Transcription
AI coding assistants are broken. They can write code, sure, but they forget what you told them five minutes ago. They lose track of your project and they have zero clue about your documentation or teamwork flow.
Sure, you can use MCP servers like Context 7 or others, but you totally rely on external resources which you have not the control and they don't always have the knowledge you actually need. That's why I want to introduce you to Archon version two, an open-source project built by Cole and the Dynamos community. He found it. You can find a link below that I think might be one of the most exciting tools for AI coding assistance this year.
By the end of this video, you will see why I believe Archon could become the operating system for AI coding and how you can try it yourself today. So, let's get started. In a minute, I spin up Archon at a real project and show you how an AI assistant can immediately pull in exact documentation it needs, create and track tasks in real time and collaborate without losing the flow.
But before we do this, let's quickly take a look how to set up Aron, install it, and run it with Docker because it's surprisingly simple. The first thing we need to do is go to the GitHub repository and clone the project. Therefore, we go here on code, copy the URL, go to a directory of your choice. I'm in my YouTube folder. I just click here in the command line, type cmd. Then we can just type get clone and the URL. Or you just copy it from the task list. It's linked in the description as well. And in my case, it's already there. So, let's quickly do additional folder. That was wrong. And do it again just you can see it from scratch. Okay, we can now list and we got the aron folder here. We go into the aron folder and then we can go on with the next steps.
We need to copy the env file example and configure our own env file. You can do this here with copy dot env.example to env or you just do it directly in the folder. can go here temp aron and you got here the example file just Ctrl C to copy it Ctrl + V to paste it and then you can just rename it. Yes, we want to rename and then you can just open it in an editor of your choice. So I have it here in Notepad++ and you can see all configurations here. Just change the highlighting here to YAML. And then we have to configure a few things here. So you need a superbase URL, the superbase service key. And that's everything you have to set. You can leave everything else as default. If you want, you can dive deeper. For example, if you want to change the ports because they are not free on your system, can change them here as well or change the lock level, but it's not needed for the first start normally.
So you have to grab your keys and the URL from superbase.com. If you use the cloud version, if you use the self-hosted version, you got it in your env file. If you don't know about that, check out my other videos where I showed this often enough I guess. So I don't show it detail right now. So for superbase.com, you can get it here on project settings, then go to API settings, and then you got here the service key role. Um, you can just reveal it and then copy. I will just do this and come back. Then we have to do it for the URL as well. You can find it here under project settings data API and then here's the URL. Just copy it, paste it as URL and then you got both service key and URL. Save it. And then we can go to the next step which is building the docker images and start the stack.
Back to our command line. We can now type in docker compose up minus d and then minus minus build because we have to build the docker containers first because that's that are not uh rebuilt containers you download from the docker hub. So we have to build them. You can also copy it from the task list. So, as you can see, it's now building the images. This can take a few minutes depending on your system specs. So, just wait until it's finished and then we will see it in our Docker desktop um starting up. You can see here I have a Aron V2 alpha which is the alpha version and not the beta we just checked out. So, we will get the new one here as well.
So while this is running we have to do an additional step in superbase. We have to create the tables and schemas and therefore we have to go to the migration folder here in the repository. Open it and then go to complete setup SQL. Open this and then click here copy raw file. So it's now copied. We go back to our superbase instance. Go here to the SQL editor. paste everything in and just fire it up. It will fail for me because I have I use here a project which has already their aron tables. So that's okay. But for you it will create all the tables. We can take a quick look here. So you will get all these archon tables. I have here another project as well for research and YouTube channel. you will get all aron prefixed tables which will store all our project settings and so on for Aron and now back to the command line we see everything built and started so we can now open or let's quickly go to docker desktop we see here aron can see everything up and running and we can now open here the front end so we can click here directly on the front end URL and port and we are in. You can also type it manually localhost 3737 and you get the aron start page.
The first thing you should do here is so you see here a few items in the knowledge base that's because I already indexed some stuff as said that's why my superbase tables were not empty but for you it will not show anything here. So you go to settings here on the left side and add your openme key here. This is needed for indexing crawling the knowledge base items you want. So you have some additional settings here on the bottom. So you can choose OpenAI or Gemini. So if you want to use Gemini, just put in your Google API key on top instead of OpenAI or put in both if you want to test out both. And you could also use Olama. It says here coming soon, but it's working, but I have to say it's not fully tested. You have maybe to do some manual adjustments for the embedding size, but uh it's basically working. I tested it, but a fully functional version will come in the future. I know someone from the Dynamus community is already working on it with a lot of more features and stuff and tested all embedding models. So stay tuned for this. I will do an additional video in the future for sure.
So you can do some additional settings like contextual embeddings if you want and can also do the parallel processing here the max count I leave it here with three and you can here do some other options also for the crawling performance. So there's a batch size max concurrent time outs and a rain delay I think the default is 0.5. Then there's also storage performance settings which I would say you can leave everything as default for the beginning and if you have problems you can adjust it later. So then here's an additional setting for code extraction because let's quickly go back to the knowledge base and to the existing items. As you can see here for the first one the superbase documentation I crawled you get also here these code examples which then can be used for your AI assistant. So the format is not pretty good for the superbase examples at the moment with these lines, but you get how it works. So it just ex extracts from the crawling the code examples and they appear here and can be used later to do proper coding for your needs. I would say that's definitely something we need to work on. uh for the official launch are goingon is still in beta and it's a little bit bad but it can be optimized and improved before we launch the full version.
Back to the settings you can do a few options here. So the minimum length of characters for the code blocks it and extracts the maximum length and some detection features here as well. They are all toggled on. So you can also play around here a bit and test out what works best for you. So here on the bottom you can also turn on bug reporting and if you find a bug you can just click here report bug and use this form to report the bug you found. So here on the top left we got also a few options. You can switch between light and dark mode. You can toggle the projects on or off. By default everything is on. Projects are here on top. the archon icon. It's project management. And this is the to-do list I mentioned in the beginning. So the AI assistant can use Archon for the project management. So as you can see here, here's an example for something I built and the clot in this case used the MCP server of Archon to generate these tasks and I can manually edit them here as well. that works also live while it is coding and you see something wrong here, you can just adjust it yourself and the AI coding system will take it. So you can also assign here between user arcom or AI IDE agent which is then the MCP server. You can also move around the task as you know from Chyra Asana or other project management tools. So it's a kind of Canva board you can work with. You can also see here on top the open to-do. So it's two task in doing seven. So it's in process and refue and done zero. Can pin projects if you want and you can also switch to a table view if you want to see more or if you want to enter task yourself fast. So you don't have to rely on the AI coding system to to do the task. So you can do them yourself. Just put it in, open them, edit them and write something.
So for creating new projects, you can either use the new project here on top and create it yourself as mentioned or via the MCP server. The coding IDE can do it completely on its own. So you have to do nothing. You just have to tell your IDE to use Aron for all project management and other stuff. So therefore you can go to the settings again and you see here the IDE global rules. So you can switch between cloud code rules. So cloud specific rules or universal agent rules you can use for curs or all other stuff you can find here. So you can see here the the directions as well where you have to put them. So you just copy them and for clot for example you put everything in the clot.md file in your project root. So we can take a quick look at aron itself where we can see here the cloud md where you would put all the rules in. So you could also put them in the global rules I guess for your users. So in the home directory so I explained this in this video. So that's also an option.
So the last thing we didn't cover yet is the MCP server itself. So it's this icon here on the left side and then you can see here the configurations for the MCP server and also the logs. So you don't need your Docker desktop CLI whatever to see the locks. You can do everything within the UI. You can also start and stop the server. You can see the status and also very important you can directly copy the configurations here for the different coding ids. So for example cloud code you just copy it and can use this in your cloud code instance. So this is more for not sure if this is correct. I think this one is you can use in cloud desktop and here the second thing is for using it in cloud code. So if you are in your terminal my case I'm on windows so I have to open my WSL with bash and then you can just do clot mcp and the rest of it and this will add it to your mcp server. So you can also check it with clot MCP list and then you can see all your configured MCP servers. I have a few Serena Context 7. I could remove this now when using fully Archon and some other stuff.
Back to the MCP servers. You can see the global IDE rules here as well. So it's the same as we just saw on the settings page. So you can either copy it from here or from the settings page. So as next step let's add a new knowledge base. So we go here to the knowledge base go to knowledge. Then we have a few options. We can choose between technical coding type or business type. So it's just uh category doesn't affect and you can also change it later. And you can also decide between URL or upload a file. Well, at the moment, you can really just uh upload one PDF, MD, doc, whatever, and not a full zip of a local documentation or a folder. But I started an implementation for multifile upload or zips, which was also working. You can see this here. So, I uploaded about 7,000 documents in batch from a zip file. It's not showing here at the moment because we cloned now the official repo and not my development state. But I have implemented here the state to upload the zip. I will show it here in the screenshot. So maybe this comes in the future into archon and you can use it as well.
So for the uls also implemented a proof of concept for a little bit more configuration for the ul. So you can um use include and exclude patterns or domains. So sometimes you just have a subdomain you want to stay on this or at least you want to maybe crawl the whole domain root domain but just index uh a specific subdomain. Maybe you are on the linkchain.com domain and find links for python. Linkchain there. This one you want to index but not java. Linkchain for example. doesn't make sense in this case I guess but you get it. So this is also not in but you can uh choose the crawl depth. You get also some description here what level you should choose for your documentation. You can try it out and see yourself depends on the documentation. And the most important thing is not because not just use level five because it's the highest and you think it will give the best results. That isn't the case which is also mentioned here in the caption. Then you can also give here tier AI text if you want or leave it blank and the AI will do recommended text itself. So we leave it blank. We go here with level two at source and you see it crawls link chain documentation. So web crawling first gets all URLs 55 as it looks like and will then start content processing source creation and then store the documents in the database. So therefore the embedding model and provider is needed in my case open AAI and as you can see it's now creating the chunks. So as we are only on level two it was really fast and we can now see the link chain API reference here. So we got 43 documents and yeah at the moment it's not possible to click here. That's thing I tested out. I did a proof of concept for a document browser similar to code examples so you can see what's actually indexed. I will insert a short B-roll here so you can see how it could look like if we implement it in the future in the official repository. So for other stuff, it would be really cool if you let me know if this are features you want to see in the official build. So that's also important for the for the Aron team to decide what to build in and what not.
So let's do a simple test and see how aron works with cloud code. So I created here a simple application PDF chat app. We will not fully implement it. I just want to show you how to configure it. So I will open the cd and then quickly open it in windsurf in my case. This will open the IDE and we got an empty project. To set up the project, we will use the contact context engineering hub from Cole, which is also built by the Dynamus community. There's also official one. I just cloned the internal one from the Dynamus community in this case. Um, but this is not the topic of this video, but you can let me know if you want to see more about this. And also, if you want to join the Dynamos community, you can find the link below to get a 10% discount. So I just copied the PRP templates here. You can see everything here with examples, templates, base and we also need the cloud code commands. We can also copy from the context engineering hub and we got the cloud folder now here with the execute PRP command and the generate PRP command.
So next step is in my case to go to WSL with bash. I'm on Windows. If you are on Linux or Mac, you don't have to do it. Just open your terminal, open clot. Yes, I allow it. And then we can start implementing our PP. So in this case, I will use the initial here. You can fill out everything which is mentioned here. There's also an example but I will keep it short and just remove everything and put a oneliner here. It's just for demonstration so you can see everything working with Aron. So the next step is go back to Aron and we need to copy the rules file. So we go to settings called cold rules copy rules. Go back to our application new file clot. MD paste everything in and we are ready to rumble. So I have to quickly restart. So I Ctrl C out of cloud restart it. So it lo so it loads all the commands as well and as we can see now and I type slash you can see the execute PRP and generate PRP. We start with generate PRP. I press tap and then I just do right click on the initial copy relative path paste it here and fire it up and as we now will see it will use the aron mcp server to plan the project. So in my case I have another MCP server Serena which do does some stuff. So we see this first to find files activate a project. So Serena will also create a folder here. So let me quickly do this on the left side and archon on the right side and open the project management here and we hopeullfully will see a new project in here. So let's just remove this one. We don't need I don't need this anymore. A quick while this is running because I showed this before but not again yet. You have to add the MCP server with the command I show you showed you here in the MCP servers under cloud code and then the second option here. So copy it, paste it to your bash to open the MCP to add the MCP server to your project otherwise aron is not added to your project and it will not use it.
So it now finished creating the PP. So generate PP and we can now execute it. So this is the new file it created with all the requirements and documentations, references, structure and so on it will need for implementing the app we want to do. So we got also a confidence score for oneshot implementation nine of 10 and some description why it thinks it's a nine. So let's just fire this up and we should see it creates the task in archon. Let's put it on the left side again. On the right side we can see live aron. Make this a little bit bigger again. And as you can see here, it asks me if it is allowed to use arcon have check. In this case, I allow everything here. So it also pulls the rules we set up in the cloud MD. Just have to load the other uh commands. And as you can see it now created the new project for our application here automatically. We can open it and see the kban board now. And we should see soon all the tasks appearing here. So I guess this is the command now for creating the tasks. And as you can see here it now created two tasks. One is already done in and in refu and the other one is in process. So now both both are done in refu. So you can refue them manually and move it to a complete or put them back in progress, comment them and do it again. As you can see, it creates more and more tasks and works on them as well.
So maybe one more quick thing here. As you can see here is also a docs tab. If you go there, you can add documentation here as well. So you can add a template feature template and add some stuff here as well. So this is pretty new to me. I didn't see this yet and I didn't test it. I think this was implemented by Rasmos just maybe yesterday or brought it to the project. So, as you can see, there are daily new features, new stuff coming to Archon, and you have to pull the latest updates to keep up to date. And that's what I said before. This the PL template stuff, architecture, and planning. Also can add meeting notes here. So, pretty awesome stuff. And you can also add new sources for technical knowledge or business knowledge here. Just add sources here from the sources we indexed before. As you can see technical knowledge, we got some and I could add here now the length chain API reference. I think it should have done this itself when we created the project. I think that's a bug we might have to take a look into because it should use for sure the link chain API reference because because it was there before we started the project or at least it should ask us if we want to edit if it finds it.
Okay, that's everything I wanted to show for the demo part. It was pretty basic and we can dive deeper in the future if you want just let me know in the comments. So what is the future and vision for Aron? So there are a few points which are already known we want to implement in the future or they are already there as we have seen with the PIP templates. They were also new for me but this was one thing. Uh a few other things are the custom rack strategies per project a full process visualization and also sub agents with defined roles. So Aron is an open beta. So I would would recommend you check it out yourself. Test everything and let me know in the comments how it works for you if you need more tutorials and also report if you have any bugs or features. And if you want to see to build a real project from start to finish, I would recommend watching the lounge party of Archon by Cole on his channel. I will link it here on top. So, it's on Saturday 9:00 a.m. Central time. I don't know if my video is before or after, but anyway, the video is already created and announced, so I can link it and you can watch it at any time.
If you'd like to join the Dynamus community, you can use my affiliate link if you want to get 10% discount. You can find it in the description. I would really appreciate to support the channel. And with this, I wish you a very nice day or evening. And I hope to see you in the next one. Bye.