n8n Livestream: Product Updates, Community Creator & the latest Community Challenge

right away there. Brilliant. Um, let's bring in Nick and David, shall we? Hey, Nick. — Hello. — How you doing? — I'm good. How are you? — I'm good. Yeah, it's been three months, right? You were here in January as well. — Is it three months already again? — Yeah. We're going We're getting into a good rhythm now, man. I like it. — Yeah, I come every three months now. — Let's make that a — Okay. What are you going to be talking about today, Nick? sharing with us? — Um, yeah. Yeah, I think we have uh four super interesting topics. One is the topic of MCP or general building from where you are building Naden workflows. Um and then I bring um two things more or less directly from the engineers to this meeting. Some sneak peeks uh one uh about agents as a first class citizen and the other one about the AI assistant um which I'm going to be talking more about. And then I think we also have David um talking about Net and Connect later and what that um actually is. — All right. [snorts] Should I go to the next one? Yeah. — Yeah. So, let's let's kick ourselves off. Um I think last time we had Sin here on the community live stream and she already talked a little bit about MCP and that we just released the very first version to beta. Um I think since then we actually worked quite a lot on MCP and improved it quite a bit. Um to a point where I would say it's pretty good at the moment at uh building Netn workflows. Um I think many internally at NN are already using it also um to um build the first workflow or even iterate over it. I think we worked on many improvements, bug fixes, new tools to make that more efficient. And what we are also um launching or maybe just launched is a skill as well um to make this MCP work even better. Um it's still currently in preview even though it works quite well. Um but that basically comes with a promise that we're iterating super quickly on this. We're going to be adding much more. Um, we're looking into different things to improve the experience with building from the outside even more. Um, so expect to have many more updates follow here. — And Nick, you said it's in beta and it's in preview mode. What does that mean? What do people need to do to try this out themselves? — Yeah, so basically you can try out the newest version of MCP in your current version of NADN, right? Um, the feature itself is in preview mode, which means it's fairly new. We're still working actively on it, but we think it's already so cool that we didn't want to like keep this closed doors anymore. Wanted to bring it out. — Um, so you can just update to the newest version. Um, beta or stable both have kind of the MCP, but of course, as we iterate on this uh fairly quickly, every new version comes with updates for the MCP, making it even better every time. All right, great. And then just for people who are not super familiar, once you have updated to the latest version, you have the MCP. So, do you need to first activate it in your instance as well or — um so yeah, you need to go to the settings. There's an MCP settings. So, by default, MCP can't just access all your workflows for security reasons, right? Um so, you can kind of uh go to your instance settings. Uh check out MCP there. Um there you also find some uh some instructions on how to connect MCP to your coding agents or to cloud or whatever you want to use. — Yeah. — To then get going with it. — Okay. So then basically after that you go into your AI tool for example cloud. You could connect it to your local uh to your uh instance URL and off you go. — Correct. There's one small caveat. We have this uh built-in connector at uh cloud at the moment. For some reason, Anthropic is quite slow with updating our MCP. So that still doesn't have these new tools um that allow you to build uh but we are discussing uh as we speak with them on how to get this updated as quickly as possible. — And Nick, how is that new skill feature going to help? Like what does it do for you if you use that? — Yeah. So basically skills in general are there to help LLMs perform certain tasks better, right? And of course as we um kind of use MCP include or your other coding agents they don't know much about NN from the get-go right so what MCP basically does is that it gives tools uh to these coding agents to build NN workflows a certain way and these uh skills or this skill kind of optimizes on some weaknesses that we still have with these LLMs on some of the tool usage right so for example example, it in the past preferred using a code node over a set node when trying to do some data structure and so forth. So the skill is just optimizing on some things um that maybe coding agents otherwise are slightly confused about. — And so we are going to be providing a skill that makes LLM better at creating workflows. — Exactly. In combination with MCP, — right, cool. Do you hear that? just kids on the streets put — fireworks. I don't know. — All right. Um All right. Brilliant. I'm going to try this out myself today as well. Thanks. — Yeah, you should. Uh it's working pretty well. All right. What's — next one? — Yeah, let's go. Let's go to the next one. um which is something uh yeah from the forge I guess uh working on something there which is like we basically refer to it as agents as first class citizens. So how agents right now work in NN and I'm sure most of you are aware is that they are a note inside a workflow right we believe with how important agents have become but also are becoming in the future that they deserve to be a first class citizen next to workflows. Of course, you can still use them in workflows. Um, but they are big enough to be kind of its own entity next to workflow, right? And with that, they don't just get a new place in the UI, but we also optimize on what agents actually mean, right? So, they are designed to be much more powerful agents. um like longunning memory, state management, uh contraction stat strategies, routing, um many of the things that you expect from these like top tier agents basically. And I think I've also been here in the past where we kind of spoke about how how's our AI workflow builder built and I think I mentioned back then, oh yeah, unfortunately it's not built with NLN. — I think this is kind of the level where these new agents should come to, right? like they should empower you to build super complex agents such as an AI workflow build inside NN and there's also going to be some other things right like the the experience is probably going to be an AI first creation experience so that you kind of chat with the agent to set itself up uh in a way um because we of course also know that most people prefer to build things with AI today um so uh yeah pretty excited about this one uh we're probably going to test it internally next week. Um, [clears throat] so even the NN team hasn't fully seen it yet. So that's going to be really exciting. — Yeah, I'm really curious because this is going to deviate a bit from the standard workflow approach, right? This is a different paradigm of building agents for us. So — correct. It's a little bit different but still kind of the same, right? Because you can also build agents right now. But right now it's a node later. That's more than just a node potentially. It also comes with a slightly different execution lock. Um because one thing that we have seen with nodes especially on a chatbased agent is that our execution logs are not optimal for that because every chat message is a new execution right you can't really see the whole chat session so that is probably also going to be different uh in these new agents but yeah lots of things work in progress here. Um also the screenshot that you can see here is like one of the internal designs floating around. the actual first version might look a little bit different once we close that — probably. All right, thanks. — Cool. And then I guess the third topic and probably the biggest one um where I even had to talk to marketing if I can already talk about this today. Um, but of course I I wanted to share some news about the a AI assistant. And I think Sin already gave a really short teaser last month about a better AI workflow building experience and that we're working on something there. — Um, which is going to be this AI assistant. And this AI assistant can do Man, your fireworks are going crazy. This AI assistant can not only build workflows but it can do more than that. Um we kind of believe in this becoming like a super agent that can do various tasks for you. One thing is of course building testing iterating workflows for you and of course building other agents as well right can do that but it also can do one of tasks. will have a much wider toolkit that you can use such as access to your computer, access to your file system, browser use, skill support, um like all these things, multimodal, like it's really going to be able to do lots of things besides just building workflows. It should also become proactive and learn about you and kind of come with suggestions um on how to optimize your automations, how to do certain processes that you haven't automated yet. Um how to do that and then basically just click and it kind of goes building that for you. Plus it will also be reachable wherever you are. What this means exactly is that there's also going to be a mobile experience um for this agent. Um exact details we're going to see. — I have questions now, but I'm But what this means is most likely of course Telegram, WhatsApp, like the usual things, but maybe even an app as part of this, — right? Cool. Did you notice every time you're saying something awesome, there's fireworks going off now? That's — Oh, yeah. Like maybe this is just the special effects. — I heart I'm paying these kids. — Yeah. And um yeah, to potentially make uh things even more risky here, I thought I just share a little video. um as a sneak peek from one of our developers who runs with a very early version um that we're testing right now um to maybe just show a little bit where this is going. It's of course not where this should end or where this will go in the long term, but I still thought would be super cool to show a little bit of a sneak peek um of where this AI system — and just like for a bit of context for people, this is a video that we used internally to inform the team. So it's not like a smooth well produced or like it's not a marketing video but it's an internal thing. This is how we communicate internally. Right. — Exactly. It's literally a video from one of our engineers who built this and wanted to show it to some other folks at NM. — All right. Brilliant. Here we go. — Hi team. I want to show you a quick demo of the current state of instance AI. Uh so I'm going to ask you to build a workflow and then we'll verify that workflow using the browser use. So the workflow is just a form that accepts full name uh natural language time uh email and the meeting topic and then it should use OpenAI to convert that natural language to easel time stamp and create a Google calendar my calendar and send the Gmail confirmation. Uh I'm going to pause here and let it build the workflow. All right. So now we have a workflow. So it's using Open AI to parse that text the current time stamp and then using code node to extract timestamp from the parsed text and then creating a Google calendar event and finally sending the confirmation. So, let's see if this works. Well, first we need to publish it. I guess this wouldn't work because it's not published yet. Yeah. So, publish it. Sure. And then we can tell it use my browser to test it. Okay. So yeah, it's connecting to my browser. Uh it opened that URL. Now it should enter the details. It's asking me the email address and name. Uh just well test tomorrow 3 p. m. Mark the rest. Okay. So, now it's filling up the form. Cool. [clears throat] And the form was submitted. So, now it's going to check the execution. The execution was successful. So, we should see an event. And there it is. Q2 planning sync uh meeting requested by test via the booking form. — Yeah, that's it. Thank you. — Yeah. So, as you can see um of course um still a bit of work to do for us but uh already really exciting. There's also an internal version floating around um at NN that for example I played a lot with um so already a lot of fun um to build with this and super excited um where this is going. It's also really exciting to see that it's now able to control your browser for you, right? — Correct. So, it's going to have a large tool set like browser is one of them. Computer use in general, file system use, um I mentioned a few things. Man, every time I speak, that's amazing. — True. — Yeah. Um so, um yeah, super exciting stuff um to see where this is going. — All right. Um well, let's see what's next. This was the sneak peek. That was a reminder for ourselves. So that brings us to David. And that's the end of your section, Nick. [clears throat] Thanks, man. See you soon. — Hey, David. [clears throat] — Fireworks for you as well. So, welcome. Welcome back to the show. — Appreciate it. — Yes. — Thanks for having me again. — Definitely. Yeah. What are you going to be sharing with us today, David? — Yeah. Um I'm going to talk about N8 and connect which is uh internally I was talking to a colleague uh before this I summed it up as batteries included. — Mhm. — Um — what does that mean? Yes, good question. I mean so you know if you come into any today one of the you know the hurdles the stumbling blocks we've never been able to get users across just as smoothly as we'd want is the credential setup part right you come into an end you want to build something cool or maybe you come in with a template and you want it to run right away um and we've done things in the past like managed credentials on a net cloud out where for the Google services you have these buttons where you can just one click connect or in uh last month's live stream I think we announced uh the partnership with firecrawl for example where you can kind of one click create an account get some free credits and so we've done these things but they always still required you to take action and before you could really execute something or run something and so what we're doing with nend connect is we are basically first enabling uh AI token consumption directly through NAN. So you come in, you can build an agent, you can directly talk to it, it'll work, no need to go to an external provider page, create your API key, kind of break the flow of building something. And the second um kind of scope of this is we're also looking to extend this with kind of a range of services so that if you say you come in you want to build an agent that is able to do cool things it can also do a range of it has a range of capabilities out of the box like web search web scrape browser use uh OCR contact enrichment you know all of these things that you would kind of expect an agent to be good at as it's spawned. Um yeah and that is N and connect in a nutshell. — Brilliant. It's one of the things we run in a lot when we're doing workshops for example that's you know people we don't want to spend a lot of time explaining how to connect to other services and if you can skip that step you can go straight to the actually interesting stuff which is how to make it do something. — Exactly. And I mean you know also we internally in product development product and engineering what we do a lot is we spin up test instances. they'll not have all credentials. So I just for that I've been loving it the past uh week where I got to play with this an early version. Um so yeah it's feeling really powerful just being able to go and do stuff. — Um and I guess on that note I should mention this is already live for a select percentage of new cloud users right now. So we're running this as an experiment. Um, and uh, it's also live just with a limited set of AI providers at the moment, — but as we learn kind of the ins and outs of it, we're looking to yeah, improve it, add more providers, and then hopefully have it go G this quarter. — Brilliant. Okay. It's also going to be a really powerful tool to quickly test out different models and do like evaluations on them and see which one performs best for your use case at the best price. Right. It's — Yeah, definitely. Exactly. I mean, what we observe with NAN users is they'll have some brand awareness usually of the big model providers and what we think we can do with NAN connect is actually help them educate them about the variety of providers and get them kind of to explore different versions. Maybe there's a cheaper model that does the same thing that you want for this use case just as well. Maybe there's a model that does it better so that you Yeah, you're encouraged and you don't have any barriers to kind of trying out a broad um range of models and providers. — Brilliant. All right. So, already testing this out. So, some people might already see this. Usually with these tests, they might run for a few weeks, it might disappear, it might change. So, this is stuff that's in flux, but at least everyone knows something cool is coming. Yeah, exactly. I'm scanning the chat with half an eye. I'm seeing, you know, can I get access and is it coming to self hosted? Is am I okay to just jump on these questions or Yes, of course. We have time. — Okay. Well, about the can I get access part, you know what we just talked about, it's a limited experiment. We're assigning at random to not bias the data in any way. Hence, we can't hand out access uh on demand. Mhm. — Um and the other bit about will this be available for self-hosted. We want to get there. Um there is uh yeah we want to get there for the start it's n cloud only. — No. Yeah. Because obviously with this there is cost involved right and so there that's something we need to figure out. So — yeah exactly on nadm cloud there's like the you know clean connection between an account and the instance for self-hosted we don't have that. — Yeah. Okay that makes sense. I think that's it's a good thing good point to reiterate that you know all these new things that we're exploring often they start on cloud because that's for us the easiest platform to deploy things on but we do want to bring all of these to self-hosted as well and we are very committed to that internally like as I mentioned we were at the company retreat in Berlin a couple of weeks ago and Jan reiterated that as well like it's community first we like it's in our DNA we want to make sure the community gets the largest feature set possible but sometimes It's just not feasible for us to release it at the same time, but it doesn't mean we're not working on it. — Yeah. Exactly. [snorts] — All right. Well, David, I'm looking forward to this one as well. So, there's a lot of new goodies to play with soon. — Yeah, definitely. Thank you so much. — Okay, man. Thank you. And I will see you in the next one. — See you then. Have fun. — Cheers. — Yep. — Uh all right. Um then we have our

community section today. I'm talking with Zayn Khan who is going to be showcasing one of the projects that he did for uh a client. And if all is well, he is in the back room waiting. There we go. Hey Zane, how are you doing? — Hi. I'm good. What about you? — I'm great. I have lots of firework. There we go. So, every time you say something cool, I will play fireworks. Okay, that's the deal today. Yeah. — Um Yeah, man. Thanks for coming to the show. And um maybe for our viewers maybe you can introduce yourself say a bit like what you do and how you discovered anything. — Sure thing. So basically I'm a developer from background turned into automation uh developer. Okay. And basically two years back I was in basically introduced to NAN and quickly it was my main tool for building automations. Okay. So I have worked with a lot of uh companies and uh most of my clients I mostly suggest them to use nit just because it's flexible as compared to other tools and uh it's uh very easy to build the workflows here. Other than that I have collaborated with the jot form decoder to for n for the workflows also basically I am the creator on naden okay so I my goal is basically to publish as much as workflows for the community as possible also I have some premium workflows some free workflows so today what I'm going to show you is a workflow that I built for my client to manage their support uh tickets. Okay. So, uh let me basically share my screen and show you that let me Okay, hang on. Let me see where your screen is. I don't see your screen yet. Zane, are you sure you're sharing it? — Yes, I'm just going to share it. — Okay. — Yep. — Okay. Can you see my screen now? — Yep. I'll add it right now. — Okay. — Here we go. — Sounds good. So basically this workflow is for teams who regularly got support tickets from their production products or even if they have a company and they provide services they can also use this workflow. Okay. So I'm currently using jot form here and the reason behind jot form is uh like it's very flexible as compared to other forms platform. I have used Google forms, type form but when I switched to jot form it's very easy flexible has a lot of integrations also there are uh very new features coming in like agents you can apply conditions into the form so it was very helpful so that's the reason I'm using jot form and uh the idea behind using it also basically what I'm doing here is uh when a support tickets come from jot form then uh let me first show you the jot form. Okay. — So I will go here — and maybe you can tell me tell us a few words about like the type of client that you're doing this for like what kind of business are we talking about here. — Okay. Sure. So basically this setup was for a SAS company. Okay. So they have a live product and they wanted me to manage their support tickets automatically which we can do. Other than that uh sorry so managing the support tickets and uh we have a database where we have all the FAQs all the generic questions. Okay. So if a support ticket is low or medium uh severity then we reply them automatically using our workflow and if it has a high severity like client is facing an issue or uh is uh any other thing then for high severity we trigger a notification to slack okay — so this is how it's working also let me show you the support form right now Okay, it would be on your screen now. — Yes, we see it. — We have name, last name, email address, issue type, issue description and a file upload where users can upload their file of the screenshot what the issue they are facing. Okay. So, regarding the database, they were using air table. Other than that, we can use any CRM that we want. Okay. Hub staff or GHL or any other thing. We just need to replace the nodes. Okay, on my screen you will be able to see the air table uh database that we are currently using. We have a users table and support entries. Right now uh I have like duplicated it. It's not the real one. So I will go back to my N workflow. Okay. So basically when a leads comes in we basically search that into our database like the user uh from which we have received uh the request is using the right email or not. Okay. So basically after that our AI agent figure out the severity okay and the sentiment of the customer. After that we also update that record into our database. Also we have another agent which basically check our internal uh vector database. Okay. So it checks that this support ticket can be replied automatically or not. Okay. And this is just for the low and medium uh severity tickets. So after that our AI agent basically reply to that email automatically and update that they got. That's it. But if AI agent thinks that this low severity and medium severity ticket cannot be resolved uh resolved by its context or its memory then it uh also send the notification to Slack so team can handle it. Okay. So this was something that was very linear but the tricky part was here uh the this section below. Okay. So suppose a user uh submitted a request but he is using any other email to submit the request. Okay. And uh if you have been into teams we need the real email that he is using or she is using uh for the account on our platform. Okay. — Right. — So if the email is not found in our database then we send an email to that person. Hey uh you need to provide us your account email that you are using on our platform. So that was the part we are doing here. Okay. And when they reply we have a below section first check that this reply is for which ticket. Okay. Then we map them out and then send a message to the Slack team that hey you need to uh do this. Okay. You need to figure out the solution and send that to the client. So that's how the whole workflow is set up. And uh regarding the models, you can change it to anything. You can change the database from air table to HubSpot or any other uh thing. Okay. You can change Slack to any other uh platform as well if required. So that was the whole idea of this uh workflow. Okay. So right now let me submit a request and show it uh to you how it's working. Okay. — Y — So I will go back to my form. I have my information here and let's put uh something like a technical problem. Okay. So hey I'm not able to log in. The error is showing on the screen which should not. I cannot get in. Let's see what happens. — So I suppose that should classify as a high severity issue, right? If someone cannot use the service. Yeah. — Yes, you're right. So yes, let me submit the workflow now. — Done. — You can also add attachments. So maybe like a screenshot or something. Is that automatically routed into the agent? — Correct. Not into the agent right now. It's saved in the database currently. — Yeah. — Okay. So now if I go back to Slack, let me first show you the workflow. Okay. So what [clears throat] happens is it's first checked our database, the user was there, a agent detected the severity and sentiment. Then the severity was high. Okay. So then it's so it's directly send it to the team. And if I switch to my Slack uh tab, so this is how it basically send the message. It's just not send the details of the issue. It also map out the user details, its current plan, when he started or and we can also add the uh value he has provided until now. Okay, it really depends. Okay, so this is how this workflow is working. We can also submit a load severity issue and check if the auto reply like go on. — Yeah. — Okay. — And you also mentioned that you're doing sentiment analysis, right? — So how does that factor into the routing? — So based on sentiment if the client is frustrated then we also consider that part in the severity as well. But it's not like uh if user is asking a general query and he's frustrated then obviously it's not a high seability issue. It's just a general query. — So it depends. — Gotcha. That makes sense. That's a nice touch. — Well, so do you want me to submit the low severity issue? — That's right. Yeah. — Sure. So what I will do is I will go back to my form and let's submit an account question. Okay. So I have a question about how can I switch my plan. Okay. So we [clears throat] are supposing that someone was not able to find that in our blog section. So let's see. They want to downgrade their plan. I'm sure that's not a high priority question. Okay. So, let's submit this one and I will go back to my NAND screen. All right. So now it's going into the rag branch. — Also, one thing I wanted to show you here. So, uh people might be confused on like why I'm using pine cone and this stuff and how I save the data into the pine cone database. Okay. Mhm. So basically if you want to save all the FAQs that you have on your website, if you have a PDF, something like that, I have a separate workflow on my creator page. Let me sh change the tab. Yeah, this one. So here uh you can submit all the URLs that you want to be scrapped and saved into your database and it will be done. Okay, this is an other workflow that I have published — uh and you can use that directly but I am uh using the database here. So I will not go into this one because this is just updating the database and I think the email is incorrect that I have the credentials are incorrect here. So the email is sent back. Okay. So I will go back to my email. Okay. So let me share the next tab. So on my email as you can see I have got a reply like steps to switches uh selling plan. We can uh format the email but that's not the case here. — So here are the steps that user can uh perform to change his plan. Okay. So that's how our agent is automatically replying to that email. Okay. So by the way uh this workflow will be live today on my creator's profile. Also the [clears throat] rag pipeline which basically scraps the data and save into pine con is also there. So — I have a slide after this with a link to your template so people can grab it right away if they want. — Okay. Sounds good. — All right. Very interesting. It is a very like common use case of course but it's nice to see how you split it up into like severity and uh and sentiment detection and using that to route it in the proper way. I think that's uh that makes a lot of sense. Uh I heard from the team backstage that we have a question for you from the viewers on uh on YouTube. — Sure. — And it's Petra asks quick question on error handling. Is it just not visible here or how is it handled? And now I assume this is like a demo workflow so you didn't need to go super deep into that but maybe you can talk a bit about how you would handle that in general. — Okay. Uh so basically if I talk about AI agents okay I know that we are using AI there. So the response that we got from the AI agent can vary. Okay. But as you can see I am using a structured output parser here. Okay. So I'm defining that I need output in this specific format. Okay. So here there will be like zero chance that uh the output I will get will be different and that will cause an error. Other than that uh in the workflows as you can see I can have updated uh the workflow. Okay, like I have used the I can use the air table uh update tool here. Okay, but I have not used it here because uh the decision power of updating any record sending an email go to that AI agent. So currently I'm using uh some strict patterns like a structured outputs uh like that to basically try to not getting the error in a live environment. — Mhm. Yeah. I think that that makes sense. Maybe there's also a level where you know services could fail, right? Maybe Gmail is offline or your pine cone is unavailable for a few seconds. There's lots of — an example regarding that as well. So right after our call like two hours ago, I was having issues with Gemini. Okay. So it was just saying service not available, service not available. Sometime it works, sometime it's not. So that's a case as well. Okay. But now I'm using cloud is working perfectly. So it depends on the services as well. — We can what we can do is we can enable the retries like I have enabled the retries here. — So if it fails it will retry three times after every 5 seconds. Okay. So this is how we can uh retry automatically for the errors. — That makes sense. Yeah. And in agents you can also have a fallback model. Uh so if one service is not available, you can just switch to uh it's going to automatically grab the next. But that requires a lot more testing to make sure that your prompts work equally well in other platforms, of course. — Um all right, Zane, with an eye on the time, uh I think we need to start wrapping this up, but uh thanks so much for your contribution here. Um as I mentioned, I have Where did I keep my slides? Here we No, now we're all the way back at the start. Sorry about that. And it's not very fast today. And almost. So, if people want to grab this workflow, this is where they can get it. [clears throat] Go to go. n. io/tick-templates or scan the QR code. And uh on that page, this leads to a page in our template gallery that can also click on your username and see all the other templates that you created for NAN. And um I heard that Jot Form also did a quick interview with you uh which is now live on their blog. So people can go and check that out as well. So they can learn how you work with Jot Form, the type of automations that you build and people can go try it out for themselves as well. So they can head over to link. jotform. com. jotformum. com/natan- community-livestream or again just grab the QR code and uh maybe the team will drop the link in the chat as well for easy access. Um well that's it. It's uh thanks so much for joining us today. I know it's quite a bit lazier for you uh where you're based. So I appreciate it a lot. Uh and I hope to see you again in the future then. Thank you. — That's totally fine. Thank you. Have a good day. — Okay. Bye.

All right. And that brings us to the community challenge section. And it's going to be a full house because I have Tino, Stfano, and Liam here with me. So, looks like we're going to have fun. There's no more fireworks, guys. I'm really sorry about that. That's — It's okay. You got these three guys now on stage. So, uh, no more fireworks needed. All good. — You are the fireworks. There we go. — We are the fireworks. — All right, Tino, tell us. — Awesome. Yes. Uh, another month, another community challenge. So, um, before kicking this off, I last the last two months, uh, I just want to give a quick introduction in what the community challenge is, what we've been doing. Um, but before we go into that, I want to make a quick shout out to everyone who joined the challenge, everyone who uh, started and submitted stuff for the challenge, and especially those of you that left amazing feedback. Um, we've seen so many cool things from people finding out about the challenge in uh interesting ways to people coming up with bugs because unfortunately we have some of those floating around, but you help us fix those and uh some super cool use cases. So, please keep that up. We read all of the feedback that we receive. Um, just want to make sure we point that out. So, what is the community challenge? It's something we created. It's a month-long build and learn kind of like hackathon meets hands-on learning uh style uh education experience. Every month we're coming with a new challenge in the form of a new theme and a new use case. A lot of people asked us, I love some of the features you have, but I have no idea how to put them into practice. And that's where this came in. So, we're going to give you a realistic use case for you to solve. Entirely fictional. So, if you happen to know someone who owns the same company, we are absolutely talking about a fictive fictional company there. I'm still hoping for the moment someone sends a message saying like, "Hey, that's us. " But so far hasn't happened yet. Um, your role in all of this depends per case, but your assignment is to solve it with an NN workflow. What's in it for you? So the main thing is we offer you an opportunity to learn new skills. See it as the crossword puzzle at the back of the magazine that you fill in because well you kind of would love to do it. You would never make the crossword puzzle yourself uh or design it yourself but you would love to fill it in. That's kind of how you should look at this as well. And perhaps the coolest thing you may receive an email from Bart saying I would love to have you on the live stream. Of course, there's a lot of other people in the community as well. Uh, so we invite you all to join us in the Discord channel and connect with everyone out there. How do you join? Very straightforward. You scan the QR code that you will see later. You go and read the full challenge, download the resources, and of course, you're going to join the internet discord to stay in touch with everyone from the community and also to ask us questions if you have any questions. Then the fun part, you build the workflow and then you submit it using the link on the challenges page. And once again, if you fill in extensive feedback, we read all of it and it is extremely well appreciated. — That's it. — To the challenge page will be at the end of this section, so stay tuned. There's also going to be a QR code, so it's going to be really easy for everyone to find out. — I'll make sure to drop it in the YouTube chat as well. So, uh, if you don't want to scan the QR code, I'll make sure it's there as well. — Off to Savano for this month's winners. — Yeah. So, good. Thanks for the introduction, Tino. I must say, uh, every time the challenge is a really cool experiment, I must say we do the use case and then we see what people just build out. And it's always impresses me of how do people think about a challenge and what do they do? Uh, yeah. To solve what we set them up for basically. And we've seen all these kind of things from simple workflows to overengineered workflows. Um but not always the ones that we choose to win are the ones that solve the problem the best and that impresses us as well. Um to give a recap, the last April challenge was about firecrol to build a webcrawler agent. Uh and we asked um our builders to create something for one of the case studies that we provided them. Um so you could choose from easy to more difficult and based on that they built a workflow and submitted it uh to solve that case study. So for this challenge this month we had uh three case studies. So we go to the next one. We have three winners for those as well. Um so shout out to all of these three that had the best ones. So um what we want to do also which is also really nice of this community challenge is to show what they have built. Um so you can also get inspiration but also see how they try to solve the case uh in their minds basically. So I think we can go one by one uh but I think you can put them in and we can [clears throat] start the package. — So we're starting with Devon Shu who is going to be showing us an npm package manager script after that off to Tilman who is doing job scouting by scraping the web and then finally Naz is showing like local scouting for I think it was like delivery or meals something. Well, we we'll learn like we're I'm going to play all three of them. So, enjoy. — Every time a developer wants to use a new npm package, they manually check GitHub, npm and documentation, which takes time and doesn't scale. So, I built an AI powered package intelligence agent using niten and fire crawl that automates this entire process. The workflow start with simple package input and fire crawl dynamically discovered the correct npm and GitHub urls. So there is no hard coding required. Then I use GitHub npm APIs to fetch realtime data like stars, open issues and last commit activity and weekly download. This data is combined to compute key signals like issues to star ratio, activity status and the workflow also handle invalid or missing package gracefully. An AI agent then analyze all this data to generate insights such as risk score adoption level alternatives and provide a clear recommendation. Instead of raw data, it generate a clean sack ready summary which observe fact inferred insight and final decision like use, consider or avoid. This transform manual package research into fast reliable decision-m system by combining fire cloud for discovery API for equilman's submission. And my computer is a bit slow. — This is my approach for a job crawler. We start over here with raw values of the candidate, everything he can do and everything he's looking for in a job. Then we are building queries which is quite a classical approach but it gets us all the links we need for the next step. So scraping the page and bringing it into a unified format so we can work with it. After that we are bringing the salary to euro if it's not in euro and bringing the date into a uni a un unified format as well. Then we're moving on to the harsh uh of the nitn workflow raking the suitability of the job and candidate. So it goes from zero to 100 and of course it evaluates if there are deal breakers involved and how good the candidate uh matches uh the job. If it's above a certain threshold then we are moving on and we can really burn some tokens. We try to find funings additional information and um look up reviews for example on Kunu if it's a good employee employer or not. After that, we're sending a mail and this arrives at your inbox so you can see everything neatly on one page and if it's a really great job you also have a complete solution. Nice. Um, finally we have Naz. — Hi, my name is Naz and this is my submission for Michael's use case. It starts with the weekly scheduleuler and then the rotation seat kicks in which basically changes the search parameters for Berlin neighborhoods and business categories weekly so that we get a fresh new leads for Marco every week. And then the discovery agent goes out and finds 20 leads based on assessing its relevance and filtering down only ones that are worth looking at by assigning them a preliminary score for the rest of the payload. Then we parse the agents output into a workable JSON format. Then we assign uh a unique ID to each lead so that the system can recognize it across different weeks. Then those results get dduplicated against the database which looks like this. From those we only select the top 10 ones for enrichment to keep the workflow fast and cost efficient. Then the enrichment agent uses fire crawl tools again to visit each website and extract real signals and assign a final score. A high score is only possible when there's a real verifiable problem such as a broken contact form or a dead link. Yeah. Enrich leads then are stored cleanly in the structured database ready to be referenced, filtered and exported later on. And then we craft a clean, scannable, actionable email for Marco with enough context to get him reach out to those leads immediately. and then the email gets sent. So I prototyped as fast as I could and I tested it vigorously and I iterated all my findings and this is how this workflow came to be. I hope you liked it. — That was super cool and congratulations everyone for winning this uh this round. I just learned that we're going to be adding a challenge section to our template library where all these winning uh workflows will be available as well. So you get some extra uh visibility there uh as a winner of our monthly challenge. Um great. With that said, what are we going to be doing this month, Stefano? Yeah. So this month uh we'll do something which is uh I would say quite complex but one of the most important things you can do uh in building workflows and definitely if you're working uh to in B2B for businesses uh because we all know how good AI can be but we also know the weaknesses of AI. Um, we can build any kind of build big workflows using AI. Uh, but we always know there are uh the hallucinations that can happen that result in a worse workflow than you actually want. So for this month, we'll go for the human in the loop automation. And before we go to the case study, I want to give um yeah the field to Liam uh to explain also a bit more about what human in the loop is and give you a quick demo also of it. And then we'll go to the case study. Liam, I see you are the human in the loop advocate. Now — I am. That is my job title for the day. Can you guys see my screen and hear me? Okay. Is that uh — we don't have your but hang on where? — Let me add it. — Okay. — There we go. Yep. — Should be shared. Perfect. So I am the human in the loop advocate today as my official job title. Um and then just a state of mind for the rest of the days. And it really isn't that hard. The team has made an amazing job of making this um easier in recent uh releases. And I'm going to move pretty quickly here to show you guys a few examples of this. I have I had five minutes. Now I probably have two. So you'll see here these human review steps on tools. Essentially, you can put any tools like normal. And let's say this is a case where we're chatting with something that can like cancel subscriptions, refund a customer. All you need to do to add a human review step in the middle here is there's this new plus button right here. You just click on here and you add in a step. And then inside of this, I'll show you one I already set up. There's this new money sign tool where you can put in the actual parameters. So you can see what that actually looks like. So that's all it takes. And when you put that in there and have it respond to you, it can't use that tool without first being approved. And this is really cool in this case because it's chat based here. So these will send back to the same chat. Let's see if I can get this started. Refund John Smith uh 20% or not refund make coupon bond for John Smith for 20%. And while that loads, you'll see here this one's select. So this is different. Imagine like you're here working with this, but then in order to make a um refund a customer, your manager has to approve or something like that. So this could actually go off somewhere else and someone else has to approve. So this goes right in the chat. This would send back to Slack. And then we can see this came back since I modified that with the actual parameters which we can see right here. like John Smith 80%. We get this in here and we'll say what the heck I said 20%. So we can say decline for instance. This is right now the easiest way to do this if it's just tools but there's some other options which I will show you very quickly like here where it doesn't have to be a chatbased setup. It could be event driven. So it could be something like a form. It could also be something like when a customer submits an order on your store. It can then go and trigger something like in this case inspired by the challenge from this month. If you submit if someone submits something creatively, it can go in review that. And let's say that I don't really care about leaving positive feedback on stuff, but like I don't want to leave critical feedback on a co-orker's work without knowing about it. But then instead of actually going in and doing the work, I can be sipping a coconut by the beach and just press uh accept on Slack for the feedback there. Right? So, this is kind of the same idea, but it doesn't have to be chat. Oops. Doesn't have to be chat, right? I think I'm moving right along here. Here's another one where human in the loop isn't just for agents. You can not trust your co-workers, too. Sorry, guys. Uh, imagine if you need to approve something, right? Like you have someone working under you or a co-orker and you guys need to approve each other's work before you send it to a client or just make sure it's good. Like in this example here, I have someone submits a form to send a price to a customer. It can go and get it, send it to you in Slack through here, and then you can approve it, which will send it automatically, or deny it, and then it'll just send them right there. And then last step, you can do stuff much more advanced. And this is the cool stuff. I hope we see some of this in the challenge where you can make a subworkflow and now any nodes in a subworkflow now work through the tool. So when you go through here and the agent calls this subwork, which I forgot to rename, that should say human review. Let's say this is an instance where you say, "Okay, I want to refund this customer or give this customer whatever. It can come through and do any complex routing, any math, any whatever. " And it can go to different review tiers. It can do different things. You know, you're it's just a workflow. You can make whatever you want your creativity to limit. Um, and I uh can take any questions or anything if we have time. I think Tino still has some work to do telling people what we have to do, right? — I think so too. But um never trust your coworker. So that's why I'm handing it over to someone that is trustworthy, Steano. — Um who will be introducing the next month's challenge. — Yes. So um this month um you will be joining the company Relay. Uh it's a social media content agency and they need your help to scale their content pipeline while keeping the humans in the loop. Um they have grown from a small company to now around 20 plus fullfledged uh marketing agency. Uh and they want to automate and speed up the processes for the creative team. Um, if you go one further, Bart, you will see the team here. And what we want you to do is to build a workflow. Uh, and I hope already that what lean has showed you can already go your uh, creative brain going and think about how you want to solve this. Um, but we have three people in the team who all have their own place in the process. We have Sophia who is the strategy part. She's the trigger actually from this process and she thinks about what can I create a post for our client. Um, if that's approved, she goes to it goes to Marcus, which is the creative person who creates the idea into a post. If that's then approved, it goes to Taylor, which is the last step, the reviewer who does the final check and decides that this can go to the social media uh directly or plan it um all based on the clients that they work for. So the whole challenge is based around the human loop. Uh we don't really look at what the output will be of uh of the social media asset. But the idea is of how can we create a process in which these three come back uh where they can revise, give feedback and go to the next step in the process. Uh if we go one slide further as well, you can see here a simple diagram of how that looks like. And to automate and make this process also faster, we want you to use AI in this uh in these steps. So also there you're quite free uh be creative of how you want to use AI and build this workflow. Uh but at least it needs to have this at least two human in the loop steps. So at least two of these people that we have defined for you. Uh and use AI to generate the output. uh and as much yeah how good the feedback will be and how good you do the revisions in this process that's how you get your scores uh if you sign up for this challenge you will get a resource pack with all the information that you need to solve this challenge to understand what you need to do in that you can find the QR here um or go to the community challenge page um yeah and then you get all the data sets all the information you need you can work first from for one client for relay uh and then adapt it to any client that could in a later stage. Uh so um yeah, I'm curious to see what you're you guys are going to build uh with this human in the loop. It's nice that already makes some things easier. Uh but I think in you can do so many things to make this process better. Um and that's also what I'm curious at. So you can follow really this in the same way the same process, but if you can also find a way to make this process even more efficient while the output is even higher, I would say that gets the highest points of this challenge. Um and the winners again indeed will be in the next live stream showcase. Uh and also again their uh workflow will be added to the official NIDAN template library which is uh really cool to see and any anybody can use it in the community. Um — yeah that's it. — I think I already found the solutions. They found out that the best solution is to just trust your teammates. No, and then you can skip all that stuff. Every time Liam sends me ask me something, I'm going to be very suspicious of like what's behind it. — That's it, Bart. You're invited to the next community live stream. Congrats. — All right, guys. Well, thanks so much for for all this. Uh I'm also really excited to see what's going to turn up and uh we'll see the winners in the show again in a month. — Absolutely. — Take care. See you soon. Bye. — Great. With that, uh, it is time for me to wrap up. And I just have a few things to mention. So, uh, if you follow us on social media, you may have noticed that we started doing Fridays, uh, for AI days, um, which is a way to share cool stuff that you built. It's very lightweight. So, every week we post like a prompt, uh, and we ask you like, hey, share a cool workflow that you built for this and tell us a bit about it. Uh last week we asked about agentic workflows and the solutions that we saw. We got about 45 people replying. Florian first uh he made his family life a bit easier by scanning school emails and extracting relevant dates and calendar appointments out of them. Malcolm was uh monitoring and documenting his NAN workflows with an NN workflow and then storing all that in notion. Marian Nijan, he was looking for contractor jobs uh online scraping data and then compiling a report out of that. Um if you want to see these like more about these entries, take a look on our LinkedIn account and you will find all of these responses there. Uh we'll be sending all you guys uh gift cards uh so you can get some cool merch from our store and we will have a new challenge for you all tomorrow. So stay tuned for that. Then as always, we're hiring a lot. Uh, and we have a new role in the community team as well. So if you would like to work with me and Melinda and Adelene and Tino and Stefano and do all these cool things like challenges and meetups and live streams, uh, please do respond. uh we are looking specifically for someone who loves to connect with people who loves to build uh local events in the in West Coast area but also in the United States in general and who would like to help us build out the ambassador program in the United States as well. Um we have more roles. I think right now we have over 20 uh positions open. So take a look at nan. io/careers for more information. Then you might not be aware but we have a newsletter. We send one every month. Uh, and it contains a lot of interesting updates about the events that we do, the jobs that we run, but we also usually talk a bit about like uh like what we call hero stories that could be something cool that happened in the community like a specific event. We will highlight cool workflows etc. So uh take a look on n. iosletter if you're interested in that and u sign up. And then talking about events, I just checked our calendar. We have over 25 events scheduled for the month of May already and I think a few more are coming up. Um so uh I would say go to n. io/ community/events and browse through it or click on the little map there to see uh what's in your area. Uh one extra cool one that I would like to highlight is the women in automation event that uh the women in the N10 team started running recently. We had an event in Berlin last month and it was so successful that they wanted to do it again and we're getting lots of requests from other communities who want to do the same thing. So if you are interested in that, please reach out to us and we'll support you to run such more diverse events as well. And then finally, if you want to become part of our ambassador program and help us host these events, we're always looking for new ambassadors. uh we are mostly looking for people in like capitals or state capitals in Europe, United States but also around the world. So if you live in let's say uh New York or Seattle uh or in London, uh reach out to us uh see if you are interested in doing events with us and meeting lots of other community members at the same time. I think that is all I had for you today. Sorry for going over time a little bit, but I hope you had a good one and I will hope we will see you again next month. Thank you. Bye. [snorts] — Hi team. I want to show you a quick demo. — Nope, not that. Hey. Heat. N.