Ah ok... thanks a lot :-)
Amazin
am I too stupid? - i did all points explained in your tutorial. my chat does not take into account the content of any file that I upload...no errors in logs or anything. Just ignoring any information presented...
I did'nt know that git is a basic command of osx
I am getting an error, please execute the whole workflow, rather than just the node. (existing execution data is too large)
same can not pass from the embeding part no matter what, and if I do a smaller file, it tells me that there is no connection to this windows, connect the last node to this window, I dont get it how it works for him and not for me.
But that msg I believe is from the file you are uploading, it is to big
ohh btw, I found out how to fix my problem too, needed to set the qdrant block to always output data, on the settings for the qdrant block ;)
otherwise I was getting this error
[No response. Make sure the last executed node outputs the content to display here]
but yeah the error you mentioned, existing execution is to large for me was because I was trying to give it a 25mb pdf, and it can not handle it I think, because with other documents it runs just fine
Use AnythingLLM instead
DONT FORGET TO DOWNLOAD GIT DAMMMMMM
Just a piece of advice. If you get “intimidated” by code, don’t work with technology. I mean, there are many other fields you can explore. Just forget about tech.
Not knowing code is completely fine. But being afraid of code and trying to do stuff in tech, makes no sense.
Tech has nothing to do with code, you can do a lot with a computer without knowing any dev language. All you need is tools that do the code for you. Even Unreal Engine has a way to develop games without code, with blueprints only, so I don't see what is the problem with people not wanting to touch in code.
I love it, but tools like this are amazing when you don't want to lose hours debug in your code just to make something simple like a chatbot or some kind of workflow like the ones you can do with this. To make this with code, you would spend hours when you can do it in minutes with n8n and similar tools
@@kiiikoooPT true. But if you are intimidated by code than you should probably not choose a profession that revolves around it
@@rfn900 like I said not all in tech depends on code, my uncle used to work in a library with Mac and freehand Corel drawn to make the pages of the books, he didn't knew anything about code but he worked with computers every single day, using code is something that we can retire specially nowadays with the tech we have, the same way you have high lvl languages like python and java, we can also have platforms with higher lvl like n8n where the code is done already and all you need is to play with the GUI no need for code at all... Well we need it but it is already done in the background... Your mentality is what makes people more afraid of trying stuff.
As someone who has worked in IT for about 40 years your statement is complete nonsense. I have written code on PCs, mainframes and mini-systems in all sorts of languages and scripts, working my way up to Senior Analyst/Programmer. However, I also worked my way up to IT Director level for a multi-billion dollar company being lead technical in the role of IT Manager for most of my career. Coding/scripting was a smaller part of my career, but something I really enjoyed. When I left my last position, the guy who took over was very knowledgeable about servers, virtualisation, storage, networks, vlans, phone systems, applications, cybersecurity, you name it he was pretty good at it. The one thing he couldn't do was write code. He had tried, but just didn't 'get it', and the thought of having to do some sort of scripting/coding would terrify him. I have no doubt that he knows more about technology than you do! I also have a friend that I went to school with who wrote a Space Invader clone in assembly language, and later worked at Bloomberg for many years as a senior developer. He would often write things like device drivers, so he was a very technical programmer. However, he wouldn't have a clue how to set up the IT infrastructure that a company uses. Setting up email servers, hypervisors, SANs, vlans? Not a clue! When it came to that, he was just like a very advanced home user. So does he really work in technology if all he can do is code?!
i did not get ollama