Transcription
Hello, friends. As they say, long time no see, so without wasting time, let's get straight to the point. Today is December 11th, twenty-fifth year, and I want to talk today about NAN, about the second version of N. And, as I've said before, I always announce dates because things are constantly changing. And if you recall, language models generally appeared in November twenty-second year, and now, when you look back, so much has happened that these 3 years that have passed have brought a lot of various changes. Therefore, if you are watching this, keep in mind that something might have changed. You always need to check the latest sources. And on the fifteenth, the second version of NA10 is released. This is a big release. It is currently available in beta. Well, I want to briefly talk about some important points, in my opinion, that are worth paying attention to, not all of them, there are many. About significant important points for those who are not aware of what Nent is in general, in principle, NAD is an orchestrator for executing workflows. They are called Workflow and Nat is called, I call it Nat. Some say N8N, some say Nathan, it's different. I call it Nat. It's more convenient for me for some reason. And it's a very popular tool, although its history began in the nineteenth year. It started as a kind of integration hub, connecting everything with each other using squares and rectangles into workflows. And since, probably, November twenty-fourth year, it gained enormous popularity because everything related to nodes, i.e., nodes connected to AI, was brought there. And the madness began here. Everyone dived into N. You see on YouTube and wherever you watch, a lot of different workflows, Zero to Hero, how I got rich, how I offloaded all the work to agents, and so on. But this, in fact, is not exactly true. The tool is powerful, and the tool can help you solve a large number of tasks, but you need to master it. In my experience, to understand N, to start doing something with it, you need to spend, well, I don't know, a month or two of intensive, persistent work. Well, and traditionally, as with AI tools in general related to our industry, the process is as follows: first, nothing works, and then suddenly everything starts to work out sharply, and you're like, oops, and you're already a new person, everything works well for you. But up to this point, we lived in version N10, which was version one. This is the workflow. It looks something like this. And if you look at its standard view, from left to right are the nodes that are executed, and the workflow goes through them. And through this workflow, we can track the behavior, actually, the points of the workflow and see which node or which node was executed. In N, there are quite a lot of things, a lot of nodes, a lot of nodes, and so on. And so the guys, after updating the first version, which was a couple of years ago, are releasing the second one. We will talk about it today, actually. I won't delve into the essence of the tool itself, it's quite extensive. I'll just talk about important points that are worth paying attention to. Firstly, where to start, if you are already using Naiten, and you already have an installation and so on. There is a natural site at DOCS, which has an article Breaking Changes, where they actually describe what has changed in the second version. Well, this is not an exhaustive list, but it is a list worth paying attention to. And if you already have an anan, for example, like me, I have it installed, you can go to settings. And in the settings, there is such a thing called Migration Report. It will show which versions, what problems you might have with Workflows, because not all Workflow versions will be compatible with version 2 due to changes. This is the first thing. You need to pay attention to these workflows. And the second is instance issues. What do you need to do for the migration to the second version to happen painlessly? The Nend guys themselves write something like this: that on the fifteenth, they will have a release, and after that, for 3 months, they will continue to support the first version, but will not release any updates for it. That is, everything will happen in the second version. Therefore, you need to migrate to it as soon as possible. And those who have already installed it themselves or are using the community edition, which the guys offer here, well, here they are, will be able to update and start using the second version. What will we talk about today? I want to talk about, probably, my overall impression is the following. Nend is moving from the level of a toy for home use to something towards a very serious product, in my opinion. Well, judging by the changes that are here. The changes are more technical rather than visual, but nevertheless very pleasant and very useful. Well, let's start with the fact that this is what the interface of the second version looks like. Here the nodes have become cute, flat. So it reminded me a bit of when I was involved with mobile applications. The transition to iOS 7.0, when flat design appeared from the skeuomorphism that was there before. So it became flat. Nodes are executed exactly as usual. Hello. What are you going to do? How are you? Wonderful animation. We will all get used to it soon. Everything works pretty much the same as before, with some exceptions. Firstly, pay attention that when you migrate your old workflows to new ones, you need to pay attention, first of all, to the node, because unlike a tool like flowwise, which can upgrade or migrate you to new versions of the tool, N cannot do this automatically with new versions of the node. Here, everything must be done manually. How to do it? You can go into any of the nodes. For example, I have an AI Agent and click on Settings. Here you will see that the AI agent Node version is 3.1 latest. This is the latest version of the node. If you look at some previous one, well, for example, I can open something from my previous installation of NA10 here and look at some node. Here is AI agent. It's quite old for me because it's an old workflow. And here is Node 1.7 and the latest version is 3. Nend generally supports backward compatibility with previous nodes, so it will work for you. But some new features and bells and whistles, of course, won't work, so you'll have to keep an eye on that. So, what is worth paying attention to here? With the advent of the second version, there have appeared, well, they were actually in the first version too, very interesting possibilities of language model nodes. For example, there is Opena, which has now become version 1.3, I have version 1.3 latest, supporting what is called Responses API. This is a new version of the API that came out after chat completions, traditionally widespread. And here, directly through the node itself, you can call tools, you can call file search, and you can call the code interpreter. This is implemented by OpenAI on their side. Well, that is, now there is an opportunity to perform searches in very simple ways, for example. What's new in the world of AI today? And, accordingly, the OpenAI node itself will now perform a search within itself and tell us what's new in the world without any additional tools that we attach as tools. You need to pay attention to this. And this can be used for your own purposes. This is specific to the OpenAI node. Just know that such an opportunity has appeared. I want to point out that you just saw the interface of NTN 2.0. The active on/off button is now missing. So now all workflows, to be available, must be published. That is, we click publish, and here we have some version, we publish it. Now the workflow is available. And the reverse operation is behind the three dots, called unpublish. And an important addition is that all child workflows that are called, well, for example, I call the Replicate Generator image generator, it must also be published, because previously NA allowed calling unpublished workflows. Now all workflows that we will call must be published. This is an interesting approach, an interesting new concept. Well, let's look at what can be done here. Generate a sunset image. While it's generating, I want to say that this is my favorite generator, which I use through replicate. Replicate is a service that hosts a large number of models, similar to KI, to a bunch of similar services for API to various models. There are image models and various video models here. I often use it for generating images in some simple projects. And it will be called by me within, actually, the generator. Well, it generated an image for me through Flux 2. This is one of the latest models, by the way, that Flux released either a week or two weeks ago. It works very well. And what's interesting, why I decided to talk about this at all? If we dive into the workflow that actually generates the image, here it is, the replicate generator generator. Let's go into it, and we will see an amazing new thing here, which makes me very happy, because, if you remember, Python, Python support in NTEN was always in beta in version 1.0. And it was actually implemented through POE - this is a library that implemented Python in WebAssembly through WebAssembly. But now they have made a new native system for implementing code in code nodes in Python. On the one hand, it's great, on the other hand, new additions have appeared that are required for NTEN. They are called task runners. That is, now you need to explicitly add special runners that will execute Python code. Well, including JavaScript. Previously, they worked within the main instance, but now you need to work with them separately, declare and update them separately, and manage them within the settings of the ten itself. But what we get in the end is an isolated Python runner, which can not only execute Python natively, not through some PO, but now it's Native Execution, excuse my French. And we can also specify the number of threads, i.e., manage the parallelization of code nodes. This is a huge step forward in terms of performance and security, in my opinion. This is great. Well, and here is the result of the image that you saw, this is precisely the Python Runner, which returned the URL from replicate to me and gave it to the parent workflow, which showed me the image in the chat. So, here is this picture. Well, and a couple more points that need attention. So, firstly, Native Python, I said about My Sigal. Some people used MySigal in SNiten, they are trying to get rid of MySigal and Maria DB altogether and are rapidly moving towards Postgres. The thing is that Postgres is also used by them for enterprise setups, i.e., when there is Nat main, i.e., the main Nat instance and worker instances, i.e., Nten worker instances that share tasks at the workflow level. Only Postgres is used there. And therefore, the guys now, I think, are moving towards Enterprise. And within this enterprise, they will further develop the product. That is, they are becoming, as I said at the very beginning, very serious, in my opinion. Besides this, what else have they done? They are limiting access to the file system at the level of the Nat Files folder, which is located nearby. And an important thing that you should pay attention to, if you used it, in N, previously nodes called execute Command. They are located, well, that is, these are commands that allow executing shell commands at the level of the Nat container itself. I think it was called either Shell or Execute. M, Shell Execute Command. Here it is. So here you can directly execute a command within N itself. Now they are prohibiting this. Obviously, there are flags that will allow the execution of such commands, but by default, it will all be turned off. And so, if your Workflows use such things, they need to be removed and eliminated. Well, and the last thing I want to say. In N, by the way, there are many interesting things that they don't explicitly talk about, but they are in an experimental stage, experimental features. For example, one of these experimental features, for the future, was data tables. If we go into personal in the second version of ten, or in the first one too, we get data tables. That is, this is actually a native table that allows working with data within NN. So you don't need to use any sub-bases, attach spreadsheets, airtables, what else. You can create data table tables right here, call it data, and work with it. There are specialized nodes for working with this data. Well, this data table was also an experimental feature for some time. And in conclusion of our today's short neuro-five-minute session, I want to say a couple more things. Currently, NEN has an experimental feature called chat. It's called chat module or chat chat module, I think it's called, yes? So, it looks like a traditional chat. Hello. You can chat with different models here, if you have keys set up. Most likely, such chats will be made later so that you can stream and transfer such interfaces to external systems. But the most interesting thing is that you can, within your own workflows, for example, if we go back to my generator, which I had, to my generator, which I had, here it is, the generator, and here is a node called chat message, chat trigger built-in. It has a checkbox make available in NA chat. Previously, it was only publicly available. And now, if you go into chat and go into custom agents, you can call entire workflows through these chats. Hello. You can do a multi-step action. Search for AI news for today and make a picture about them. Period. So, what should happen? Now the workflow that I just showed you, the replicate generator, which will call the search using responsai AI, should be called. After that, the OpenAI node itself will create a prompt for generation, and the generation will call, generate an image, and ideally return it to this chat. And I want to point out that I now have a Runner running and active that allows executing Python code, which actually prepares the results of replicate's work. Well, it takes some time, and I think 80% of the time is taken by replicate itself, preparing the image. So this is the longest time that is required. Well, my friend, where is the picture? The next event, I don't see the picture. Open workflow, but this is a demonstration thing. Of course, it always happens like this when you show something, it naturally stops working. Let's see what was in Executions 16:01 - that's right now. And what did they answer us? The picture was still some kind of humanoid robot. Yes, of course, exactly. Here's the picture. I don't know why it didn't appear in the chat, honestly. When I ran this scenario, it displayed perfectly, but now for some reason it decided not to show it to me, although it should appear in the chat. So, where, where is the picture? Ah, here's the picture. Thank you very much. Well, I just had to say, well, models are like that. And chats support pictures. I repeat, this is an experimental feature, so you won't see this thing yourself unless you explicitly enable it. It will appear within chats. And the last thing before we wrap up the five-minute session, pay attention that both the first and second versions of N have started supporting MCP servers. That is, now Nat can be used as a source to which you can connect and execute natively at the level of Nat itself. What does this mean? You can create some clever workflow within NAN and take any tool that can communicate with an MCP server, for example, clot code or clot AI desktop or any application, in fact, that can work with MCP, and make Nat act as an MCP server with the ability to execute these workflows and return results back. For this to work for you, besides enabling it and setting up access to MCP, as is usually done, you need to specify the workflows that are available for calling. Because MCP servers occupy a lot of space in the context of models, and it's done smartly here that selective MCs are available. And if you do this, pay attention that it is very important to make a description, and in your edit description, your workflow description must be very clear and specific about what it is about, so that the tool calling the MCP server understands what this workflow does. But this is for the future. When the second version is released, we will definitely go over, possibly some more changes, because this is currently a beta, it's not a release version yet, it's just a beta that you can play with and see what's inside. We will go over the release changes, talk about MCP servers, and generally torture NEN. I must tell you, I look very positively at the changes that the team behind the product is making. Not only is there at least one or two releases of some bug fix updates per week, but the changes that are happening now seem very interesting and very serious. And I, as a person who spends a lot of time with this tool, see a very positive trend. Well, and of course, the entry barrier to the tool is decreasing. And the people who experience the greatest joy are non-technical people who can dive into it. If you are starting to use something like NTН at a company level, consider the possibility of an Enterprise subscription. The Enterprise subscription, of course, costs money, not like the free Community Edition, but it offers much broader capabilities, including working with Git, workflow version management. Well, there's a whole lot of stuff there. And within a company, an organization, it gives a big boost. So, figure it out and dare to try, and we'll see you. Have a great day, everyone.