November 4, 2025 • 39 mins
Join the Tool Use Discord: https://discord.gg/PnEGyXpjaX
Master n8n workflow automation in this complete guide. If you're tired of manual tasks, n8n (an open-source Zapier alternative) might be the solution. Richardson Dackam, founder and AI builder, joins Mike to explore how developers can save time using n8n. We dive into powerful use cases like email management, social media automation, and prototyping. Learn how to integrate AI models (OpenAI, Anthropic, Gemini), run custom code scripts, use webhooks, connect to databases, and implement human-in-the-loop validation. Richardson also shares common pitfalls, observability strategies, and the difference between self-hosting and the cloud.
Learn more from Richardson:
https://richdackam.com
https://www.nshipyard.com
Connect with us
https://x.com/ToolUsePodcast
https://x.com/MikeBirdTech
https://x.com/RichardsonDx
00:00:00 - Intro
00:00:32 - What is n8n? (Open-Source Automation)
00:02:23 - Common n8n Use Cases (Email, Social Media)
00:03:36 - Automating Email: AI Tone & Human in the Loop
00:07:02 - Observability, Logging, and Error Handling
00:22:00 - Using Custom Code & JavaScript Nodes
00:29:50 - n8n Cloud vs. Self-Hosting (Features & Security)
00:37:01 - How to Get Started with n8n (Beginner's Roadmap)
Subscribe for more insights on AI tools, productivity, and automation.
Tool Use is a weekly conversation with the top AI experts.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
After three consecutive guests, I've talked about the value that
they got from N8 NI hit my breaking point.
I just had to try it. But The thing is I didn't want
to go in blind. I wanted to find the best person
I could to teach me about N8 N before I actually used it.
And I'm going to bring along forthe journey.
So in episode 6 of our tool use,I'm joined by Richardson to
come. He's a founder and an AI builder
(00:20):
and he's hosted and it had master classes before so today
he's going to cover why every developer should consider using
N8 N. Some great use cases to really
save you a lot of time and some pitfalls and gotchas that you
need to look out for. So NNN is essentially like an
open source workflow automation builder.
Its advantage is that you cannotcreate like workflow and it was
designed with developer in mind.So like you can actually see
(00:44):
just an output, you can even edit the response, you can add
code to it. So it's pretty intuitive.
So not just developer can use it, but it really give people
that know how to code the ability to do even more complex
stuff. And on top of that, because it's
open source, it has a pretty bigcommunity and you can build
custom node. So if you take something like
(01:06):
ZAP here, for example, if, if there's a feature that you want
to have on it that is not there yet, you will have to wait for
ZAP here to build it. But on something like N8 N, the
community can build it or you can even hire a developer to
build custom node for you. So that's flexibility is what
makes it official. That's so cool.
And I mean, we were talking before.
(01:26):
I've been a software developer for 10 years, big and open
source too. But there's something about this
drag and drop interface that I just didn't really like.
I like the idea to code everything, control everything,
transparency all the way down. But more or more developers are
saying the same thing where there's like it abstracts away
the integration work. A lot of the the drudgery with
this job of trying to make systems talk to each other.
(01:46):
So who? Who else gets to benefit from
this? So anyone that has like, like a
work for within the company thatthey would like to automate.
So you take a task and you thinkabout how you can make that task
into a workflow that is repeatable, then you can really
like take advantage of that And it's powerful enough that you
can create enough like systems to replace an entire department,
(02:10):
if any new company. So some people have built a
really complex workflow that take care of the wall of the
wall thing. Especially when you combine that
with AI that can be the decisionmaker, then you can it give you
a lot of power to do like reallycomplex automation.
What are some common or popular workflows that people will
automate on N8 N? So like a common one will be
automating emails management. So like they say even for
(02:32):
refund, if you're getting refunding you instead of hiring
someone that is going to go through the e-mail, parse them,
label them, you can use a workflow to do that.
So what you do, you first think in steps like what happened in
every single step and you're thinking input output.
So what's coming in and what's coming out?
So then you can take something like an e-mail.
(02:53):
You can say read my e-mail, findany e-mail that mention refund,
label them refund and then writea response to to to the customer
based on this set of criteria. So emails, e-mail management is
a big one. Customer support is a really big
one as well. Social media management is
really big. It's getting bigger and bigger
(03:13):
because you can automate postinginto multiple across multiple
social network, especially when you have like, you know, people
making content on YouTube and they want that same content to
appear on TikTok and you know, all the other platforms.
So you can use N8 N to actually automate that entire social
media posting. I definitely need to start
looking into this more because I'm manually posting a lot of
(03:35):
cross-platform stuff, but let's let's hone in on just the e-mail
example. Just try to give people a little
more inspiration as to what theycan build.
I hesitate having AI write emails because I don't know what
the quality is like, is it goingto match my tone?
So can we get through a little granular?
How do we get human in the loop?How do we kind of set the AI for
tone? Does it have to be a certain
LLM? Like how?
(03:56):
How do we get that workflow moremore polished?
Yes. So the feeling you just
mentioned human in the loop is abig one.
You can have a human verifying astep before before it's
submitted. So you can actually be part of
that workflow as human in the loop.
The other thing you can do is for having templates.
So for for a lot of work operation workflow, often time
(04:20):
you can just, you know, simplifythem into templates because it's
often times similar responses. So you can have pre written
template that you just use instead of of the the tool and
just add or update the variable that you want to edit inside of
that template. And that's the way that you give
you the consistency of what to expect, which is good for brands
and tone because you don't want the language of your brain to
(04:43):
change based on every single response, right?
So I think template is like a really good good tool.
Nice. And for for the aspect of
templates, I actually had Lee Russell on a couple episodes ago
and we were talking about the idea of when you just customize
the the user's name, Hi, first name it.
It's kind of expected. Now, it's not really that
magical experience. Can you have it where with along
(05:03):
with this template, you actuallywill have something more custom.
So let's say like I first name and then a paragraph related to
their specific instance and thenthe generic template.
Does that work? Yes, this is this is where stuff
that work for the use AI comes in into place.
Because now you can take a take a portion of the input being,
for example, the the job of the person, the position or
(05:26):
something that they've accomplished and have AI
customize that, that intro section specifically for for the
person that's going to receive the e-mail.
So you can actually make it morecustom by having AI deciding how
to rewrite those different parts.
OK, cool. And for the AI, do you, do you
have to use open AI kids you have different, is it just like
(05:46):
an API call? How do you actually plug an AI
into the system? So it's pretty flexible.
So you have it give you access to different type of AI models
that you can use to can use anywhere from Grog to Entropic,
Google, Jimini, they have so many now.
So you can use any of those those tool in it.
And it gives you that ability and flexibility to choose what
AI model you want to use. And the way you do it is you,
(06:08):
you, you go on Entropic, for example, which is the one that
build cloud. And then you take your API key
and you add it inside of your NNN credentials.
And then you can now use that model specifically, specifically
for one node or different node that you're adding to it.
So so yeah, you can bring, bringin different AI models as, as
based on on your need. You're not limited to a single
(06:30):
one. And you can even use custom one
if you use something like Olama which is something that a lot of
developer like to use or open routers and stuff like that.
And and sorry, how would that work?
Set a custom API URL and just point it towards my server and
then just like regular request response.
Exactly and or like if you're using open router then you have
even more flexibility because they have such a large database
(06:51):
of models so you can pick different model that you want to
use. And kind of leading back to to
human in the loop. I think it's super important for
you to keep people informed as to what they're doing and
insight as to what's happening. Observability in general is a
concern when using systems like this because if a pipeline is
just running, it's in the background, maybe there's a step
along the way, causes an error and by the time it reaches human
in the loop, it's already cascaded.
(07:13):
Do you have any advice or strategies on how if someone
uses anything, they can get better observability, logging,
tracing into the process? Yeah, that's actually a really
good plan because error and logging is a thing that a lot of
people that begin to use and always forget to do.
So you can actually like making sure to have no to just handle
errors and how to deal with errors and where you want this
(07:33):
error to go to. So one way to handle that is if
you're put a human in the loop, you can have some of the error
going to the human in the loop to review them, or you can have
a sender log via e-mail or to toa platform.
So me personally, for some of mytool, I like to build my own
dashboard. So I will get these on my own
(07:54):
dashboard that I can then, you know, review and you can also
send it to Slack, which is very common for a lot of companies.
So they have all those error directly coming on Slack and
they know how to take action whenever those come in.
Is it possible I just use Slack as an example?
Error comes in, human acknowledges it, they know the
solution is it. Are you able to automatically
(08:15):
have triggers from outside of NAN send information back in?
So say I say like close ticket or respond with with whatever
response. Can that be done in Slack and
send back to NAN? You you have this thing called
webhook that essentially are like listeners that can listen
to even outside of NAN. And so that will be fit into the
category of triggers. So NAN has like different type
(08:36):
of triggers that you can have like time bays that react to
responses and so on. And so for example, if you, you
wanted to respond to event happening on Slack, it could be
for example, when I say yes, send an event to that webhook
and take action based on that. So it would be, it would work
usually as if it's a trigger as part of a workflow.
(08:57):
I have, like they said, the begin of the workflow.
This is what happened. I haven't seen a lot of people
using it as it being in the middle of a workflow itself.
So usually it's more done as a trigger.
OK, so custom triggers make a lot of sense.
Love the idea of webhooks. It just allows you kind of
percolate into any system no matter how custom.
Can you kind of do similar thing?
Webhooks on the other side whereNAN is able to notify a webhook
(09:18):
when it's completes a task or stuck in a task.
So that's a very common use for notifications.
Also, for example, people that want to use stuff like Telegram
or WhatsApp, they can send a webhook to some of those
platform to get to submit something.
I thought even with type form. Type form is a very common use
with web hook as well. You can receive the web hook
(09:39):
there as well. And if you want to send it, you
can use it to for example, any time a user take an action, you
want them for example, to to getaccess to a form.
You can also send, send an e-mail.
You know there's so many variables that you can do with
that in and out. Nice.
And I kind of assuming it's yes,I just want to make sure.
Can it connect to databases? Like can I have it actually
(09:59):
write entries to the table? Yes.
So a lot of people connect even different type of databases.
So yes, the answer is yes. Some people like to use Excel to
keep it simple. Some people use like super base.
You can also use RADIUS like a cache database vector is a
really big use case if you're using AI because in NAA there's
(10:21):
this thing called clusters that are often built for reusability.
So they will have the cluster connected to a vector database
if you need to store some information and process for
making it easier for the AI to have memory essentially.
Sir, could you go into detail what is a cluster?
(10:43):
So a cluster is kind of like a group that make an action
reasonable. Instead of NADN there's
different type of cluster. You can have a chat cluster that
essentially is composed of an AImodel how how the AI model is
going to get the message, a database which is where the
memory is going to be stored. So instead of having to build
(11:05):
every single different componentyourself, you have that cluster
that everything already pre built and ready to use.
And I just make it a lot faster to build more complex nodes.
So that's something that a lot of pro tend to use it because it
makes things a lot faster than having to set up everything from
scratch. Yeah.
So instead of regular atomic units, you have this this Group
A cluster to to plug and play. Does the community develop
(11:26):
clusters like is there? Is there like a library I can go
to be like oh this matches my use case insert?
So that's what Jenna, I'm not sure, but usually when I see
when I see community contribution is usually notes.
I've personally used a lot of notes from the community.
I haven't seen a cluster. I haven't I haven't need to
discuss that. I'm not 100% sure.
So this sounded really exciting.I love the idea of being able to
(11:48):
just automate a lot of workflows, but I imagine as
people start exploring this technology for the first time,
myself included, there didn't bemistakes.
What type of common mistakes do you see people make?
And when they make these mistakes, what's the debugging
process like? I would say like one of the most
common mistake is approaching NNN as like just like making
tasks. So finger of tasks instead of
finger of flow, because it's very important to think about
(12:12):
the information that comes in. So one way, one way I like to
put it to visualize it is it's kind of like they like kind of
like Legos legal with data, right?
So every time you create one legal to the other one, you
passing the data to the other Lego.
And then and then as you're doing that, you cannot be
building this, this supply chainor, you know, or if I put it in
(12:33):
the sense of like four assembly line, right?
So every assembly line is getting different data.
So instead of making, let's say,your car, instead of making
object, you, you're taking the data and then put passing into
the next thing. So a lot of people tend to think
of and and as just that, but it's really important to think
about it as an assembly line when you're passing that out
from one end to the other, because this is what allow you
(12:56):
to actually transform the information and the data and
then get to an output. So also that's like probably one
of the most common one. And in terms of the the pipeline
aspect, the assembly aspect, is there a lot of requirements for
data transformation? If I have data coming in a
certain format, do I have to go in between every node and map
fields to each other? Or is there any type of
automation there? Yes you do you.
(13:16):
You have to map field sometime. What you do is you can.
So, so you come in the format called Jason, and then you can
kind of like decide, say that for example, this node is going
to output this different field. And on the next node, you can
grab the value from from the input or the value from the
response and pass and pass that to where you're manipulating the
(13:39):
data. So for example, if you want to
add someone's input from a PROM inside of instead of to an AI
model, you will take that input and and and and pass it to the
AI model. When the AI model generate a
response, you will use that response and pass it to the next
node and so on. So you can transfer the data
that way with it. Nice.
And just for that process, do I need to know from either
(14:02):
previous experience or other experiments what the data fields
are? Or will any then display?
This? Is the output this is the
required input? Like match things.
So there's two things. One, you can decide how you want
the data to be outputted, which is very important to have the
file say you need to think aboutinput and output because you
need to know how you want your data to be presented for the
next node. And and once once you've created
(14:25):
this, this, I guess this preliminary, then the next node
would know what that I can it can it can take in and actually
this is a tension, but Asian builder by open a actually do
that really well. The way they allow you to
visualize what comes in, what comes out.
I think they did a really good job of that.
Is there 'cause you mentioned open source so people will build
(14:45):
their own nodes I guess? Is there any work done on the
interface? Like do you think that any then
will catch up to open AI's agentbuilder in terms of UI?
That that it doesn't seem that that's the priority right now,
because I was reading something that the CEO of and, and what I
think his name is Jen open, openiser like I have lost his name
(15:08):
wrong. But technically what what he
said is that their vision is to build like the excel of AI.
So they really want to build something that people can use at
the orchestration layer. So like they want to really
focus on that location layer, while open AI is really focused
on building good product that are very easy to use.
The other thing that they're going to do is they're going to
(15:29):
add a way to generate workflow and that's going to be purely
for marketing purposes to in order to get more people to use
N8 N, but not they're not reallygoing to dive into that
generation, but mostly to if someone need to quickly have a
word from to it instead of N8 N,they'll use a prompt and write
what they want to build and they'll pre set up the
different, the different notes for them.
(15:51):
So that's their current focus right now.
I don't. And also because they target
market is is usually more technical, while Asian builder
is targeting the average consumer, the average like your
customer and company. So like I don't think that
they'll try to do something similar to open the eye.
Pivoting slightly, you mentionedemails, a good use case writing
(16:11):
to accelerate database. Could you go over a few other
like integrations nodes, maybe some that aren't so popular that
you found it but actually been very helpful?
Like what? What type of services should
people keep in mind when they'resetting up their workflows?
So something that I people don'treally think about this like RSS
feed, you know, a lot of people,for example, listen to podcasts.
There's a lot of ways that you can utilize RSS feed or for news
(16:35):
or for podcasts. I think that's like a really
interesting use case. If you want to have like an
agents that check your, your preferred news source all the
time or preferred podcast. And if you want to only, you
know, check the pockets you're actually interested in, you can
have filters. So that's like a really like
creative way to make it into your little, you know, a robot I
can like prepare the news for you and so on.
(16:56):
The other one will be I mean foam trigger is a common use
case as well, but people don't really see that they can use
that for for more than just forms.
You can do that as like a payment form as well, because
you can have a form to take on payment and you can have a
webhook to respond to the Stripewebhook request.
So you can actually build entirepayment system just from NNN.
(17:19):
Often time like as as a developer, the thing that bring
me to use NNNN is that if I think about a workflow that I
could build into a website and Iwant to quickly see how that
would function. I will quickly build a prototype
on NAN and and see it working and then allow me to quickly
test something and having an endpoint for someone else to
(17:39):
utilize. So I could for example give that
end point to my app if I wanted to do something complicated.
I don't want to build full edge yet, but just create a quick
note to see how it would work. You're not the first person to
bring up the idea of using it for prototyping.
Do you find you'll hit a ceilingor a certain level of
functionality where any end's nolonger appropriate, and then you
take the concept and put it all into code?
(18:00):
Or have you? Is there even a ceiling there?
There's a ceiling. So, so sometime if your task,
for example, involve like using browser AI agents or doing
scraping or stuff that have to manipulate the browsers and
stuff like that, I think it's a lot easier to build it in code
because you mentioned observability early on and
that's like a really big factor.And I think that N 8 N sometime
(18:23):
lack of visibility because it doesn't shows you the the error
right away. So it's sometime easier to
actually build that inside of like, you know, anyone if I'm in
cursor or you name it, because then you'll be able to, to
really see the error output and a debugging output.
So if something get too complex where being being able to see
the debugging log is important, I think I would say stay stay
(18:46):
away from N8N if you can. Just like programmatically set
up things and where the input and output are very clear that
it's an really easy tool to use to build quick prototypes.
OK, cool. And and that that's good to
know. And you had mentioned that there
are some very complex workflows that exist.
Do you have any advice on how people architect or lay out
their NA then set up to help with maintain ability where
(19:09):
instead of just like a random cluster of nodes that can be
overwhelming not knowing where what where what is any way of
like organizing or structuring anything like that?
Yes, so you can think about, forexample, there's, there's,
there's stickers note that you can add so where you can explain
what different area or part of the workflow actually doing.
(19:29):
Those are very common. It makes it also easier for you
yourself when you look at your work for after a while, because
sometime you may forget, you know, what is this work for
doing again? Now on social media, I've seen a
lot of of like images that tend to go viral of people creating
these massive and workflows thatare very, very complicated.
And I often see NNN commenting on it and say, have you ever
(19:51):
heard of cluster or have you ever heard of stub nodes?
You know, so there's a way to simplify it by looking at
cluster nodes and looking at Subs node.
Sometimes Subs node can allow you to do some of those stuff
with a cluster. Are there any other like meta
nodes that don't necessarily help with functionality but help
with the developer experience? Do you call them stickers where
the the node that kind of saves?Is there anything like that
(20:13):
which doesn't impact the flow ofthe data but helps me
understand? The workflow, I think, I guess
one thing that you could do is updating the title of the node
itself. That's that would be like the
easiest, easiest go to. So making sure that the name of
the node itself represent an action that it's actually doing.
Let's say if it's deleting your e-mail, you might say delete
(20:36):
e-mail or update e-mail. And one thing to keep in mind is
that every single node takes a specific action.
So you can have different nodes for those different action like
update, delete, create. So it's really if you when
you're looking at your NNN boardand you see your logo, if you
don't, if you don't specify the action, it can become very
confusing really quickly. You just see Gmail, Gmail,
(20:58):
Gmail, Gmail have no idea if it's reading, updating, or
creating. So I would say combining the the
action with the name of the nodeare other ways to make it to add
clarity to your workflow. I've recently heard NNN allows
you to do natural language prompt to like like to a
workflow, but sometimes it needssome tweaking like it just gets
you started. Have you played around with that
yet? I was going to say to give you
(21:19):
some background, the way I got into NNN was really
conventional. So I wanted to Start learning
how to use NNN and I and I wanted a quick way to do it.
And I realized that they weren'ta tool that where I could just
write a prompt to use it. So I decided to build that tool
and on the process of building that tool, this is how I got to
learn about how NNNN works. And so from doing that, I
(21:41):
started building more things andthen I started doing some small
contract to better understand how it works.
And it kind of went from that and I went from building
automation to building my own tool, kind of how I got into it.
That's a cool. For me, the best way to learn is
just by doing. But if you do something even
meta like that that that's awesome.
One thing that I've been, it's almost like the tipping point
that made me really interested in NADN was the ability to have
(22:02):
nodes that run custom code. So could you walk through that
at all? Like is there limitations to
what type of code? Could I have a full app running
in there? Or is it just like scripts?
Like how does that process work?Yeah, so you needed mostly
scripts. So you can build like really
like powerful script. And again you have to keep in
mind input and output. So you can write a script and
and and say what is going to output.
(22:23):
You can use for example a scriptin JavaScript and and detect
what that script is going to returns.
So it's also a very powerful wayto use which actually a lot of
developer like because it give you a lot of flexibility.
From my personal experience, oneof the thing that frustrates me
when doing that is again, like the console and the log, I
(22:43):
sometimes feel that I'm not getting enough like feedback or
response. So one good thing is you might
have to write your script somewhere else 1st and then
bring it inside of NNN. That's something I'd like to do.
But yeah, it's a really cool wayto add more customization to
your workflow. Have you done it personally?
Like do you have scripts runningin any of your workflows?
(23:04):
Only one, but it's technically Ihad one in.
Sorry. Only one type of script.
So technically I have one with like for example, if I want to
match manipulate data, I'll transform it in a certain way,
then I will use a simple JavaScript to do that.
I don't use it as often, so it'susually more if I want to match,
manipulate the data in a specific way, let's say rejects,
(23:27):
which is like detecting words ina sentence.
If I'm not using AI, if I want to save money on token.
So I will use it just for that, for like text manipulation.
But there's a there's a more complex way to use that that I
personally don't have haven't done before, but those are
usually my my my most good to use case when I'm building
workflow. Nice.
And you had mentioned you see some things on social media,
(23:48):
even though admittedly you're a lot of it self-taught.
Do you have go to resources or certain like influencers or
people that you would listen to to learn more about N8N or or
what's your learning? Process.
There's a there's a couple. There's this guy thinks name is
Nate Hurst. And then I think it's Nate, Nate
hurt. And then he has like a really
good channel. Yeah.
So has a school community where it's called automation society
(24:12):
where he actually like showed different workflow.
So I've learned a lot from watching his videos and how it
works because I saw John Califf that has some good videos and
also the NNN official documentation itself has a lot
of very viable information that you can learn from.
The other thing is in the open source community on GitHub, even
if you're not a developer, if you just search NAD N community
(24:34):
nodes, you're going to find a lot of nodes that that will be
super helpful because it's categorized in terms of what
what those nodes do, like productivity or e-mail, webhook,
SSP and so on. So these are like really good
resources that I will use. There's also library on GitHub
of people that have built full on workflow and it gives you
(24:57):
access to the entire workflow. And because NAN is actually what
he returns is a in a format called Jason.
So the entire node that you've seen, NAN, you can do command A
and copy in the entire node and paste it in your text file and
it will return it. Jason, because of that, it's
really easy for people to put a lot of nodes inside of GitHub.
(25:17):
So there's a lot of GitHub application that has like a long
list of different nodes that youcan use that I already pre made.
And you just got me thinking about the idea of putting into
CICD so it can actually run someactions when I'm doing a push.
So super exciting actually. Sorry.
Before I forget, you mentioned Jason.
Is that configurable? Can I do markdown as well as an
output? You can do different output.
You can do like text output and decide to return markdown if you
(25:39):
wanted to. Actually Markdown is really
common format that to use because a lot of MLM, I
understand Markdown and XML really well.
So and just and Jason, especially if you're doing image
generation. So sometimes those are like the
the output that you you could gowith.
Yeah, because I I found Markdownhas just been phenomenal for
interact with AII use Obsidian for my my vault locally, for my
(26:01):
notes locally and I still have cloud code ring there generating
dynamic dashboards throughout the day for different things.
It's it's been really cool. And I just think the more
standardized we can get it, especially I think with Markdown
with very little decoration to text, yes, it's just been great.
There's actually a really good team with Obsidian that people
don't utilize a lot. So you know, like when you
generate an output of NLMS, likewith ChatGPT for example, if you
(26:22):
just copy it and paste it somewhere else, you're not going
to get the markdown format. But with Obsidian, if you copy
it and put it in Obsidian, it would return you the markdown
format. So often time if you're coding
and you need to you don't like you want to keep that mark down,
it's extremely useful. So I would say like if you have
the Obsidian plug in and this isnot sponsor obviously, but I
(26:44):
think it's pretty, it's pretty useful.
That's so cool. And actually just since we're
talking about Obsidian slight pivot, I built an Obsidian plug
in generator. We're able to have a quick chat
with an LLM and it'll actually output an Obsidian plug in to do
your functionality. So like I have it set up as my
CRM. It does all my meeting notes, it
does all my episode stuff, and Ijust feel like NAN might be able
to plug into that as well and just really make the system a
(27:05):
lot more powerful. Really cool.
Yeah. But I mean, that's the Sky's the
limit right now. I will I'll say like if you
can't think of the workflow you can can build it.
One note on that, for branching SO1 input, is there a limit to
how many like routes it can take?
Or do I need to just make new workflows if it's going to be a
lot of branching? I I haven't, I've never reached
a limit, but you can do quite a lot of branching.
(27:27):
You can do you have if node you have one node, you can do like
so many different branching and I've never personally reached a
limit on. I tried to avoid building those
complex like crazy node that yousee.
So I always like tried to be very parse with my logic.
I push any end the way I would have pushed an application.
So I tried to keep it to like two or three logics when I'm
(27:50):
branching out so I'm not one. I haven't reached a limit yet so
I wouldn't know if this leave it.
So do you do you have different workflows with the same trigger
then? Like receive e-mail?
Will that set off different workflows?
Yes, so So what I do is in the file I might have different
triggers for different things. So I might so receive e-mail,
then if something else happened I can have another web to do
(28:11):
something else. So that one way to approach is
having multiple trigger on the file, but different approach.
Then you can group them and then, you know, put like sticky
note to say this one does X, this one does this.
So that's one way to do it. The other thing is doing a long
chain where you, you can put thedomain in the loop.
Usually the, the way I do it, I,I always focus on one very
(28:32):
specific workflow that I want toget done.
And I, I keep it to that. I don't make, I don't
overcomplicate it because you still have to think about
maintainability. Like if you make it too complex,
that is really hard to maintain.And often time it's for quick
use cases that will make it easyto do a specific workflow.
Then I'll be able to transfer toa client and I want them to be
(28:53):
able to understand it. So, so, so keeping it like
making sure it's focused on the on it's own task, but not over
complicated. I think it's it's an important
thing. Also, just on the note of the
branching limit or or lack thereof, do you know are there
any rate limits? Like could I have it pull
something multiple times a second?
Or are there like request limitswhere I can only say like, you
(29:14):
know, once a second? Like, do you know anything about
that? No, I don't know about the rate
limits of of and then I've only.So technically instead of my app
8 nodes, there's you can connectan end to to 8 nodes.
So there are rate limits there in terms of how many records I'm
making. But I haven't, I'm not familiar
with the rate limit of like whenyou're making like requests and
(29:36):
stuff like that. I think it also it'd be
dependent on the nodes and the integration because different
integration have different limits.
So, so I think it must like it will be two layers, so the N81
and then the actual integration that is connected to it.
One thing that I'm very interested in, I'm actually
surprised it took this long to come up, is the idea of self
hosting. N8 N have you have you compared
(29:58):
and contrast the cloud version versus self hosted?
Or is it is a feature parody? Like what do people need to keep
in mind if they're considering self hosting?
Technically the cloud version has has more features.
I believe that in the cloud version, I don't remember the
top of my head, but there's I think there's credential that
was one of it that you can use. You cannot use a credential
(30:19):
which make it super easy to actually when for example,
you're using let's say open AI and you want to make sure that
it goes really easy, you can do that.
I often use primarily the cloud version where I can host myself.
Never had the need to host it but technically 11 quick way to
do it locally. I use ngrok where I can just
(30:41):
like host it temporarily to testmy my endpoints for clients.
I've recommended to you hosted on personal or you know, neglify
into like an instance or an app or something like that.
If someone is dealing with sensitive data as an example, is
the cloud version safe or shouldthey just self host it?
Yes. I mean if you don't want people
are like an end to be aware of you that I would say probably
(31:01):
you want to host it yourself. I think that's one of the use
case for a lot of companies thatdon't want to share the data
with third party or an end and they will decide to host them
themselves. Yes, the cloud is really good
for general use. If you want to have quickly
access to it and not having to deploy it and being able to
create a workflow, make it readyand available right away so that
you can use it right away. I think the cloud version might
(31:23):
be more technical because you still need to run it locally,
deploy it and style, run a comment and study locally and
all that, which can be very intimidating to a lot of people.
That one I'm not familiar with coding and 2nd not familiar with
GitHub and you know. One thing that I'm just curious
about for for you is as a builder, as a developer, someone
who actually makes his work flows.
Do you have any recommendations for APIs that you found useful
(31:46):
or been helpful in your in your work flows?
Me like my go to API for cash isRedis is is is upstash.
I find upstash is like so easy to and quick to get set up with
release database. So that's always been my go to,
Something I use it for for more than just storing cache for
(32:06):
example I can store a lot of information on it.
So I would say that's my go to. Superbase is good for databases
for vector dynamics scape me right now, but you could use
probably as well. I think that would be.
That's about it. That would be like my main go to
that I usually use for APIs. I'm just curious for AI LLM
(32:28):
model, do you have a go to or doyou like Depending on the use
case we use a different one I. Think it's the latter depending
on the use case. My go to is usually open AI.
I usually go with it, but then depending if I want to do
sharing images or sharing video I could do Gemini.
Nanobanai is pretty good right now for manipulating images or I
(32:49):
can do VO3 for video. I've used Graph in the past.
I find that X in general can be very strict with the APIs.
So, so from, from like history, so like I tend to go with like
the other like cloud, you know, open AI and so on.
And. Similar to the API's, do you use
or recommend any external tools that help with Naden?
(33:11):
Yeah, so I will have to mention my own tool which is 8 nodes
because it allow you to essentially run the prompt and
generate A workflow. So what it can do is that you
can upload an image of another workflow and it can literally
copy that exact workflow and recreate the exact same nodes.
Or you can just describe exactlywhat you want to do and then
it'll do all the thinking and planning for you and I'll
(33:32):
generate a work for you that youcan get started with.
And once you have that, you can just copy that Jason and putting
that inside of NAN and start using NAN there.
And often time it's already ready.
You have all the connection, theparameter, the notes are good to
go. So that should be the one that I
used and would recommend. I know NAN is working toward
(33:55):
building something similar, but the thing that eight node
focuses on is really being able to analyze images and generate
workflow from those images really well.
I actually didn't know that you could do a Jason option.
So there's that classic drag anddrop interface, but there's also
the ability to write or create Jason to set up these workflows.
Yeah. So, so the way those workflow
(34:15):
work is that they are actually Jason behind the scene.
So even if you look at NNN and you copy a workflow, you you'll
be a Jason once you paste that into a text file.
So because of that, you're able to use an LLM like ChatGPT
cloud, have it generate the Jason and paste that instead of
NNN and it really works. But oftentimes you run into a
(34:38):
lot of errors because of contextengineering.
They don't have all the information that they need in
order to build those work for properly.
So what I did with eight node isI spend a lot of time doing a
lot of context engineering to give all the context the AI need
in order to build nodes properly.
The right connection, the right parameter, the right keys.
And then now when the Jason output comes in, it's actually
(34:59):
good to go and to play some NNN and then you won't get any
issues or or misconnected nodes and stuff like that.
I really hope someone fine tunesa small model on a bunch of
successful N 8 N workflows so wecan just kind of chat with this
thing and just have this spit itout.
But good job with the prompt engineering.
Yeah. Oh, for sure.
Yeah. Especially like you said, with
it being open source, that always draws a community.
(35:20):
So I really hope they come together and make that happen.
I've tried to do it by the way, but it's just the cost of
training your own model was justadding up and I was like, I'm
not going to do that alone, so. Yeah, need to crowd source the
solution. Exactly.
This has been awesome. I feel like I've learned a lot
from you. Could you share a little like
personal insights? Like what work flows do you have
(35:40):
you found have been the most productive for you?
I think the most probably cheapest e-mail is always my go
to whenever I build workflow. That's the one that gives people
the sense of having an AI doing things for them really quickly.
Everyone is familiar with emailsand then you kind of know
already how to break down the steps of your e-mail.
So read my e-mail, classify them, respond to this type of
(36:03):
e-mail. For me, those are always my most
good two use cases for clients. The most common one that I've
seen is social media. They want NATN to write posts
for them, for linking, for Instagram, for posting, for
scheduling. So these tends to be the most
common one I see from clients. And so I would say probably
(36:23):
those two use cases will be likethe most common that I
originally. And are there any nodes that you
wish existed that don't currently exist?
Probably better Stripe payment nodes will be helpful.
I think Stripe has did a great job making it easier for people
to process payment. But I still think that there are
(36:43):
gap in holes that could be improved.
So I will say, for example, likebeing able to somehow like do
payment or manage payment better.
So yeah, I would say it's a bit broad here, but I would say
payment will be one that I wouldprobably want to see more things
done to it. Perfect and last question for
me, someone's just getting started.
(37:04):
What should they do to learn NA then what do you recommend as
resources or a strategy or road map?
How can someone get up to speed?So I'm a big believer in
learning by doing. So I would say if you want to
learn NNN I would say install itand try to build your first
workflow. Start by building work for
around problem that you have. If you're not sure, instead of
(37:24):
emails, it's a good one to get started with and try to think in
flow rather than in tasks. So if you have a task, try to
break it down and figure out what are the different flow and
steps of that task and try to figure out how it goes from
point A to point B again by thinking about the Legos, that
whole you're just moving it fromone point to the other.
(37:46):
So I'll say that's how we'll probably get started if I would
just start it again. That's amazing, Richardson.
This was a blast. How can people keep up with it?
Where can they follow you? So I'm at Richardson DX pretty
much everywhere on X. Also I have you can find me on
my website at rich.com.com whereI have all my resources, links,
(38:06):
social media profile and everything.
So yeah, feel free to send me a message or reach out to me on
Twitter. Correct.
Awesome. Well.
Thank you for coming on teach just about NNN.
Thanks. Thanks, Mike.
Thank you for listening to the N8N master class with Richardson
to COM. I can't believe it took me this
long to actually try N 8. N Cursor and Cloud Code have
been amazing at rapidly spinningup little scripts and
(38:28):
automations, but to actually develop the integrations with
the services I use on a day-to-day basis, that always
takes a little extra time. There's always extra debugging,
so the idea of a very clean interface that I can just easily
spin up a prototype and see how it goes before I pull it into my
system sounds very appealing. I'm curious, have you used N8N?
If not, has this episode one youover?
And if you have, what workflows do you like on the tool use
(38:50):
Discord server? We're sharing some and workflows
with each other. So by all means, come and join
us and share your insight, learnfrom the community.
And I really want to give a shadow to Richardson for coming
on this episode and teaching us from his experience.
Try out his tool, check out his YouTube channel.
He's a really good guy. And I'll see you next week.
After three consecutive guests, I've talked about the value that
they got from N8 NI hit my breaking point.
I just had to try it. But The thing is I didn't want
to go in blind. I wanted to find the best person
I could to teach me about N8 N before I actually used it.
And I'm going to bring along forthe journey.
So in episode 6 of our tool use,I'm joined by Richardson to
come. He's a founder and an AI builder
(00:20):
and he's hosted and it had master classes before so today
he's going to cover why every developer should consider using
N8 N. Some great use cases to really
save you a lot of time and some pitfalls and gotchas that you
need to look out for. So NNN is essentially like an
open source workflow automation builder.
Its advantage is that you cannotcreate like workflow and it was
designed with developer in mind.So like you can actually see
(00:44):
just an output, you can even edit the response, you can add
code to it. So it's pretty intuitive.
So not just developer can use it, but it really give people
that know how to code the ability to do even more complex
stuff. And on top of that, because it's
open source, it has a pretty bigcommunity and you can build
custom node. So if you take something like
(01:06):
ZAP here, for example, if, if there's a feature that you want
to have on it that is not there yet, you will have to wait for
ZAP here to build it. But on something like N8 N, the
community can build it or you can even hire a developer to
build custom node for you. So that's flexibility is what
makes it official. That's so cool.
And I mean, we were talking before.
(01:26):
I've been a software developer for 10 years, big and open
source too. But there's something about this
drag and drop interface that I just didn't really like.
I like the idea to code everything, control everything,
transparency all the way down. But more or more developers are
saying the same thing where there's like it abstracts away
the integration work. A lot of the the drudgery with
this job of trying to make systems talk to each other.
(01:46):
So who? Who else gets to benefit from
this? So anyone that has like, like a
work for within the company thatthey would like to automate.
So you take a task and you thinkabout how you can make that task
into a workflow that is repeatable, then you can really
like take advantage of that And it's powerful enough that you
can create enough like systems to replace an entire department,
(02:10):
if any new company. So some people have built a
really complex workflow that take care of the wall of the
wall thing. Especially when you combine that
with AI that can be the decisionmaker, then you can it give you
a lot of power to do like reallycomplex automation.
What are some common or popular workflows that people will
automate on N8 N? So like a common one will be
automating emails management. So like they say even for
(02:32):
refund, if you're getting refunding you instead of hiring
someone that is going to go through the e-mail, parse them,
label them, you can use a workflow to do that.
So what you do, you first think in steps like what happened in
every single step and you're thinking input output.
So what's coming in and what's coming out?
So then you can take something like an e-mail.
(02:53):
You can say read my e-mail, findany e-mail that mention refund,
label them refund and then writea response to to to the customer
based on this set of criteria. So emails, e-mail management is
a big one. Customer support is a really big
one as well. Social media management is
really big. It's getting bigger and bigger
(03:13):
because you can automate postinginto multiple across multiple
social network, especially when you have like, you know, people
making content on YouTube and they want that same content to
appear on TikTok and you know, all the other platforms.
So you can use N8 N to actually automate that entire social
media posting. I definitely need to start
looking into this more because I'm manually posting a lot of
(03:35):
cross-platform stuff, but let's let's hone in on just the e-mail
example. Just try to give people a little
more inspiration as to what theycan build.
I hesitate having AI write emails because I don't know what
the quality is like, is it goingto match my tone?
So can we get through a little granular?
How do we get human in the loop?How do we kind of set the AI for
tone? Does it have to be a certain
LLM? Like how?
(03:56):
How do we get that workflow moremore polished?
Yes. So the feeling you just
mentioned human in the loop is abig one.
You can have a human verifying astep before before it's
submitted. So you can actually be part of
that workflow as human in the loop.
The other thing you can do is for having templates.
So for for a lot of work operation workflow, often time
(04:20):
you can just, you know, simplifythem into templates because it's
often times similar responses. So you can have pre written
template that you just use instead of of the the tool and
just add or update the variable that you want to edit inside of
that template. And that's the way that you give
you the consistency of what to expect, which is good for brands
and tone because you don't want the language of your brain to
(04:43):
change based on every single response, right?
So I think template is like a really good good tool.
Nice. And for for the aspect of
templates, I actually had Lee Russell on a couple episodes ago
and we were talking about the idea of when you just customize
the the user's name, Hi, first name it.
It's kind of expected. Now, it's not really that
magical experience. Can you have it where with along
(05:03):
with this template, you actuallywill have something more custom.
So let's say like I first name and then a paragraph related to
their specific instance and thenthe generic template.
Does that work? Yes, this is this is where stuff
that work for the use AI comes in into place.
Because now you can take a take a portion of the input being,
for example, the the job of the person, the position or
(05:26):
something that they've accomplished and have AI
customize that, that intro section specifically for for the
person that's going to receive the e-mail.
So you can actually make it morecustom by having AI deciding how
to rewrite those different parts.
OK, cool. And for the AI, do you, do you
have to use open AI kids you have different, is it just like
(05:46):
an API call? How do you actually plug an AI
into the system? So it's pretty flexible.
So you have it give you access to different type of AI models
that you can use to can use anywhere from Grog to Entropic,
Google, Jimini, they have so many now.
So you can use any of those those tool in it.
And it gives you that ability and flexibility to choose what
AI model you want to use. And the way you do it is you,
(06:08):
you, you go on Entropic, for example, which is the one that
build cloud. And then you take your API key
and you add it inside of your NNN credentials.
And then you can now use that model specifically, specifically
for one node or different node that you're adding to it.
So so yeah, you can bring, bringin different AI models as, as
based on on your need. You're not limited to a single
(06:30):
one. And you can even use custom one
if you use something like Olama which is something that a lot of
developer like to use or open routers and stuff like that.
And and sorry, how would that work?
Set a custom API URL and just point it towards my server and
then just like regular request response.
Exactly and or like if you're using open router then you have
even more flexibility because they have such a large database
(06:51):
of models so you can pick different model that you want to
use. And kind of leading back to to
human in the loop. I think it's super important for
you to keep people informed as to what they're doing and
insight as to what's happening. Observability in general is a
concern when using systems like this because if a pipeline is
just running, it's in the background, maybe there's a step
along the way, causes an error and by the time it reaches human
in the loop, it's already cascaded.
(07:13):
Do you have any advice or strategies on how if someone
uses anything, they can get better observability, logging,
tracing into the process? Yeah, that's actually a really
good plan because error and logging is a thing that a lot of
people that begin to use and always forget to do.
So you can actually like making sure to have no to just handle
errors and how to deal with errors and where you want this
(07:33):
error to go to. So one way to handle that is if
you're put a human in the loop, you can have some of the error
going to the human in the loop to review them, or you can have
a sender log via e-mail or to toa platform.
So me personally, for some of mytool, I like to build my own
dashboard. So I will get these on my own
(07:54):
dashboard that I can then, you know, review and you can also
send it to Slack, which is very common for a lot of companies.
So they have all those error directly coming on Slack and
they know how to take action whenever those come in.
Is it possible I just use Slack as an example?
Error comes in, human acknowledges it, they know the
solution is it. Are you able to automatically
(08:15):
have triggers from outside of NAN send information back in?
So say I say like close ticket or respond with with whatever
response. Can that be done in Slack and
send back to NAN? You you have this thing called
webhook that essentially are like listeners that can listen
to even outside of NAN. And so that will be fit into the
category of triggers. So NAN has like different type
(08:36):
of triggers that you can have like time bays that react to
responses and so on. And so for example, if you, you
wanted to respond to event happening on Slack, it could be
for example, when I say yes, send an event to that webhook
and take action based on that. So it would be, it would work
usually as if it's a trigger as part of a workflow.
(08:57):
I have, like they said, the begin of the workflow.
This is what happened. I haven't seen a lot of people
using it as it being in the middle of a workflow itself.
So usually it's more done as a trigger.
OK, so custom triggers make a lot of sense.
Love the idea of webhooks. It just allows you kind of
percolate into any system no matter how custom.
Can you kind of do similar thing?
Webhooks on the other side whereNAN is able to notify a webhook
(09:18):
when it's completes a task or stuck in a task.
So that's a very common use for notifications.
Also, for example, people that want to use stuff like Telegram
or WhatsApp, they can send a webhook to some of those
platform to get to submit something.
I thought even with type form. Type form is a very common use
with web hook as well. You can receive the web hook
(09:39):
there as well. And if you want to send it, you
can use it to for example, any time a user take an action, you
want them for example, to to getaccess to a form.
You can also send, send an e-mail.
You know there's so many variables that you can do with
that in and out. Nice.
And I kind of assuming it's yes,I just want to make sure.
Can it connect to databases? Like can I have it actually
(09:59):
write entries to the table? Yes.
So a lot of people connect even different type of databases.
So yes, the answer is yes. Some people like to use Excel to
keep it simple. Some people use like super base.
You can also use RADIUS like a cache database vector is a
really big use case if you're using AI because in NAA there's
(10:21):
this thing called clusters that are often built for reusability.
So they will have the cluster connected to a vector database
if you need to store some information and process for
making it easier for the AI to have memory essentially.
Sir, could you go into detail what is a cluster?
(10:43):
So a cluster is kind of like a group that make an action
reasonable. Instead of NADN there's
different type of cluster. You can have a chat cluster that
essentially is composed of an AImodel how how the AI model is
going to get the message, a database which is where the
memory is going to be stored. So instead of having to build
(11:05):
every single different componentyourself, you have that cluster
that everything already pre built and ready to use.
And I just make it a lot faster to build more complex nodes.
So that's something that a lot of pro tend to use it because it
makes things a lot faster than having to set up everything from
scratch. Yeah.
So instead of regular atomic units, you have this this Group
A cluster to to plug and play. Does the community develop
(11:26):
clusters like is there? Is there like a library I can go
to be like oh this matches my use case insert?
So that's what Jenna, I'm not sure, but usually when I see
when I see community contribution is usually notes.
I've personally used a lot of notes from the community.
I haven't seen a cluster. I haven't I haven't need to
discuss that. I'm not 100% sure.
So this sounded really exciting.I love the idea of being able to
(11:48):
just automate a lot of workflows, but I imagine as
people start exploring this technology for the first time,
myself included, there didn't bemistakes.
What type of common mistakes do you see people make?
And when they make these mistakes, what's the debugging
process like? I would say like one of the most
common mistake is approaching NNN as like just like making
tasks. So finger of tasks instead of
finger of flow, because it's very important to think about
(12:12):
the information that comes in. So one way, one way I like to
put it to visualize it is it's kind of like they like kind of
like Legos legal with data, right?
So every time you create one legal to the other one, you
passing the data to the other Lego.
And then and then as you're doing that, you cannot be
building this, this supply chainor, you know, or if I put it in
(12:33):
the sense of like four assembly line, right?
So every assembly line is getting different data.
So instead of making, let's say,your car, instead of making
object, you, you're taking the data and then put passing into
the next thing. So a lot of people tend to think
of and and as just that, but it's really important to think
about it as an assembly line when you're passing that out
from one end to the other, because this is what allow you
(12:56):
to actually transform the information and the data and
then get to an output. So also that's like probably one
of the most common one. And in terms of the the pipeline
aspect, the assembly aspect, is there a lot of requirements for
data transformation? If I have data coming in a
certain format, do I have to go in between every node and map
fields to each other? Or is there any type of
automation there? Yes you do you.
(13:16):
You have to map field sometime. What you do is you can.
So, so you come in the format called Jason, and then you can
kind of like decide, say that for example, this node is going
to output this different field. And on the next node, you can
grab the value from from the input or the value from the
response and pass and pass that to where you're manipulating the
(13:39):
data. So for example, if you want to
add someone's input from a PROM inside of instead of to an AI
model, you will take that input and and and and pass it to the
AI model. When the AI model generate a
response, you will use that response and pass it to the next
node and so on. So you can transfer the data
that way with it. Nice.
And just for that process, do I need to know from either
(14:02):
previous experience or other experiments what the data fields
are? Or will any then display?
This? Is the output this is the
required input? Like match things.
So there's two things. One, you can decide how you want
the data to be outputted, which is very important to have the
file say you need to think aboutinput and output because you
need to know how you want your data to be presented for the
next node. And and once once you've created
(14:25):
this, this, I guess this preliminary, then the next node
would know what that I can it can it can take in and actually
this is a tension, but Asian builder by open a actually do
that really well. The way they allow you to
visualize what comes in, what comes out.
I think they did a really good job of that.
Is there 'cause you mentioned open source so people will build
(14:45):
their own nodes I guess? Is there any work done on the
interface? Like do you think that any then
will catch up to open AI's agentbuilder in terms of UI?
That that it doesn't seem that that's the priority right now,
because I was reading something that the CEO of and, and what I
think his name is Jen open, openiser like I have lost his name
(15:08):
wrong. But technically what what he
said is that their vision is to build like the excel of AI.
So they really want to build something that people can use at
the orchestration layer. So like they want to really
focus on that location layer, while open AI is really focused
on building good product that are very easy to use.
The other thing that they're going to do is they're going to
(15:29):
add a way to generate workflow and that's going to be purely
for marketing purposes to in order to get more people to use
N8 N, but not they're not reallygoing to dive into that
generation, but mostly to if someone need to quickly have a
word from to it instead of N8 N,they'll use a prompt and write
what they want to build and they'll pre set up the
different, the different notes for them.
(15:51):
So that's their current focus right now.
I don't. And also because they target
market is is usually more technical, while Asian builder
is targeting the average consumer, the average like your
customer and company. So like I don't think that
they'll try to do something similar to open the eye.
Pivoting slightly, you mentionedemails, a good use case writing
(16:11):
to accelerate database. Could you go over a few other
like integrations nodes, maybe some that aren't so popular that
you found it but actually been very helpful?
Like what? What type of services should
people keep in mind when they'resetting up their workflows?
So something that I people don'treally think about this like RSS
feed, you know, a lot of people,for example, listen to podcasts.
There's a lot of ways that you can utilize RSS feed or for news
(16:35):
or for podcasts. I think that's like a really
interesting use case. If you want to have like an
agents that check your, your preferred news source all the
time or preferred podcast. And if you want to only, you
know, check the pockets you're actually interested in, you can
have filters. So that's like a really like
creative way to make it into your little, you know, a robot I
can like prepare the news for you and so on.
(16:56):
The other one will be I mean foam trigger is a common use
case as well, but people don't really see that they can use
that for for more than just forms.
You can do that as like a payment form as well, because
you can have a form to take on payment and you can have a
webhook to respond to the Stripewebhook request.
So you can actually build entirepayment system just from NNN.
(17:19):
Often time like as as a developer, the thing that bring
me to use NNNN is that if I think about a workflow that I
could build into a website and Iwant to quickly see how that
would function. I will quickly build a prototype
on NAN and and see it working and then allow me to quickly
test something and having an endpoint for someone else to
(17:39):
utilize. So I could for example give that
end point to my app if I wanted to do something complicated.
I don't want to build full edge yet, but just create a quick
note to see how it would work. You're not the first person to
bring up the idea of using it for prototyping.
Do you find you'll hit a ceilingor a certain level of
functionality where any end's nolonger appropriate, and then you
take the concept and put it all into code?
(18:00):
Or have you? Is there even a ceiling there?
There's a ceiling. So, so sometime if your task,
for example, involve like using browser AI agents or doing
scraping or stuff that have to manipulate the browsers and
stuff like that, I think it's a lot easier to build it in code
because you mentioned observability early on and
that's like a really big factor.And I think that N 8 N sometime
(18:23):
lack of visibility because it doesn't shows you the the error
right away. So it's sometime easier to
actually build that inside of like, you know, anyone if I'm in
cursor or you name it, because then you'll be able to, to
really see the error output and a debugging output.
So if something get too complex where being being able to see
the debugging log is important, I think I would say stay stay
(18:46):
away from N8N if you can. Just like programmatically set
up things and where the input and output are very clear that
it's an really easy tool to use to build quick prototypes.
OK, cool. And and that that's good to
know. And you had mentioned that there
are some very complex workflows that exist.
Do you have any advice on how people architect or lay out
their NA then set up to help with maintain ability where
(19:09):
instead of just like a random cluster of nodes that can be
overwhelming not knowing where what where what is any way of
like organizing or structuring anything like that?
Yes, so you can think about, forexample, there's, there's,
there's stickers note that you can add so where you can explain
what different area or part of the workflow actually doing.
(19:29):
Those are very common. It makes it also easier for you
yourself when you look at your work for after a while, because
sometime you may forget, you know, what is this work for
doing again? Now on social media, I've seen a
lot of of like images that tend to go viral of people creating
these massive and workflows thatare very, very complicated.
And I often see NNN commenting on it and say, have you ever
(19:51):
heard of cluster or have you ever heard of stub nodes?
You know, so there's a way to simplify it by looking at
cluster nodes and looking at Subs node.
Sometimes Subs node can allow you to do some of those stuff
with a cluster. Are there any other like meta
nodes that don't necessarily help with functionality but help
with the developer experience? Do you call them stickers where
the the node that kind of saves?Is there anything like that
(20:13):
which doesn't impact the flow ofthe data but helps me
understand? The workflow, I think, I guess
one thing that you could do is updating the title of the node
itself. That's that would be like the
easiest, easiest go to. So making sure that the name of
the node itself represent an action that it's actually doing.
Let's say if it's deleting your e-mail, you might say delete
(20:36):
e-mail or update e-mail. And one thing to keep in mind is
that every single node takes a specific action.
So you can have different nodes for those different action like
update, delete, create. So it's really if you when
you're looking at your NNN boardand you see your logo, if you
don't, if you don't specify the action, it can become very
confusing really quickly. You just see Gmail, Gmail,
(20:58):
Gmail, Gmail have no idea if it's reading, updating, or
creating. So I would say combining the the
action with the name of the nodeare other ways to make it to add
clarity to your workflow. I've recently heard NNN allows
you to do natural language prompt to like like to a
workflow, but sometimes it needssome tweaking like it just gets
you started. Have you played around with that
yet? I was going to say to give you
(21:19):
some background, the way I got into NNN was really
conventional. So I wanted to Start learning
how to use NNN and I and I wanted a quick way to do it.
And I realized that they weren'ta tool that where I could just
write a prompt to use it. So I decided to build that tool
and on the process of building that tool, this is how I got to
learn about how NNNN works. And so from doing that, I
(21:41):
started building more things andthen I started doing some small
contract to better understand how it works.
And it kind of went from that and I went from building
automation to building my own tool, kind of how I got into it.
That's a cool. For me, the best way to learn is
just by doing. But if you do something even
meta like that that that's awesome.
One thing that I've been, it's almost like the tipping point
that made me really interested in NADN was the ability to have
(22:02):
nodes that run custom code. So could you walk through that
at all? Like is there limitations to
what type of code? Could I have a full app running
in there? Or is it just like scripts?
Like how does that process work?Yeah, so you needed mostly
scripts. So you can build like really
like powerful script. And again you have to keep in
mind input and output. So you can write a script and
and and say what is going to output.
(22:23):
You can use for example a scriptin JavaScript and and detect
what that script is going to returns.
So it's also a very powerful wayto use which actually a lot of
developer like because it give you a lot of flexibility.
From my personal experience, oneof the thing that frustrates me
when doing that is again, like the console and the log, I
(22:43):
sometimes feel that I'm not getting enough like feedback or
response. So one good thing is you might
have to write your script somewhere else 1st and then
bring it inside of NNN. That's something I'd like to do.
But yeah, it's a really cool wayto add more customization to
your workflow. Have you done it personally?
Like do you have scripts runningin any of your workflows?
(23:04):
Only one, but it's technically Ihad one in.
Sorry. Only one type of script.
So technically I have one with like for example, if I want to
match manipulate data, I'll transform it in a certain way,
then I will use a simple JavaScript to do that.
I don't use it as often, so it'susually more if I want to match,
manipulate the data in a specific way, let's say rejects,
(23:27):
which is like detecting words ina sentence.
If I'm not using AI, if I want to save money on token.
So I will use it just for that, for like text manipulation.
But there's a there's a more complex way to use that that I
personally don't have haven't done before, but those are
usually my my my most good to use case when I'm building
workflow. Nice.
And you had mentioned you see some things on social media,
(23:48):
even though admittedly you're a lot of it self-taught.
Do you have go to resources or certain like influencers or
people that you would listen to to learn more about N8N or or
what's your learning? Process.
There's a there's a couple. There's this guy thinks name is
Nate Hurst. And then I think it's Nate, Nate
hurt. And then he has like a really
good channel. Yeah.
So has a school community where it's called automation society
(24:12):
where he actually like showed different workflow.
So I've learned a lot from watching his videos and how it
works because I saw John Califf that has some good videos and
also the NNN official documentation itself has a lot
of very viable information that you can learn from.
The other thing is in the open source community on GitHub, even
if you're not a developer, if you just search NAD N community
(24:34):
nodes, you're going to find a lot of nodes that that will be
super helpful because it's categorized in terms of what
what those nodes do, like productivity or e-mail, webhook,
SSP and so on. So these are like really good
resources that I will use. There's also library on GitHub
of people that have built full on workflow and it gives you
(24:57):
access to the entire workflow. And because NAN is actually what
he returns is a in a format called Jason.
So the entire node that you've seen, NAN, you can do command A
and copy in the entire node and paste it in your text file and
it will return it. Jason, because of that, it's
really easy for people to put a lot of nodes inside of GitHub.
(25:17):
So there's a lot of GitHub application that has like a long
list of different nodes that youcan use that I already pre made.
And you just got me thinking about the idea of putting into
CICD so it can actually run someactions when I'm doing a push.
So super exciting actually. Sorry.
Before I forget, you mentioned Jason.
Is that configurable? Can I do markdown as well as an
output? You can do different output.
You can do like text output and decide to return markdown if you
(25:39):
wanted to. Actually Markdown is really
common format that to use because a lot of MLM, I
understand Markdown and XML really well.
So and just and Jason, especially if you're doing image
generation. So sometimes those are like the
the output that you you could gowith.
Yeah, because I I found Markdownhas just been phenomenal for
interact with AII use Obsidian for my my vault locally, for my
(26:01):
notes locally and I still have cloud code ring there generating
dynamic dashboards throughout the day for different things.
It's it's been really cool. And I just think the more
standardized we can get it, especially I think with Markdown
with very little decoration to text, yes, it's just been great.
There's actually a really good team with Obsidian that people
don't utilize a lot. So you know, like when you
generate an output of NLMS, likewith ChatGPT for example, if you
(26:22):
just copy it and paste it somewhere else, you're not going
to get the markdown format. But with Obsidian, if you copy
it and put it in Obsidian, it would return you the markdown
format. So often time if you're coding
and you need to you don't like you want to keep that mark down,
it's extremely useful. So I would say like if you have
the Obsidian plug in and this isnot sponsor obviously, but I
(26:44):
think it's pretty, it's pretty useful.
That's so cool. And actually just since we're
talking about Obsidian slight pivot, I built an Obsidian plug
in generator. We're able to have a quick chat
with an LLM and it'll actually output an Obsidian plug in to do
your functionality. So like I have it set up as my
CRM. It does all my meeting notes, it
does all my episode stuff, and Ijust feel like NAN might be able
to plug into that as well and just really make the system a
(27:05):
lot more powerful. Really cool.
Yeah. But I mean, that's the Sky's the
limit right now. I will I'll say like if you
can't think of the workflow you can can build it.
One note on that, for branching SO1 input, is there a limit to
how many like routes it can take?
Or do I need to just make new workflows if it's going to be a
lot of branching? I I haven't, I've never reached
a limit, but you can do quite a lot of branching.
(27:27):
You can do you have if node you have one node, you can do like
so many different branching and I've never personally reached a
limit on. I tried to avoid building those
complex like crazy node that yousee.
So I always like tried to be very parse with my logic.
I push any end the way I would have pushed an application.
So I tried to keep it to like two or three logics when I'm
(27:50):
branching out so I'm not one. I haven't reached a limit yet so
I wouldn't know if this leave it.
So do you do you have different workflows with the same trigger
then? Like receive e-mail?
Will that set off different workflows?
Yes, so So what I do is in the file I might have different
triggers for different things. So I might so receive e-mail,
then if something else happened I can have another web to do
(28:11):
something else. So that one way to approach is
having multiple trigger on the file, but different approach.
Then you can group them and then, you know, put like sticky
note to say this one does X, this one does this.
So that's one way to do it. The other thing is doing a long
chain where you, you can put thedomain in the loop.
Usually the, the way I do it, I,I always focus on one very
(28:32):
specific workflow that I want toget done.
And I, I keep it to that. I don't make, I don't
overcomplicate it because you still have to think about
maintainability. Like if you make it too complex,
that is really hard to maintain.And often time it's for quick
use cases that will make it easyto do a specific workflow.
Then I'll be able to transfer toa client and I want them to be
(28:53):
able to understand it. So, so, so keeping it like
making sure it's focused on the on it's own task, but not over
complicated. I think it's it's an important
thing. Also, just on the note of the
branching limit or or lack thereof, do you know are there
any rate limits? Like could I have it pull
something multiple times a second?
Or are there like request limitswhere I can only say like, you
(29:14):
know, once a second? Like, do you know anything about
that? No, I don't know about the rate
limits of of and then I've only.So technically instead of my app
8 nodes, there's you can connectan end to to 8 nodes.
So there are rate limits there in terms of how many records I'm
making. But I haven't, I'm not familiar
with the rate limit of like whenyou're making like requests and
(29:36):
stuff like that. I think it also it'd be
dependent on the nodes and the integration because different
integration have different limits.
So, so I think it must like it will be two layers, so the N81
and then the actual integration that is connected to it.
One thing that I'm very interested in, I'm actually
surprised it took this long to come up, is the idea of self
hosting. N8 N have you have you compared
(29:58):
and contrast the cloud version versus self hosted?
Or is it is a feature parody? Like what do people need to keep
in mind if they're considering self hosting?
Technically the cloud version has has more features.
I believe that in the cloud version, I don't remember the
top of my head, but there's I think there's credential that
was one of it that you can use. You cannot use a credential
(30:19):
which make it super easy to actually when for example,
you're using let's say open AI and you want to make sure that
it goes really easy, you can do that.
I often use primarily the cloud version where I can host myself.
Never had the need to host it but technically 11 quick way to
do it locally. I use ngrok where I can just
(30:41):
like host it temporarily to testmy my endpoints for clients.
I've recommended to you hosted on personal or you know, neglify
into like an instance or an app or something like that.
If someone is dealing with sensitive data as an example, is
the cloud version safe or shouldthey just self host it?
Yes. I mean if you don't want people
are like an end to be aware of you that I would say probably
(31:01):
you want to host it yourself. I think that's one of the use
case for a lot of companies thatdon't want to share the data
with third party or an end and they will decide to host them
themselves. Yes, the cloud is really good
for general use. If you want to have quickly
access to it and not having to deploy it and being able to
create a workflow, make it readyand available right away so that
you can use it right away. I think the cloud version might
(31:23):
be more technical because you still need to run it locally,
deploy it and style, run a comment and study locally and
all that, which can be very intimidating to a lot of people.
That one I'm not familiar with coding and 2nd not familiar with
GitHub and you know. One thing that I'm just curious
about for for you is as a builder, as a developer, someone
who actually makes his work flows.
Do you have any recommendations for APIs that you found useful
(31:46):
or been helpful in your in your work flows?
Me like my go to API for cash isRedis is is is upstash.
I find upstash is like so easy to and quick to get set up with
release database. So that's always been my go to,
Something I use it for for more than just storing cache for
(32:06):
example I can store a lot of information on it.
So I would say that's my go to. Superbase is good for databases
for vector dynamics scape me right now, but you could use
probably as well. I think that would be.
That's about it. That would be like my main go to
that I usually use for APIs. I'm just curious for AI LLM
(32:28):
model, do you have a go to or doyou like Depending on the use
case we use a different one I. Think it's the latter depending
on the use case. My go to is usually open AI.
I usually go with it, but then depending if I want to do
sharing images or sharing video I could do Gemini.
Nanobanai is pretty good right now for manipulating images or I
(32:49):
can do VO3 for video. I've used Graph in the past.
I find that X in general can be very strict with the APIs.
So, so from, from like history, so like I tend to go with like
the other like cloud, you know, open AI and so on.
And. Similar to the API's, do you use
or recommend any external tools that help with Naden?
(33:11):
Yeah, so I will have to mention my own tool which is 8 nodes
because it allow you to essentially run the prompt and
generate A workflow. So what it can do is that you
can upload an image of another workflow and it can literally
copy that exact workflow and recreate the exact same nodes.
Or you can just describe exactlywhat you want to do and then
it'll do all the thinking and planning for you and I'll
(33:32):
generate a work for you that youcan get started with.
And once you have that, you can just copy that Jason and putting
that inside of NAN and start using NAN there.
And often time it's already ready.
You have all the connection, theparameter, the notes are good to
go. So that should be the one that I
used and would recommend. I know NAN is working toward
(33:55):
building something similar, but the thing that eight node
focuses on is really being able to analyze images and generate
workflow from those images really well.
I actually didn't know that you could do a Jason option.
So there's that classic drag anddrop interface, but there's also
the ability to write or create Jason to set up these workflows.
Yeah. So, so the way those workflow
(34:15):
work is that they are actually Jason behind the scene.
So even if you look at NNN and you copy a workflow, you you'll
be a Jason once you paste that into a text file.
So because of that, you're able to use an LLM like ChatGPT
cloud, have it generate the Jason and paste that instead of
NNN and it really works. But oftentimes you run into a
(34:38):
lot of errors because of contextengineering.
They don't have all the information that they need in
order to build those work for properly.
So what I did with eight node isI spend a lot of time doing a
lot of context engineering to give all the context the AI need
in order to build nodes properly.
The right connection, the right parameter, the right keys.
And then now when the Jason output comes in, it's actually
(34:59):
good to go and to play some NNN and then you won't get any
issues or or misconnected nodes and stuff like that.
I really hope someone fine tunesa small model on a bunch of
successful N 8 N workflows so wecan just kind of chat with this
thing and just have this spit itout.
But good job with the prompt engineering.
Yeah. Oh, for sure.
Yeah. Especially like you said, with
it being open source, that always draws a community.
(35:20):
So I really hope they come together and make that happen.
I've tried to do it by the way, but it's just the cost of
training your own model was justadding up and I was like, I'm
not going to do that alone, so. Yeah, need to crowd source the
solution. Exactly.
This has been awesome. I feel like I've learned a lot
from you. Could you share a little like
personal insights? Like what work flows do you have
(35:40):
you found have been the most productive for you?
I think the most probably cheapest e-mail is always my go
to whenever I build workflow. That's the one that gives people
the sense of having an AI doing things for them really quickly.
Everyone is familiar with emailsand then you kind of know
already how to break down the steps of your e-mail.
So read my e-mail, classify them, respond to this type of
(36:03):
e-mail. For me, those are always my most
good two use cases for clients. The most common one that I've
seen is social media. They want NATN to write posts
for them, for linking, for Instagram, for posting, for
scheduling. So these tends to be the most
common one I see from clients. And so I would say probably
(36:23):
those two use cases will be likethe most common that I
originally. And are there any nodes that you
wish existed that don't currently exist?
Probably better Stripe payment nodes will be helpful.
I think Stripe has did a great job making it easier for people
to process payment. But I still think that there are
(36:43):
gap in holes that could be improved.
So I will say, for example, likebeing able to somehow like do
payment or manage payment better.
So yeah, I would say it's a bit broad here, but I would say
payment will be one that I wouldprobably want to see more things
done to it. Perfect and last question for
me, someone's just getting started.
(37:04):
What should they do to learn NA then what do you recommend as
resources or a strategy or road map?
How can someone get up to speed?So I'm a big believer in
learning by doing. So I would say if you want to
learn NNN I would say install itand try to build your first
workflow. Start by building work for
around problem that you have. If you're not sure, instead of
(37:24):
emails, it's a good one to get started with and try to think in
flow rather than in tasks. So if you have a task, try to
break it down and figure out what are the different flow and
steps of that task and try to figure out how it goes from
point A to point B again by thinking about the Legos, that
whole you're just moving it fromone point to the other.
(37:46):
So I'll say that's how we'll probably get started if I would
just start it again. That's amazing, Richardson.
This was a blast. How can people keep up with it?
Where can they follow you? So I'm at Richardson DX pretty
much everywhere on X. Also I have you can find me on
my website at rich.com.com whereI have all my resources, links,
(38:06):
social media profile and everything.
So yeah, feel free to send me a message or reach out to me on
Twitter. Correct.
Awesome. Well.
Thank you for coming on teach just about NNN.
Thanks. Thanks, Mike.
Thank you for listening to the N8N master class with Richardson
to COM. I can't believe it took me this
long to actually try N 8. N Cursor and Cloud Code have
been amazing at rapidly spinningup little scripts and
(38:28):
automations, but to actually develop the integrations with
the services I use on a day-to-day basis, that always
takes a little extra time. There's always extra debugging,
so the idea of a very clean interface that I can just easily
spin up a prototype and see how it goes before I pull it into my
system sounds very appealing. I'm curious, have you used N8N?
If not, has this episode one youover?
And if you have, what workflows do you like on the tool use
(38:50):
Discord server? We're sharing some and workflows
with each other. So by all means, come and join
us and share your insight, learnfrom the community.
And I really want to give a shadow to Richardson for coming
on this episode and teaching us from his experience.
Try out his tool, check out his YouTube channel.
He's a really good guy. And I'll see you next week.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Paper Ghosts: The Texas Teen Murders
Paper Ghosts: The Texas Teen Murders takes you back to 1983, when two teenagers were found murdered, execution-style, on a quiet Texas hill. What followed was decades of rumors, false leads, and a case that law enforcement could never seem to close. Now, veteran investigative journalist M. William Phelps reopens the file — uncovering new witnesses, hidden evidence, and a shocking web of deaths that may all be connected. Over nine gripping episodes, Paper Ghosts: The Texas Teen Murders unravels a story 42 years in the making… and asks the question: who’s really been hiding the truth?
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com