Über dieses Video
Privatsphäre ist der ultimative Luxus im Zeitalter der KI. Dieses Video führt Sie durch die brandneue Adysius-Einrichtung unter der aktualisierten URL und enthält die Ankündigung des Ajax-9b-Parametermodells. Erfahren Sie, wie Sie lokale Agenten einsetzen können, die Ihre E-Mails lesen, im Internet suchen und sich Ihre persönlichen Ziele merken, ohne jemals Daten in die Cloud zu senden. Die wichtigsten Erkenntnisse: - Umfassende Anleitung zum neuen Adysius.dev-Repository und zur Docker-Einrichtung. - So konfigurieren Sie das Ajax-9b-Modell und integrieren Olama für lokale LLMs. - Einrichten von KI-Agenten für die Verwaltung von Gmail, die Websuche und das Kalendermanagement. - Nutzung der „Brain“-Speicherfunktion für ein personalisiertes KI-Erlebnis. - Tipps zur Leistungsoptimierung auf dem Mac mit MLX und dem Warp-Terminal.
Der Datenschutz ist dahin, es sei denn, Sie führen Ihre KI lokal aus.
Die Ära, in der jeder private Gedanke an eine Unternehmens-Cloud gesendet wird, neigt sich dem Ende zu. Mit dem neuesten Update für Adysius und der Ankündigung des Ajax-Modells – einem leistungsstarken Modell mit 9 Milliarden Parametern, das auf Qwen 3.5 trainiert wurde – liegt die Macht autonomer Agenten nun offiziell wieder in Ihren Händen. Wenn Ihnen Ihre Daten am Herzen liegen, ist es an der Zeit, Ihre Abläufe auf einen lokalen Rechner zu verlagern.
Die Rückkehr von Adysius
Adysius ist unter der neuen Adresse Adysius.dev zu finden, wodurch sich der Einrichtungsprozess auf wenige einfache Befehle vereinfacht hat. Ganz gleich, ob Sie einen M5 Max mit 128 GB RAM oder eine bescheidenere Konfiguration nutzen – die neue Architektur ist auf Geschwindigkeit und Zuverlässigkeit ausgelegt.
Wichtige Schritte bei der Installation
- Das Repo klonen: Wechseln Sie in das Verzeichnis Ihrer Wahl und führen Sie den Befehl „git clone“ aus, um die neuesten Dateien abzurufen.
- Umgebungskonfiguration: Kopieren Sie die Beispiel-env-Datei und passen Sie Ihre Geheimnisse an. Stellen Sie sicher, dass Sie ein Administratorpasswort festlegen und die Olama-Basis-URL auskommentieren.
- Docker-Bereitstellung: Verwendung
docker compose up d buildum Ihre Container zu starten. Das übernimmt für Sie die gesamte aufwendige Arbeit im Zusammenhang mit den Abhängigkeiten.
Mit Ajax und Olama durchstarten
Während das Ajax-Modell derzeit für die Veröffentlichung vorbereitet wird, können Sie andere Modelle mithilfe von Olama sofort integrieren. So können Sie Modelle wie Gemma 27b oder spezielle MLX-Versionen für den Mac direkt in Ihren Arbeitsablauf einbinden. Durch die Einbindung dieser lokalen Modelle erhalten Sie die Möglichkeit, im Internet zu suchen, Ihren Kalender zu verwalten und sogar Ihre E-Mails zu automatisieren, ohne dass Ihre Daten jemals Ihr Netzwerk verlassen.
Die Zukunft des lokalen Gedächtnisses
Eines der beeindruckendsten Features dieses Updates ist das „Brain“ bzw. das Speichermodul. Während Sie mit Adysius interagieren, baut es eine lokale Datenbank mit Ihren Zielen, Vorlieben und Ihrem Verlauf auf. So entsteht ein personalisiertes KI-Erlebnis, das es mit Claude oder GPT-4 aufnehmen kann – und das bei 100-prozentiger Datenschutzsicherheit.
Es ist an der Zeit, nicht länger ein Produkt der großen Tech-Konzerne zu sein und die Kontrolle über die eigene Intelligenz zu übernehmen. Die Einrichtung geht schnell, die Tools sind kostenlos und die Ergebnisse sind bahnbrechend.
Transkript▾
Pewdiepie's Odysius is making a comeback because he's just announced that he's training his own Adysius specific model called Ajax. We can't actually get access to it right now. He's going to be releasing them as ready, but it's trained off of Quen 3.5 9b parameters. But given all this, we're going to the latest and greatest setup given the URLs have changed and a few things have changed on the platform. So, I'm going to update my previous video on that. So, without wasting any more time, let's get on with it. So this is on a new URL. So I'll make sure I'll link that down below. It's on Adysius Dev now. And scrolling down, they've really simplified the startup. It's literally just these lines of code here. So you want to clone the repo or you can get it up here as well. Just copy that. Now inside the terminal, I have done a master class on the terminal and that's linked down below as well. Runs you through loads of basic commands. It is a long video because there's a lot to learn, but if you want a deep dive, if some of this intrigues you, I definitely suggest going and checking that out. So, what I will do is I will go to my sites folder here. This is where I like to install and run all my website applications and stuff. You might just want it in your documents folder. I like it in my sites. Oh, and by the way, I'm using warp, which is a very, very helpful terminal for Mac. It 2 is another great terminal for Mac. Anyway, [snorts] so now we've cd inside of here. We're going to run git clone and then paste that URL there. It's going to create a folder called Odysius there. And it's just going to download all the files that are on that git repo. And if we cd into that, we change directory into that. Let's go back to the instructions. We're going to copy the env file. So copy the example into a new file called env. So you can see it's copied there. Now, this is a secret file that's going to contain all of the secrets that you might need. So, your open AI key as an example, and these are kind of settings. You can change the port if you want it to run on a different port there. Now, what I'd do is actually uncomment this line here, Adysius admin password, and then put in a password that you want to use. What you also want to do is uncomment OAMA base URL. And that'll make sense in just a second. Save it. Close it. You need to install Docker. That's super simple to do. Literally just head over to Docker's website and download it. It's free. And run docker compose up d build. Apologies, I'm not even running docker. So, we've got it up and running. I can just press up and it's going to download all of the things it needs, the packages it needs. What do you think he's thinking about right now? Smooth brain of his smooth egg. Great. And that's all done and everything's healthy. Now I had it when you visit localhost 7000 I actually get this. So it's actually better to visit under 127.0.0.1 7000. Here we can paste the password we created and we are in. So if we go to what is 2 + 2 we actually can't do anything. So if we go select model plus or you can go to settings. We actually add our own models here. Now when Ajax becomes available, we will be able to download it. I'm not exactly sure how we will be able to download it, but I'm guessing it'll be on hugging face. And whether it will be on MLX or GGF, we don't know. If that's confusing to you, don't worry so much about it. But the easiest way to get models running with Odysius is using Olama. This allows us to download and install models on our machine here. Some of them run just on the cloud, so they don't actually download to your machine, but if we go to Gemma as an example, these run locally, and in this instance, they've actually got MLX versions. Now, I would suggest a mixture of expert model. I'm going to download Gemma 426b. Inside of another terminal window, we can go Lama pull Gemma 27b. And this is going to take a moment to download. Smooth bra. And that's downloaded. Here we just go on the three dots add llama. It's going to pick up the URL that we uncommented. We've added that and it's picked up our models. It's picked up that I've got Nimble and TEV1 downloaded. We're going to use Gemma here and say, 'Hey, now how well this performs completely depends on your machine. I've got an M5 Max with 128gb of RAM. It's a very capable machine, very capable model. You can see we've got a response here. Your mileage may vary. I can't know what machine you have and what model is best to run, but we're getting a result here. Cookbook allows you to find models that are good for your machine. It does a very good job. However, it does run Llama CPP. I'm on a Mac, which means it's not particularly great for running these types of models. MLX is ideally the models that run on your Mac. This isn't a hard and fast rule, but I feel like we're getting too much in the weeds here, but you might want to go into QuickBook and just discover what models can or cannot run well on your computer. Now, something I want to try here is I'm going to use Whisperflow here. [snorts] go to ebay.co.uk and find out if there are any Mac Mini M4s available right now. And because this is an agent, I'm hopefully this will go to ebay itself, navigate to ebay, search for Mac Minis, and give me a result on that. So, we're going to allow for this chat session. Again, we're going to progressively gear towards trusting the AI. We're not just going to get all access all the time. Okay, it looks like eBay has some security in place. Let's go. Let's go to samuel gregory.co.uk and tell me a bit about them. Go. So, it's gone to the website. It's telling me a bit about it. To get the most out of Adysius and many of these things, you want to set up your email. You want to make sure that it's got access to your email. And I know that sounds daunting and and it should be really to be honest. But these this is the this is the point of all of this stuff is that you're letting local models do things with various aspects of your life. Your calendar as an example. Now I've set up a Gmail integration here for my emails. Obviously your information will be different. Just know that if you're using a Google one you need to generate an app password. I can say what was my latest email and actually action that. I could respond. I could use the information from a Google search to draft up an email and actually send the email all through Adysius. Okay. Security alert. And that was me generating that password. One thing I do like as well is the brain. It's the memories building up over time adding things as you use Adysius. It starts to remember things about you. and it can recall them or use them in reference to questions you ask it. You know, say you've got some personal goals that you want to tell it all about that you have some financial goals that anything you talk to it about that kind of thing, it will take that into account and respond. That's how kind of I use the memory feature on Claude. You've also got this document creation feature as well, which is really cool. You can access those in the library. And there you go. That's Adysius updated to use the latest versions. Obviously do an update to this video when Ajax becomes available, but ideally it'll be a similar kind of process. Like, subscribe if you haven't already. Until next time, keep on vibing.