Hi Steve, thanks for purchasing the template. Let me reach-out to you via email and get you up and running

​@derekcheungsa i figured out the issue. Nocodb recently got rid of workspaces completely for self hosted instances, so it just says No Workspace. My auth worked after deleting and creating g a new one.

@@Steve.Goldberg Awesome! Great to hear Steve.

Hi Cristina, I've sent update via email. Should be on gumroad now as well.

Thanks Cristina for your feedback. I'm on 1.72.1. I've updated the gumroad template to be backward compatible with earlier releases of n8n. Please download the v9 template.

@@derekcheungsa the v9 template is not avalaible in gumroad. Please can you help me with that?

@@CristinaPena-b2z I think the gumroad server might need just a few minutes to sync. It's there now. But let me email it to you to make it a bit easier 🙂

Thanks! Good luck 🙂

Thanks for buying the template. Unfortunately, Deepseek doesn't yet support multi-modal, so that LLM can't be used. In my evaluation of the LLM's for this use case, I found Claude Sonnet to be the best. I have also seen good results with gpt-4o which is slightly cheaper than Sonnet. I don't recommend weaker models like gpt4o-mini as the results are not as good.

@@derekcheungsa cool thanks for the reply ;)

Hi Jim, yes, absolutely. That's a good idea. There's two key things that are needed: the meta info for the table column names and then a filter mechanism. nocodb has a simple sql like filter that makes this easy, but I can see that postgres db such as supabase has a simple filter ability as well. So this should be possible to adapt. Do you have use case you are thinking of?

​@@derekcheungsa Ah yeah, I am a data analyst working for a company. It would be really convenient if I could interact with the database while conducting data analysis to gain insights

​@@derekcheungsa Ah yeah, I am a data analyst working for a company. It would be really convenient if I could interact with the database while conducting data analysis to gain insights

​@@derekcheungsaI'm thinking of using supabase and connecting nocodb to it as the data source for easier management. Have you ever done this? Also, would love to see a crash course or video going more in depth on using nocodb for using formulas, prompt templates and prompt chains instead of putting prompts in n8n, which is a pain to manage and keep track of.

Thanks Ramon!

Yes, a bit more work needed to send to telegram node that displays the image as a photo. I'm curious the use case you are thinking of.

@derekcheungsa I want to create data analysis agent which can be sent the output to telegram both chart as an image and insight, can you guide me how it can be send?

Generally, you would add a telegram send text node to the data analyst agent. And in the Visualization Tool workflow, add a telegram send photo node. There's a few tricky parts: before sending to telegram send photo node, you'll need to download image first. Additionally, you'll need to store the chat_id from the telegram trigger into a global variable so that it can be referenced in the send photo telegram node in the tool.

@@derekcheungsa thanks for ideas, do you have any video explanation on the telegram integration part?

thanks for the feedback. happy to help get this sorted out. Can you send me a screen shot of what you are seeing via the support email? Additionally, there are now additional setup steps in the gumroad page.

@@derekcheungsathanks for the quick reply! Was trying to find an email before I commented but must have missed it can you point me to where it is? Will also look at the gumroad again

@@TheIdosimon Just following up. Were you able to find the information you needed?

@@derekcheungsa Thanks for following up, got the perplexity stuff working. Is there a way to auto publish with ghost or can it only do auto draft and manual publishing?

@@TheIdosimon Yes, you can use a custom api call in n8n to directly publish without auto draft. Here's how you would go about doing this: ghost.org/docs/admin-api/

If you use it as a tool, it does :-)

Thanks very much! I appreciate it.

Thanks for the feedback. It uses the perplexity api as a tool in the automated workflow and ability specify the length, number of sections, tone of voice in generation. It also uses a form of chain of thought to break down the work first by creating a table of contents. This workflow can also be adapted to create full ebooks :-) That's something I will work a bit later.

Thanks for your feedback :-)

Thanks!

If you have a specific use case idea, please let me know :-)

@ flowise has an E2B and Anthropic flow where you can do data analysis and generate charts. Similar to Pandas AI functionality

Yup, it's already there. All my paid templates are in the course.

@@derekcheungsa Thanks Derek I signed up ;)

Do you mean to select which db at runtime? If so, yes.

I mean, if there's an update with google sheet data, can it persist(realtime update in) thru nocodb. so that i will not manually convert the googlesheet ro csv then download the file to nocodb

@@kitlee888 I see. Yes, to do this, you setup a monitor flow that monitors for changes in gsheets and then makes the update in nocodb.

@@derekcheungsa thank you, sorry i'm just a beginner in ai automation but very interested to learn(hopefully there's available n8n template for this use case, ill avail it to test and learn). thanks again.

Yes, I've updated the video description with link.

thanks for the feedback! You can hook up to gsheets with the row of websites you want and then connect to split node that effectively acts as a loop. You put these two nodes in front of the scrape workflow

@ thanks dude. Appreciate the response 😎🔥

Hi Marc, thanks for your comment. Please keep in mind that the example I'm using has 14,000 records and over 40 columns. I'm pretty sure this approach can scale to 100,000 rows or more as it uses essentially an sql approach to get exactly the data needed to add to the context window for the LLM. One of the key insights in this video is that with the nocodb, the sql needed is super easy unlike traditional sql needs essentially LLM code gen and more prone to variability.

Could you share the csv?

Thanks for the suggestion. I appreciate it.

Thanks for the question. No, csv agent won't scale to this size of data set. There's 14K rows.

On railway.app

Hi Richardo, unfortunately not.

Thanks for your feedback. n8n uses the langchain framework for it's visual AI implementation.

Thanks for the feedback. Indeed it's important to use AI with a co-pilot mindset. We have to do our own due diligence with the output to verify. But that's where the citations helps because you can then see the source.

Yes, I’m asking specifically because I have perplexity pro and it often gives fake urls

That is so true

😮😅 😊 0:21

Thanks for your feedback!

Thanks :-) Yes, I've just sent-out the newsletter with the template included. This is going to be a paid template on gumroad soon. As for comparison to langgraph, n8n is very powerful and I find more expressive -- for example, a split node already parallelizes agent execution.

@@derekcheungsathanks! Any idea how to get perplexity to stick to a time frame..Im finding it can pull really outdated info thx

I've added this to the bonus section of my udemy course and have sent out a link to download this to my newsletter community. I want to give them first access :-)

@derekcheung2598 thanks I have already purchased your course on udemy, but no link 😢

​@@AIVisionaryLab thanks for your support! It's in lecture 32, in the resources section.

Thanks Kris. Yes, it's because I need to get the citations from the output. This needs some additional processing that I can't do with just the http request tool.

Great question. There are two main factors in cost: token count and model cost per million. gpt4o-mini is very inexpensive while gpt4o is a bit more expensive. The Research Assistant agents do a lot of the heavy lifting in the workflow and they use mini. The Editor node requires a stronger model to get all the editing of the citations right, so it uses gpt4o. For an article of 1000 words, the cost is about $0.10.

@derekcheung2598 not so expensive. Nice.

Yup, I'm uploading this to the bonus section in the course along with the template. Should be available by tomorrow.

​@@derekcheungsa I bought your Udemy course some time ago. And now I'm wondering if it's the same one, or do you have multiple courses on Udemy?

I bought your Udemy course some time ago. And now I'm wondering if it's the same one, or do you have multiple courses on Udemy?

@@funnydogfargo1026 Thanks for your support. It's the same course on n8n. I'm adding the most popular tutorials and templates in the bonus section of the course. You can download the template now from the bonus section.

@@derekcheungsa nice. Thank you.

For complex tables and graphs, using LLM vision approach will yield better results. For simple text, pdf extraction node works well

Thanks for your support!

Thanks for your feedback. In your chat window, to the right, you'll see a log of all the calls that the agent is making and the output of the call. Also, at the top is part of the screen there is an Execution tab. You can use that to see the values of the nodes as well. But mostly, you should be able to see the chart visually displayed. Make sure to set the exchange variable in the settings node. it defaults to NYSE

@@derekcheungsa Got it, it was being weird because I was also changing and testing right away, but without saving it. Thought it would use the current instance. Thanks!