chapters transcript notes
click any line to jump to that moment in the video
0:14 All , I'm here with Matthias. This is our I guess second episode here. Hello Matthias, how are you doing? Hi, good to see you. Good to see you. this has been fun. This 0:24 just happened by chance. We had talked yesterday around a tool that you've been building. This very with our series that we've been talking about. I'm just starting a brand new podcast around 0:35 Agentic level things, exploring where agents are going to be used, how do you use them with Power BI and development and fabric things. And we just touched base what 2 days 0:45 ago and we're this probably makes a lot of sense. We should talk about this. For sure. And to be fair, we were I mentioned this. I thought we were doing appropriating and I were going live. that's that's really 0:56 [laughter] I it I just the energy and the enthusiasm. Let's just go. Let's just make it happen. excited about this. All , let me just give you a little context 1:06 here. one, we're introducing a tool called BI Buddy and this is going to help us adjacently build things with Power BI with an agent, which is really good to to to have in our toolbox 1:18 as a tool set. But before we go there, I want to give a little bit of history of where Matthias and I really started connecting on this AI thing and I got to give mad credit to Matthias 1:28 here because we were at remind me the place we were at Budapest. Yeah. we're at the Budapest conference speaking around Power BI and 1:38 and and things and we happen to be at the dinners for the the dinner for the speakers lunch or speakers dinner I guess it would be called. And we were just talking about things and we were sitting across from each other talking 1:48 about stuff and probably me or someone else brought up the AI. I'm oh I'm all about AI and all this and started figuring out what is it all about? 2:02 Then you blew my mind and you you you let me show you what I'm building and you you brought out this app and I had heard about it but I haven't played with it and you started showing me lovable and how lovable was you were 2:12 building these things with app registration, you were doing admin management things, you were building all these really cool things and you're it's fun. Tell me more about what you 2:22 were doing there. I've I've been running many experiments and I've honestly I was trying to push the limits of what lovable could do because it had this notion that lovable is for websites 2:34 and for basic stuff but I I questioned and realized that it can do a bunch of technical stuff. you mentioned it. You can log in through your Microsoft account. You can use SSO. 2:44 You can do a proper back end. It has its own cloud environment and we're not doing advertisement for lovable here but it it's cool for quick prototypes. Yes. Of course I'm also using other tools. Of 2:56 course also GitHub Copilot clone code and these things but lovable is still my gateway into getting a web app out really really quickly and I did a bunch of cool stuff. I've 3:06 blogged about my my AI council where I had multiple agents communicate with each other. I've made I've made even games for workshops on on lovable and a bunch of other other 3:17 cool things. It's it's really really quick to get started there. we're talking about this and you're sharing this thing with me and my mind's getting blown. I'm what the heck you're doing? 3:27 This is cool. I'm as a as a tech nerd person I was I'm there with you. I'm this is neat and then I'm rethinking how is this going to change what I do, what I work on, what 3:38 does my team work on? , when you can go from one person's idea into a functionable function functional and workable app within hours, minutes, 3:49 depends on what you were doing, which this usually would take longer periods of time, weeks or months to develop. This was just really compressing the amount of time it took me to develop these things. one, I was very pleased 3:59 to hear from you and and and teaching you this. Another thing that was happening at the same time not related to the conference was we were on the very beginning of the wave of really good AI models 4:10 to appear and I think this is when Opus was about to be announced, Sonic was getting much better, ChatGPT was getting its new version of of software out and before we 4:22 talked in December I think that stuff hit. Yeah, for me that was January. I remember I remember specifically talking to you and others about I I see this as 4:33 an army of junior developers that I can I can make them do things but they they they it feels they started at the job yesterday. They fresh out of school. They didn't learn how to do. And then 4:45 January hit and I'm honestly to me we were at the it's the senior developer. They still are new at the company I have to teach them how do we work here, what's our best practices, etc. etc. 4:56 etc. But the skill level is is senior level higher. Yeah, way higher. And the and while we were at the conference and this is what really hooked me in here, , you're 5:07 telling me about these things that you're building with AI and I said I had this idea for this game and I was describing the game we were playing with my family and he's wait a minute, let's let's make an app around it. I 5:18 described the game and and the game is my my kids and I were playing this game. Someone found it on a TikTok or a video or a short or something, 5:28 YouTube short, where someone was you would record into your phone, you would sing a song, ? You , whatever the song is, , love love me do a Beatles song, ? 5:38 And then what the song what the app would do is it would take the the lyrics and it would flip it around and invert the whole song and play it in reverse. And the the game was , that 5:50 you hear the song that was sung in reverse you got to sing the reverse of the song. you have record yourself singing the reverse and then play it back normal to 6:03 see how close it sounded the real one. And it was just a really fun silly little game. We played it for 30 minutes. We were laughing. And it took one prompt. it wasn't one of those have to do. It was one 6:14 prompt, 2 minutes and we had our game. Boom and it was incredible and I was oh my word, this is cool and I did and this is this is what we were talking about 6:24 and the discussion was if you have an idea and you have a tool with an agent that allows you to supply the design pieces behind this, I'm this 6:35 is this makes sense. This is going to be the disposable app era. If I see an idea and I can get a prompt down, anything that's too simple, it will immediately become you there's no proprietary 6:48 For sure. at the point that sometimes I make single use apps. we yes, there will be single user apps but I even have single use app , I 6:58 need to convert this PDF file to or compress it or whatever. Do I want to search the internet and and hit up some shady website with a bunch of commercials or do I want just want to 7:09 create my own single single use tool? The problem is that then the next time I have to do the same thing, I don't necessarily remember that I made the tool. Of course I will create it again it becomes really disposable 7:20 you're saying what what single single use applications even. But you know this game, it's one of the lovable things I've made that I've opened the most times. And my son will 7:30 still remind me oh shouldn't we play this game again? it was a hit at home as . Yeah, totally. exciting. , good. All , this is awesome. Very very fun. if 7:40 you do me a favor, there's a chat window down here for you Matthias, you and I. Does your lovable app get can you share it? Is it a shareable app that goes out publicly or does it have to be only on 7:51 your account? Yes, good question. Don't worry about it . Maybe we'll get it later. We'll if we can figure it out, we'll put If there's if there's a way to share it, it'd be fun. We'll put it in the show 8:01 notes if we get it to later. But if anything else, super fun, very exciting thing. , let's let's move on from this. , we've we've done this fun thing. AI is 8:11 is commandeering a lot of our world. We're thinking about it a ton . , let's fast forward . I start seeing things on LinkedIn about BI Buddy. let's step in here. What is BI Buddy? 8:22 Give me the high level version of this and maybe can we do a demo of it? Can you show us what it does? For sure. For sure. And and I'll I'll I'll start at 8:33 the beginning of where I got the idea. Yeah, love it. I was in Netherlands for and I cannot pronounce this if anyone from Netherlands is watching this, I'm sorry. 8:44 [laughter] On the Power BI Gebruikersdag, which is the Power BI user group day. that's a an and I think it's the largest Power BI only conference in Europe or even maybe 8:56 even in the world largest in person. 3 4,000 or attendees or something. It's really it's really cool. Just for Power BI. It's 9:07 amazing. Yeah. and we had this full room of of 70 80 people who were doing this mastering AI in Power BI course that I did together with Mark Mark 9:17 Lellewelt. Cool course but one of the feedbacks that we got was that or we realized was that the hurdle of even opening up VS Code or 9:29 a terminal or something that was was for a lot of the people in the room was a big blocker. They've never used an IDE before. They aren't used to this workflow. They just 9:41 work from Power BI and they would rather have UI than they would have something that smells code. And I realized that , there's a huge gap here. I mean the AI adopters are moving very 9:51 fast and and that gap is widening because we we've gated the tools where we can do great things with AI into 10:01 tools that are developer native. And and and I realized that that's that's problematic and I there's a I I wanted to see if I could do 10:11 something to bridge that gap. That was that was the idea. Can I Can I bring that was the the initial thought. Can I bring the Power BI MCP, 10:21 which is really powerful and really cool, but it exists in VS Code or some somewhere else. Yes. Too technical. Too technical. Can I bring that inside of Power BI Desktop? That was my 10:32 my initial spark. Yeah. And then quickly I realized it Of course I cannot because I'm not Microsoft and I cannot change Power BI Desktop, but I can do the next best best and put it 10:42 inside the external tools of Power BI Desktops and make it as a user experience really really streamlined and really really seamless, which I will also show. that was kind 10:53 of the I want to help bridge the gap for the people who are a little bit where it feels daunting to start going directly into VS Code and you have to do 11:03 bunch of folders and markdown files and even though the principles are simple, the UI is is is really frightening to be honest. And I remember because I've been there before I open an IDE the first 11:15 time. it's hard to remember once you get used to it, but that first time of all those windows and options and it's really not created for non-developers. Correct. 11:25 Yeah. I wanted to bridge that gap and I started of course working on some quick prototype and that prototype suddenly excited me it's it's exciting me to do more and I became almost a 11:36 little obsessed. I'm I'm sure if you ask my wife she'd said I spent many nights and many restless [laughter] Oh, I'm just going to make this feature 11:46 and then boom seven seven hours later I'm going to bed. Six more features have been added. Going to [laughter] bed at 4:00 a.m. because oh no, I also wanted to do this and that then I felt I need to do a little bit of 11:57 of unit testing because things were getting sluggish and and suddenly yeah it it slowly and steadily become an actual project where I also hit some of 12:07 the the barriers of how do I turn this into something that's not a prototype, but something that's an actual solution and suddenly I had to think about things how do you get people to download 12:17 it? That's that's harder than you'd think. Way harder. And a little bit about marketing to write the emails you don't go to spam folders. You have to know the links and and 12:27 and this is a key learning for me. Don't put EXE files directly in the mail. [laughter] 12:37 You have to know know these things that that there are there are actual patterns to to get this this first step done. And then the next step is getting people to install it and that's also harder because the first you'll see is you create a new 12:48 application people download it. When Windows Defender goes Whoops. Nope. Yeah, too new. Yeah. And then oh . we had to learn 13:00 how do you do this official certification and and and signing of install files you see you check the the this this certificate of my install files it will say coming 13:12 back to my identity and stuff that and that's a whole process. you have to go through official channels, submit your passport and stuff that to be able to to give the stamp. Yeah. 13:23 All the steps of the way that were just this this from prototype to solution maturity that I had to go through. Security obviously. How do I handle that 13:33 enterprises will not be happy if I do any logging and I honestly admit that the app very privacy minded. Yeah. but I do need to to see how much 13:43 people are spending on my API account because that's my my credit card. But if a company goes in and add their own API key, I'm not tracking anything and that's 13:55 It's just local app connecting directly to their their own API provider and I'm just out of that loop completely. I have a back end, but only use it in in certain scenarios. All of those things 14:06 Yeah. I had to do and and still I'm there where I feel this is something I feel is mature and I want to show it to the world. I'm still in in in closed 14:17 beta, but at the moment I'm I'm if anyone's listening I'm these days this this week I'm accepting everyone who signs up and 14:27 fast tracking them for access. if you're sitting here watching thinking this is cool, I would to try it. Just go to bibuddy.ai and I'll I'll fast track that that that 14:37 beta access. Awesome. You heard it here first. This is the the the release party inadvertently on a real live [laughter] stream. no yeah yeah. 14:47 if you to be I'm going to call this very clearly in the description below we have a link for bibuddy.ai. in the description of this video if you want to go check out the product that we're you're going to see , go check 14:56 it out there. Also, if you want to go look in the chat window as on YouTube and Facebook and on Twitch, we also have the link directly there as 15:07 . if you if you this product, you what you see here, there's already two links available for you already. Check them out. I've also linked Matias your LinkedIn page as . we have that for you 15:17 additionally and if you want to go get more information, if you want to go follow his page or reach out directly, go follow him on LinkedIn. You're going to learn much I have and I think 15:27 we're ready to dive in. Ready to screen share? . Yes. All , over to screen sharing. Yeah [snorts] yeah yeah. it big here. There we go. I'll start 15:37 with of course Power BI Desktop because that's where that's where we that's that's the the user group we want to to help here. we're in Power BI Desktop. You're looking at a 15:48 very rudimentary report. Doesn't really matter. There is a mess of a data model. It's a simple model, but it has a bunch of broken measures and you'll see there's also a 16:00 role of security that's not working. this could be for a non-AI user a pretty pretty Fair bit of a work. Yeah. Fair bit of work to to to start doing. 16:11 Yeah yeah. it's a good use case for something the the BI Buddy. as I said it's living in in external tools. Once you've installed it, it will automatically put itself here along 16:21 these other great tools. and you just go and you click here. if anyone's ever played with MCPs before, that MCP is really powerful, but even just getting it to 16:31 connect to your model can be a little tedious. You have to get it to connect with chat via chat. you have to say, "Hey AI, can you please connect to my local model?" And it goes , "I 16:41 don't know what local is." And you have to [laughter] my Power BI Desktop. I've seen that. Yeah, for sure. what I did what I did Mike is I kind 16:52 of said, "How can I make this as seamless as possible?" And since we're opening from Power BI Desktop, I can get that data in launch time. I'm I'm mounting the connection for the MCP as we're opening 17:04 the app. You don't have to ask the the the AI to do it for you. Oh, wonderful. Yeah, just clicking BI Buddy and it's opening the app and you see here it says 17:14 connect to model, but it's going to switch itself in a few seconds and boom it's connected to this local model. I found it. , excellent. Found it. It's connected. I can go and start saying simple things 17:26 rename my column in sales orders. Oh, there's a bit of spelling error here. It should be fine. Total dry items to 17:38 total hot items. I This doesn't mean anything to me, but it's a it's a it's a case here. . And Yes. This is a normal chat 17:49 interface. This is what we would usually use if we had Copilot or we had another chat experience. It's just it's a window where you chat to it and you talk directly to it and it gives you immediate feedback and it starts 17:59 thinking and doing things and you're you're not talking to a a large language model that's then able to then see your your Power BI data model. you'll see here that 18:14 Yeah, I'm just and as Look at it. It's already doing things. You see here it's changed the column name live. You can see it immediately in Power BI Desktop opening and it's Yep. Sure. 18:24 It's it's it's just changed That's an that's an easy thing. I also say, ", let's let's throw a big task at it and say, 'Hey hey AI, I have these bunch of measures. They aren't working. Analyze 18:34 them all, locate the issues, determine the easiest fix for all of them and then just do it. I don't want you to ask me for approval. I just just want to do 18:44 do the work for me. The thing is if I don't the change, just decide not to click save. Close my app, open it again it won't it won't it won't matter. but at the end 18:55 of the day probably should also want to consider having a some diff thing in in mobile as . it's it's found this going to get the full definitions. 19:06 Behind the scenes it's calling all those measures, getting the definitions and a bunch of metadata. It It found a core pattern, some legacy 19:19 set of measures. Here the references name that doesn't exist. I'm just wanting to validate its own work checking that , I fixed them, but let me just re 19:30 them and see that they're are fixed and that nothing are left. I know you see there's some DAX relations. it's calling 19:40 measures to see what data to retrieve to to validate its own work. And then finally it'll give me a summary of I This is what I've done. This is the problems I found. This is what I fixed and 19:50 cheers. Yeah, all that in few minutes. and honestly I already installed this app, but the the installment is as smooth. You download the file, you install it, boom, 20:00 it's in desktop and you're you're ready to go. I really have tried to make this as seamless Yeah. a lot of the let me just I get a 20:11 little bit more context for myself here. just mechanically what's happening, you've got BI Buddy sitting here on the -hand side. You're talking to directly the connection to the model. 20:22 You're doing you're using the the XMLA endpoint locally on the on your desktop to have it communicate and then make changes and edits and you're using the MCP server to do all that for on 20:34 behalf of the user . Yes. . I'd be more than happy to dive dive into some of the more technical details of of what we what what what what we do at engine level, but I want to show something some other 20:45 cool feature before we go there. Yeah, 100% This was seamless. You opened it from desktop and and connected to this already. But if you had if you had a problem with multiple desktop 20:57 models open, you could go and have it scan the different models and choose a different one. you don't have to connect to the one you open from desktop. You can also connect to others. I've never seen that before. No, no, but 21:08 there's more. There's Wait, wait, there's more. I love it. there's more. You can connect to a local PBIP file if you want to. you could connect to my Oh. 21:18 I could connect to my my PBIP folder here this Lego model. I could connect to this one instead and I could ask list my table names and it should 21:30 connect to Yeah, it's connected to this other model and it will come back with some Lego oriented table names. Wow. . Yeah. And this is the table names and yeah, 21:41 there's a an ugly date table in there in the mix, but yeah, it's it's connected to the Lego table instead. I could also choose the Fabric or the Power BI service connection. Wow. . Yes, yes, as I 21:53 said, I've I really I really put some up late. I can see this the 4:00 a.m. has been paying off here. [laughter] When you click here, you have two options, Mike. You can either just ex- 22:03 accept this as an app registration on your tenant and suddenly you can browse your stuff from within the app body. I can choose my workspace, I can choose my model and I 22:13 can just seamlessly connect to the sales model with with a few clicks. Oh, this is slick. I really this the simplicity of connecting to things is really a secret sauce here that's really 22:23 nice. You can easily get in and it it'll go into the service. Oh, of course. This is great. Yeah, thank you. I need to pull up my 22:33 phone for a bit because MFA. this is Oh, yeah, that's . It's all . It's security. It's good secure stuff. just make sure it's not going on camera, ? [laughter] 22:46 Not on camera. Not on camera. Not on camera. it's it's it's locked in and it's connected to my Fabric sales model. And if I were a user who didn't want to do it this way, I could also have chosen to type 22:57 manually and do it the old-fashioned way where I have to do the exact name, but I just personally the browsing experience better. Yeah, 100% agree with that one. 23:07 Wow. Yeah, try to make it as easy for you any need you have PBIP, Fabric hosted or Power BI service hosted models or local Power BI desktop models and 23:18 connect to all of them and then you be able to do the same operations. you asked about engine level. I I think I was focusing more on just this is all focusing on engine level. Is there anything you have and I'm not 23:29 trying to add more features to you here. I know you're trying to get out the door already, but I do see there's a couple open spaces on this visual report layer here. Because you're looking at the PBIP, do you have some Is there some 23:40 report side of this or is it only or is feature focus only on the model? ? Again, giving you some leeway here. Yeah, yeah, yeah, yeah. 23:52 [sighs and gasps] I'll I'll show you something since you're asking [laughter] this question. That is I honestly, I'm not promising that this will be included in the 24:02 product because this is really really a complicated really a complicated space. the process for Yeah, the process for for for doing what you call it 24:17 the process for doing changes to PBI are in a manner that don't break your model is is surprisingly hard. And And again, I just want to be clear. you don't have to show to me if you don't want to cuz I again, I'm pushing 24:28 you pretty hard on this one, but suffice it to say all model edits here are captured under the BI Buddy space. model edits, making folders, 24:38 any manipulations we want with the model is is fair game for BI Buddy. Yeah? Yep, for sure. And anything anything that the MCP can do, you can do with 24:48 with with BI Buddy basic Awesome. Love it. before before I show you we'll we'll we'll make it a a teaser for the end of the session. Yeah, yeah, yeah. 24:58 [laughter] Before I show you that, I want to say that I've And this this may sound too technical for the people who are not used to working natively with AI, but Fabric 25:09 made in January, I think or or December, they made a a new open standard that they called Agent Skills. You can find it on agentskills.io. Yes. . And it's a it's a cool way to 25:21 enhance whatever AI agents you're running with , skills. you can have a skill that's focused on making it better at writing DAX code or do data modeling or even just 25:32 fun skill that makes it talk in the voice of a fairy princess or something that. , it's it's At the core it's a it's a package of of a bunch of functionality that you can 25:42 then give to the AI. Sure. BI Buddy is built up around that standard. you see there's a tab called skills. I've made UI for it. there's a bunch of built-in 25:53 skills you'll get out of the box. I've made an M standard skill, DAX measure best practices, data modeling skill that will help it better even from the get-go. Yeah. But you can go and add your own 26:03 personal skills as . I could go here and write my own say, "Hey, I want a new naming convention. I'm not happy with the building one." here's a naming convention and this could be 26:13 always use snake case. I probably wouldn't do this in real life, but I think You can do it if you want. Yeah, if that's what you're if that's what you're 26:23 Again, this is this is where the AI should allow you to flexibility to adhere to your company standards. if you've already built a bunch of models that are already using snake case and that's what you say our business 26:33 does, great. You need to be able to adhere that. Yeah, the built-in skills here, all of them , they have this three toggle switch. You can switch them off, which means that the model is never 26:44 going to hear about it. even though there's a built-in naming standard that is pretty good and doesn't use snake case, Yeah. you can have it ignore it by that switch just switching it off 26:54 and then you can switch your own naming convention on if you if you want that. Love it. on off makes sense. If it's on, the agent knows about it and knows all the all the details. If it's off, it won't 27:06 even know that it exists. Yep. auto, which honestly should probably be the default for all of them, the agent will know it exists, but it will only activate it if it has to use it. if 27:17 if it's getting a question fix my measures, it's going to activate the DAX measure best practices. We can go back in the chat and see where that happened, I think. it it does a little 27:28 bit of reasoning or thinking. It tells you, "Hey, I've accessed this skill." it had loaded the tools specifically here table operations, something something. Did it load a skill as ? I'm going to just going to go 27:39 back and check. no, it probably it probably loaded them at runtime. when you ask the questions, it loaded a bunch of skills without telling us. But but it's using this auto and it will only load it 27:50 if it has to and that means that we won't create a lot of noise because if it has to have to fix measures and we give it a lot of information about 28:01 role of security and there is a lot of information. this skill has bunch of text, but it also have five underlying references of even more text. it it can it can get a 28:12 lot of knowledge there. That's a lot of noise Yes. put that one on auto and it will then trigger it if it if it gets a role of security question and then it can use those that knowledge. Very 28:23 smart. said, yeah, thank you. And as I said, it's built around the the Agent Skills standard. for the the little more tech savvy user or someone who wants to reuse some awesome community skill that 28:35 someone created, you can find the BI Buddy install folder and find them as actual Agent Skills folders. you can copy-paste a skill from outside into this folder and it would work out 28:46 of the box. With it with a few Thank you. , yeah, I I really really really it as . This is really I love the design of this. Absolutely love the fact that I can just 28:56 bring my own stuff and you're going to read through my folder, which is genius because again, sometimes I'm I'm building one thing or two one or two things here or there and I need to have a a whole pile of these things show up 29:06 at once. This makes a lot more sense. I don't have that heavy click tax of always, , going through the UI and having to type every little thing out. You can just generate in mass. This 29:16 is great. And of course, not everyone wants to go to those find those folders and and maybe they won't don't want to risk messing something up. there's of course also an import from folder. 29:26 you could just have it in your download folder, import from folder, pick your Agent Skills folder and then it would it would then then bring bring it out. in for you. Awesome. Love this. cool. 29:37 . Great. Wow. Amazing. Yeah. And then the last part is the tools and and , the main tool for everything it does is the Power BI modeling. that's the MCP. But I 29:47 also included MS Learn. , if you switch this on, you could go and ask things "Please use MS Learn to give me the 29:58 permissions of the different workspace roles." Oh, nice. And then also use it as a a a coach for 30:09 tuning into the best practices or if you don't remember some certain fact from documentation, you can have it be your your documentation fetcher agent as . Yeah. , in this case, it 30:20 should be able to usually bring back a table of of permissions and things that. But, let's let's see how it how it does. Oh, wow. 30:30 Oh, it really went all it went all in on this one. Yeah, it did. , there's these four roles. Yeah, yeah, yeah. What can they do? What are they best for? they do Yeah, went above and 30:40 beyond here. [laughter] He must be using a really good model. Yeah. Yeah. Love it. This is incredible. And yeah, and this 30:50 one this is the most ridiculous feature that I'm proud of, but but doesn't really add any core functionality. But, let's say I'm here and I'm saying "Hey BI buddy, how can 31:01 I add my own naming convention?" Let's say we're a user that didn't watch this video and just wants some help with with the the app itself. It has self-awareness if you if you switch on 31:12 the the UI guide tool. Oh, no. how can I add my own naming convention to this app? 31:26 Oh, awesome. , it has a little self-awareness of the app that it lives inside and what exists there. , it it even says go to skills. Highlight thing. , and in skills here go click this. Add the skill 31:37 description and then you're then you're there. Oh, that's amazing. Yeah. How do you I'm floored. I'm how on earth are you producing that 31:47 with this really really cool? . This is amazing. , that you've gone through skills, tools, and there's a couple of things around settings. , if we're going to do a complete review 31:57 or complete complete land here, Yeah. what key things should I look at? What are the the the settings that we're going to be able to adjust here in the other side? You said something about models and where the models are being serviced from. 32:09 how does that work? What what model are we using here or how do you set up additional models to help support AI buddy? Yeah. , there's a little bit of a an element to what I'm showing that 32:19 that this the the model picker here won't be available to the users that are on-boarded . But, there is there is two main settings here. , there's a bring your own key and there's a built-in model. , if 32:29 you're using the built-in model, it just works out of the box. You don't have to enter anything. You don't have to log in. You don't have to do anything. You can just download the app, install it, and then you can start start working 32:39 with your with your model. Yeah, yeah. there is just for the sake of me having that ultimative 32:49 stop button if something gets excessive, there is a credit system where you cannot do an infinite amount infinite amount of built-in consumption, but you can do pretty fair share of 32:59 of it. And yeah, and every everyone who's signing up here in this period also gets this this this beta onboarding bonus of I think it's a thousand credits. , 33:09 that should be good to do some testing. Great. yeah. Yeah, and then there's bring your own key where if you already have a cloud key or an OpenAI key, you can bring that in here and then 33:19 choose your own own models. And I'm waiting on the the next release, which I can't show here. It has also custom field if you want to add a model that is already not 33:30 on the picker. , you get full flexibility there. Sure. But, I've chosen the models that I know are behaving , for you. across different tests in the in this app. Awesome. Very 33:41 neat. use your own. And if you do this way, I'm not you're not going through any of my systems . The local app here will communicate directly with Claude 33:52 and OpenAI. I don't even get a blip about the fact that you ever sent a message or that you added this API key. in my in my in my back end, you'll just be seen as honestly just an 34:04 inactive user because I don't see anything happening. And this this deliberate choice because I know some enterprises, they do really want minimal don't even want to pass 34:14 Yeah, they don't even want to pass through a back end. Awesome. The built-in of course is passing through my back end because I need to be able to handle the credit, 34:24 but I'm not logging any any messages or conversations or anything about your your model. I don't even get chat chats or the actual messages. They're going to my back end. They're sent directly to 34:35 OpenAI. They're sent back to you and they're deleted the moment they leave the the chain. , there's no even even intermediate cash. , there's there's nothing nothing stored 34:45 there. It's just passed through directly. Can I ask just a brief question around the bring your own key? I I really , there's this idea I've been talking with this publicly a little bit. There's this whole idea of apps should work and 34:56 I should be able to bring my own model. ? Cuz in your example here, you're already paying for you're already paying for Claude code or something, ? you're on the 100 35:06 plan $100 plan or the or the $200 plan or or something that as . I also really using Azure Foundry. And , is that an available Could I use 35:16 Azure Foundry also in bring your own key or is that something that's a little bit too unique because Foundry is a another API or another endpoint at that point. 35:26 Yeah, obviously. , it's a it's how do you say it? It's on my in long-term not prioritized roadmap, but it wouldn't take more than one organization or user saying "Hey, 35:36 this would be a priority for me." and then I I would consider fast-tracking it because because coding-wise, it's just another API endpoint. , , it's 35:47 going to be harder to fit it in UI-wise than it's than it's going to be It's do the work to get it to go. Awesome. Make the work This is amazing. Awesome. 35:57 Yeah. Any any more demo pieces or should we start wrapping and call it quits here? I think the last thing I want to show is that I've I've when I started the beta, it was a little 36:07 rough around the edges in terms of logging in. I This is one of the most recent change I made a proper log in. And this is not the sexiest feature, but it's something I I really . , , if you go here and you 36:18 log in, you're going to go , I want to log in to this email. 36:28 And it'll send you a pass a pass code to your email. , this is modern top top security design because I I did a little bit of research and was surprised to learn that that passwords are going away for a reason and this is the most 36:40 secure way you can have people most secure and easiest way you can have people The the short-lived the the short-lived token or the short short-lived code there. , 36:50 that only lives for a couple minutes and then it's gone and then you can never log in with it again. Yeah. Exactly. And they only arrive to your email. There's nothing , you can't use it for any other users and it will I 37:00 mean, you use it it's already going to be discarded and all those things. Awesome. just want to reuse your Microsoft account because it's it's it's what you also have to use for the fabric 37:10 connection and everything, there's also Azure AD. , you can also click here and log in to to , I was already authenticated, just logged in. Yep. Boom. Easy. It doesn't use your Microsoft. Love it. 37:22 There the sneak peek for later and the thing that I don't promise to to implement, but I did [laughter] play around with. . , I have played around with a little bit of a 37:33 way of making changes to reports. , here is an example. Maybe it'll work. , it seems to be bugging out a bit. . That's unfortunate. [clears throat] 37:43 Yeah. you're experimenting with the report side, but not quite yet. , that , potentially, we'll see something in the future. , when you get something a little bit more concrete here, we 37:54 can pull you back in and say . Yeah, I think that's a good plan. , my my problem is that that making the change is not the hardest problem. The problem is that that when I make the 38:04 change for the model, it's it's seamless. You see the change in your desktop. When I make a change to the report files, you have to close Power BI Desktop Close. Yes. Correct. to see the change. Yeah. 38:14 Yeah, yeah. That makes total sense. And and until Microsoft fixes that problem with the PBIR or PBIP changes, you're dead in the water where you're constantly shutting 38:24 down and restarting Desktop, which is a pain, believe it or not. That's not very a good user experience at all by any means. , I've been doing a lot of manipulation 38:34 things there as . we're we're we're thinking of the same because a lot of times I'm also building tools that are helping us manipulate the report layer and I have found Desktop just doesn't really geared 38:46 up to to let you do external edits on stuff inside Power BI. , you you kind of find yourself building external tools or standalone tools that can do 38:56 the UI edits or changes the way you need them to. And then when you're done, package it up, give it back to Desktop, and then Desktop can handle it. Yeah, for sure. For sure. yeah, 39:06 what I what I won't be able to demo is I've made a prototype of a kind of an an external renderer that will render an intermediate page of how 39:16 does the report look for you. , you could make changes, see them live, and then when you're happy, send them to Power BI Desktop and then Oh, . But, it's very experimental and I it's a 39:27 little There's a lot there. Yeah, there's a lot there. It's a It's almost its own product. , I rather AI BI BI buddy plus. 39:38 [laughter] Yeah, exactly. Exactly. I love it. And I have I have many other better ideas on my road map of how I want to make this even better for for 39:49 users. there will be also interesting things in the future. I love it. , Matthias, thank you much for your time today. I think this was a very interesting project. Really love that you're really deeply diving into 39:59 the MCP, the agentic space. And as we were talking earlier today around, , thinking agentic, this is a great way to think agentically and you're building a 40:09 wonderful harness around large language models that are going to be easy for business users to use and consume. I just want to say thank you much for spending the time with me today. Looking forward to net new developments 40:20 that you're going to produce here. Happy to continue pushing the product, things that you're doing here. I love this idea and I'm sure I'm going to have copious amounts of additional ideas and 40:31 what we can add here. only I'll only give you ideas if I can bring you customers. How about that? [laughter] 40:41 Awesome. , thank you much for your time today. Thank you everyone for listening and watching today. We appreciate your time. This was a great demo of BI Buddy. Hope you enjoy it. Go check out our links down in the description below, also in 40:50 the chat window if you want to go check out the tool. Matthias is accepting early access. early access in the link below. Thank you much and we'll see you next time. 41:01 See you. Agentic thinking. [music]