Fable 5 Build: Agent Controlled Signal Map! Open Source

Key Takeaways

Hello all of you wonderful people. Welcome back to the Matt Vid Pro channel. This right here is my AI signal engine. It is an agent-powered signal intelligence map, and it is built entirely with the now discontinued Claude Fable 5. It's quite impressive, I think, technically being entirely vibe coded, but that is far from the only reason why this is actually interesting and special. Let me show you how this works. I'm going to press the clear button in the corner here to completely wipe this map or board, then click the demo scan button, and you'll see how this actually is supposed to work in real time as an agent filters through, researches, and considers AI news. They pop up live, and whenever a new connection is made, you can actually click on that specific connection, and you're going to get a connection insight between those two pieces of news. This is just the demo, so none of this is real information, but what this allows you to do is see the AI space from a large, wide, macro level, and then see the more micro pieces from that macro view, and see how they interconnect and interplay together. The whole idea is to give you everything at once, allow you to switch between different parts of the AI realm at a glance, understand why it matters, and then also enable filters to see key pieces of information at a glance. There are other features, and I'm going to show you them more in depth, but first I want to show you guys how to actually install and host this thing yourself. It is fully open source, so you theoretically could modify it for all kinds of use cases, toss it in Codex or Claude code, have it go wild. I'll also briefly explain why I built this. Honestly, I need it. Uh keeping up with the AI space is quite hectic, and I'm a visual learner. This is a phenomenal tool, and it gathers real insights, organizing and setting things up in a way that works best for me. A lot of you guys might say, "Hey, this looks familiar from one of your recent AI news roundups." and you would be correct. The granddaddy grandfather version of this, built by Codex for one specific week's worth of AI news, was definitely the inspiration, the starting seed for this project. But, this was static. It is not real-time live always-on infrastructure. Another reason that I built this is because I'm always trying to push the boundaries of what I can dream up and bring to life with AI technology. Whether that's a useful tool or a creative demo, but I really do think that this is another step forward on that path for me. Not just creating a custom tool, but pushing the boundaries on features, implementation. Let's talk about the system that makes this work. The whole separately hosted stack looks something like this. First, we have a main agent. I chose to build this around Hermes. In my opinion, it's the best open-source agent around right now, and it handles this pretty beautifully. News Hermes agent deployed on a Hostinger VPS via Docker Manager. It runs persistently, scans the AI landscape, reasons about what matters, and posts structured JSON events to this dashboard. It doesn't necessarily control the dashboard, it sends signals in. In order to keep everything always-on and private, we'll need a VPS, and especially one that has Hermes built right in. So, Hostinger is today's sponsor, and will be hosting our AI signal engine separate from my main PC, although it can run on a local PC or Mac. The website says Codex helped build the dashboard. Really, it was mostly Fable 5 until, you know, uh the event occurred and Codex helped finish up the rest. If you already have a Hermes account, there is a wide selection of AI models to choose from, or you can load in your own API, which is what I'll be doing. But, you need a powerful LLM under the hood to run those agentic tasks. GPT 5.5, the human-in-the-loop stays editor-in-chief. We are reviewing the connections, the information sources. It's set up to auto-create verification checklists, things that you don't want to overstate, and of course, you can remove nodes. Here is a visual diagram of the architecture. This is also set up to intake RSS feeds. So, if that's something you're already working with, it's definitely there. Anyways, Hermes agent sends in information, data is stored on the dashboard that's running. If you weren't looking at individual pieces of code, you'd think it's pretty simple. And guys, I would not be doing my job if I didn't talk a little bit on security and safety. There's a little bit of a reality check here. Autonomous agents, especially open source ones, are a little bit of a security risk. So, when setting them up, you have to be pretty responsible. Just running the agent on a separate VPS creates a cleaner boundary than running on your local main machine. This project is definitely built with safety and security in mind. Basically, you create your own little password, and you create a specific ingest token for the AI agent. Without the password, you can't see the board, and without the token, the agent can't connect to the board. Perms and approvals built right into the prompt. There's a visible log that can be toggled on the map. No destructive actions, no publishing, and verification flags on information when needed. If you want to just try this thing out with your agent, it's very easy to get this running on Mac and Windows. You don't need Docker if you're just trying it on local hardware. Docker is really only required for running this always-on through the VPS later. It's automatically installed in the VPS with Hermes and is the more permanent solution. So, with that context cleared up, to get started, you're just going to want to download the repo. You can do that up at the top of GitHub manually like so by clicking this download zip and then extracting the zip, or you can clone it with commands. So, once you download the signal engine repo, it's going to look a little something like this. There is one specific file we're going to be looking for, the .env file. We have to customize this file to set our Hermes token and the password to enter the signal engine. Right-click and edit the file with your preferred text editor, and you can see there's really not too much going on here. The app password, just type anything here to set this, and the same goes for the ingest token. Make sure you save the file before exiting. Now we're ready to actually launch the signal engine, and that's going to be different whether you're on Windows or Mac and Linux. On Windows, it is as easy as clicking this start. bat, which will automatically launch. But if you also want to see the agent visually connect and browse websites, on Windows, there's another launch browser.bat file that will open up a little headless Chrome to do just that. For macOS and Linux, you'll want to open a new terminal, navigate to the downloaded folder, the signal engine, and then run this command right here. If you don't have a Chromium-based browser, there is some additional instructions down here. This point in the installation, if you don't already have an AI agent, now would be the time to install one. Like I said, Hermes is really what this is built around. It's my favorite open-source agent available for Mac, Linux, Windows, fairly easy to install and use. For Hermes, specifically, launching it via CLI is as easy as writing Hermes directly in terminal, and as you can see, it loads right in. I've got this thing hooked up to GPT 5.5. I don't have a paid plan with News Research, so I'm using an API key from OpenRouter. It works through the Hermes agent as normal. There are different ways to access AI models through Hermes. Later during the Hostinger VPS install, I will actually walk you guys through grabbing an API key and implementing it because it is directly a part of that specific setup process. Let's sign in to the signal engine with our password. You can see there's nothing here yet. It is blank. In the top right-hand corner, I'm going to click settings. This is where we're going to access the prompt for the Hermes agent, but as you can see, there is a separate little copy-paste prompt section that works exactly the same, but it's more generalized if you wanted to try this with another agent of yours. Let's type the ingest token from the ENV file. In this case, it was Hermes Go, and copy the prompt. So, with that, we now have each individual piece ready to actually make the whole system function. In our Hermes terminal, we'll right-click, paste, press enter, and now Hermes is going to do a check to see if it can connect to the AI signal engine. Then, it's going to conduct some research and actually start to add nodes to the board. You can see, right off the bat, Hermes is preparing some skills. It's contemplating and ruminating a little bit. And, honestly, I'm going to full-screen the AI signal engine so you guys can watch it work its magic. In this user interface, I can actually hide a couple of the various tabs. So, this would be the agent log. It's directly connected to Hermes. The signal feed on the side that allows me to stack up and see all of the added nodes at a glance. And then, of course, the scan days. The agent checks what day it is, and based on the day that it's searching through nodes, it will create different maps. So, each day is a new map. You can see it is definitely browsing a few websites, Hugging Face, Reddit. It's grabbing some browser screenshots. And here, the board begins to populate with our first added node. ZAI releases GLM 5.2. A lot of you probably have heard about that. MIT licensed open weights with a million token context, seven verification flags on this particular one. And you can see there's all these little offshoots here that are basically entities associated with the singular news piece, Hugging Face, ZAI, the model itself, GLM 5.2. Let's see here. I wonder what Hermes is going to add next. So far this week, I can say has been a pretty big open-source focus. There we go. Fast context, a 4B Repo Explorer sub-agent for coding agents. It's like a unique little sub-agent release. We've actually seen a few of these lately, but you can see it has noticed these are a part of the same trend. If we want to figure out just exactly why that is, we can click it. It's a part of the same wave. Microsoft and ZAI, two independent signals pointing in the same direction. Both target long horizon coding agent workflows, but GLM 5.2 scales the solver model, while Fast Contests specializes in the exploration layer. And it's just added another competitor to ZAI Moonshot publishing Kimi K 2.7 code, 1 trillion parameter open coding model. Yes, they absolutely do compete with each other in the open source ring. So, it's great that it was able to, you know, pick up on that, and of course, give us this animated line, this ability to see it, click and immediately understand it. This gives us a little blurb why the agent linked these. Both are newly trending large open weight models aimed at long horizon coding and agentic engineering workflows. But, it's more than that, because they're on a collision course. They're converging on the same ground. Oh, and we have more additions coming through. It will populate the board with like 20 nodes. So, anyways, Google DeepMind releases Diffusion Gemma, Open Diffusion LLM. Oh, and that is connected to, yes, the Nvidia research tiles. Oh my god, everything's all connecting. But, you can see where everything connects and how it connects, allowing you to explore and understand it at your own pace. Agent Reach explodes on GitHub, zero API fee web research layer for agents. Yes, like I said, lots of open source related drops. The engine is going to continue to add nodes to the graph, but I'm going to click and show the signal feed here on the side. Like I said earlier, everything is sort of stacked up here, and you can click on any of these, and this gives you the information over on the left-hand side, novelty interest and risk score with deeper summaries, the verification claims, you know, things you don't want to overstate, the threads that it's connecting to, and you can bundle them together or choose to ignore them {slash} delete the node entirely. So, this ignore button is technically just going to send a message to steer. Removing the node keeps the topic allowed and just physically deletes it from the map. Anyways, what I would probably start doing at this point is begin to bundle these all together. So, these little select buttons on the side, I can start to click. Let's say I wanted to bundle all of the LLMs. All right, so Hermes hit its iteration budget. We've already got a pretty great view of the AI space this week. Definitely getting to that Fable 5 news. I've seen it add up to 20-plus signals. You could have this run on a loop. Maybe it clicks on three times per day. It would depend on how much you're willing to spend on the LLM inference to add to the board. Like I said, this was really built around my needs, so it's creator-focused. You can generate a brief. This sort of automatically gets you a bunch of information compiled by the system. An executive summary shows what the board is being dominated by in terms of topic. So, models, open source, and agents. Absolutely the biggest ones right now. It gives a basic creator angle. Tells us whether signals carry elevated risk. It even comes up with a few title ideas. They are mapped to the reality of the AI space right now. It also creates a few story bundles, which groups of connected signals are the strongest. Six connected model signals, of course, that we bundled together ourselves as well. Google research safety, talking points, the full verification checkpoints, and risky claims that it was able to pick up, which are lengthy. Uh you can see a lot of these are hyper-specific as well. So, we have them all up front. Honestly, exporting something like this would be best to send off to another agent for further breakdown. You can export it as a markdown file. To make it less confusing, you can also toggle off the entities, which definitely helps you see just the connections and the nodes at once. There's definitely more room to add new features, but that's why I released it open source, so you guys can modify and change it in any way you want. I think you get the picture on all this. Let's talk about hosting it on Hostinger. If you wanted to cement this into your workflow, logically have it queue and run at set points during the day, it's more permanent, it's more safe. Today's sponsor Hostinger can host an always-on VPS that already has Docker and Hermes built right in, meaning we just have to load up the signal engine, put a little bit of information in, and it's off to the races. This is affordable and without a doubt the easiest way to go about building your own hosted agent setup that has a real home separate from your local machine. Unless you've got a dedicated machine for an agent, this is the practical long-term solution. So, when you're setting this up, use my link down in the description below plus the coupon code for a discount. I have not tried the non-VPS managed Hermes agent through Hostinger. I assume you can still get this running just fine on that, but VPS is definitely going to be the way to go for this particular project. You're getting the full control, the least resistance to installing this. Choose your billing period depending on how committed you are. They do have monthly, but of course you're getting a discount per month if you go with something longer. You can also use code MattVidPro on the 12-month and the 24-month plans for a little bit of a discount. It is a hosted virtual computer, so you can use it for all sorts of projects directly related to AI. You can also get some AI credits boosted directly for the Hermes agent built right in. Honestly, you don't need to go with either of those. You can if you want, and you could probably use them with the setup, but if you already had a Hermes plan for example, or you're like me and using an open router API key, they're not needed. Operating system and any additional tools to install are up to you. Just make sure Hermes agent is selected at the bottom. Docker will be installed automatically, and the plain Ubuntu that they give you is just fine. Once you snag a plan for your server and give it about 5 minutes to boot up, the home panel or HPanel in Hostinger should look something like this. From here on the right-hand side, we're going to click the manage button, and clicking this will bring us to the server overview. Also have this little manager button over on the right-hand side, so we're going to click that. Up at the top right-hand, you'll see a terminal button, and you'll be taken [clears throat] right to the Hostinger terminal. And right through here, we're going to execute the commands to download and install the repo and get the Signal Engine up and running. The included deployment.md file, also able to be read on GitHub, gives you the full Hostinger VPS install guide. Putting the code on the server is super easy, though. You can see in step one, I'll go over here and copy all this code, right click in the terminal and paste, press enter. It's going to go ahead, download and install the Signal Engine repo. Next, up set our two secrets. These are our security tokens to get in and allow Hermes to connect. Do the same thing, copy on the side, right click and paste, press enter. And now through the terminal, we're able to set everything we need. In order to navigate this menu, you're going to have to use your arrow keys all the way to app password. Type whatever you want in, same thing for the ingest token. In the terminal, hold the control key and press O, then press enter, then press control X, and you'll be brought back to the original terminal. It's worth noting it's probably a good idea to go ahead and port forward port 3000. All you have to do is go to your firewall settings, add an accept TCP protocol for port 3000. From there, we want to move to step three, actually launch the engine, same way we've been doing, paste in the command line. And once this is done booting up, you should get these nice green check marks. And if we refresh on the Docker Manager, we should see it as an additional Docker project. So, if you actually want to go ahead and open the Signal Engine up, under access, all we have to do is press open, and you can see it brought us right to the engine. Type in your password. Keep in mind, we still have to use commands to set our Hermes API key. This can be pretty annoying, but this would be the general command to actually set an API key for Open Router. Hermes config set Open Router API key and then your API key. It has to be in quotes. From here, we can boot up Hermes normally by just typing Hermes and then pressing enter. Now that Hermes is fully booted up here, this is the other command to set the model to GPT-5.5. Honestly, you can't rely on ChatGPT or Google to find some of these commands and expect them to be correct. They must be changing all the time. I kept getting incorrect commands from ChatGPT. This is what's working right now for this particular date on this video. Anyways, that will set 5.5 to be the global model. From there, back to the signal engine, we want to copy the local IP address and port just so Hermes has that context. And of course, copy your prompt from the settings, paste it in, and watch Hermes begin to fill up your AI signal engine. It definitely will take some time. I'm seeing typically 15 to 30 minutes. And if you trigger this, let's say three times per day, you should get a very, very robust collection of news pieces and insights. Yes, what looks like to be the main story of this week has popped up yet again in our signal dashboard. Well, I'll continue to let this run in the background. Thank you guys very much for watching today's video and tutorial. If it helps you out, I really would appreciate a like on the video and a subscribe. This is definitely one of the cooler projects for me in recent memory, only being topped by that short film. useful. I'm trying to actually use it in my workflows, and I'll be interested to hear what you guys have in terms of ideas to maybe build or expand off this. Do you have your own separate agent device to host this yourself and run it 24/7, or are you going to go with the VPS method like I've got right here? I've let this set, and it has collected quite a lot of AI news. It actually does get laggy when it gets more bulked up like this. But yeah, I sincerely appreciate your viewership. Thank you so much for watching. Thank you to Hostinger for sponsoring today's video. More to come and goodbye.