WEBVTT 00:00:00.000 --> 00:00:01.639 I was looking at my business ledger the other 00:00:01.639 --> 00:00:04.120 day, and I realized something a bit disturbing. 00:00:05.120 --> 00:00:08.960 We are all paying this, well, I've started calling 00:00:08.960 --> 00:00:10.919 it the intelligence rent. The intelligence rent. 00:00:10.980 --> 00:00:13.099 That is an incredibly accurate way to describe 00:00:13.099 --> 00:00:15.660 it. Right. Because for the last three years, 00:00:15.699 --> 00:00:19.100 we've been conditioned to believe that AI is 00:00:19.100 --> 00:00:23.300 strictly a service. You want a summary, you pay 00:00:23.300 --> 00:00:26.199 the monthly sub, you want to generate code, you 00:00:26.199 --> 00:00:29.500 pay per token. Exactly. We are essentially renting 00:00:29.500 --> 00:00:32.240 a brain from a landlord we don't know, a landlord 00:00:32.240 --> 00:00:35.439 who can change the rent or the rules whenever 00:00:35.439 --> 00:00:37.659 they feel like it. And that is the trap. It's 00:00:37.659 --> 00:00:39.899 convenient, sure. But the shift we are seeing 00:00:39.899 --> 00:00:42.320 now asks a totally different question. Which 00:00:42.320 --> 00:00:45.399 is? What if you could just own the brain, not 00:00:45.399 --> 00:00:47.799 a subscription to it, the actual files, the intuition, 00:00:48.039 --> 00:00:50.119 sitting right there on your local drive, doing 00:00:50.119 --> 00:00:52.310 your bidding for free? That is exactly where 00:00:52.310 --> 00:00:55.810 we are going today. It is Friday, February 27th, 00:00:55.810 --> 00:00:59.250 2026. Time flies. The deep seek moment is a full 00:00:59.250 --> 00:01:03.289 year behind us. The landscape has settled. And 00:01:03.289 --> 00:01:06.189 the question isn't if you can run a GPT -5 class 00:01:06.189 --> 00:01:08.969 model on your laptop anymore. The question is 00:01:08.969 --> 00:01:11.549 why haven't you started yet? Honestly, if you 00:01:11.549 --> 00:01:13.450 haven't started, you're falling behind on the 00:01:13.450 --> 00:01:15.950 most important asset class of the decade. Welcome 00:01:15.950 --> 00:01:19.569 to the Deep Dive. We are exploring a really fascinating 00:01:19.569 --> 00:01:23.090 guide today by Max Ann. It's titled The Complete 00:01:23.090 --> 00:01:26.750 Guide to Open Source AI in 2026. It's a great 00:01:26.750 --> 00:01:28.609 piece. It really is. And it doesn't just read 00:01:28.609 --> 00:01:30.790 like a standard tutorial. It reads more like 00:01:30.790 --> 00:01:33.209 a survival manual for the modern Internet. Yeah, 00:01:33.230 --> 00:01:35.010 what I appreciate about this approach is that 00:01:35.010 --> 00:01:38.390 it moves us past the philosophy. We aren't just 00:01:38.390 --> 00:01:40.409 talking about the ideals of open source. Right. 00:01:40.530 --> 00:01:43.450 We are breaking down the actual stack, the software 00:01:43.450 --> 00:01:45.790 and the hardware you need to reclaim your digital 00:01:45.790 --> 00:01:48.030 sovereignty. So give us the roadmap for today. 00:01:48.170 --> 00:01:50.670 If you, the listener, want to move from being 00:01:50.670 --> 00:01:53.689 a user to an operator, what does that path look 00:01:53.689 --> 00:01:55.530 like? We are going to tackle this in three stages. 00:01:55.750 --> 00:01:57.930 First, we need to dismantle the stack itself. 00:01:58.189 --> 00:02:00.090 Understand the difference between the recipe 00:02:00.090 --> 00:02:02.560 and the meal. Second, we have to address the 00:02:02.560 --> 00:02:05.519 elephant in the room the geopolitics. We need 00:02:05.519 --> 00:02:08.340 to look at why Chinese built models are dominating 00:02:08.340 --> 00:02:11.039 the charts and dismantle the fear around using 00:02:11.039 --> 00:02:13.199 them locally. That's a massive point. I think 00:02:13.199 --> 00:02:15.719 a lot of people see made in China on software 00:02:15.719 --> 00:02:18.159 and immediately think spyware. Exactly. And we're 00:02:18.159 --> 00:02:19.840 going to debunk why that's a misunderstanding 00:02:19.840 --> 00:02:22.780 of how open weights actually work. Correct. And 00:02:22.780 --> 00:02:25.060 finally, we are going to get practical. We're 00:02:25.060 --> 00:02:27.819 going to walk through building a private financial 00:02:27.819 --> 00:02:32.699 analyst agent that runs entirely offline. No 00:02:32.699 --> 00:02:35.419 cloud, just your bank statements and your local 00:02:35.419 --> 00:02:38.460 AI. I love that. Let's start at the beginning 00:02:38.460 --> 00:02:41.219 then. Open source AI. It's a buzzword we've heard 00:02:41.219 --> 00:02:44.919 since 2023. But in 2026, the definition has really 00:02:44.919 --> 00:02:46.860 hardened. It has. What are we actually talking 00:02:46.860 --> 00:02:48.460 about here? It's not just free software, right? 00:02:48.599 --> 00:02:51.240 No, not at all. The source uses this culinary 00:02:51.240 --> 00:02:53.719 analogy that I think is just perfect. Think of 00:02:53.719 --> 00:02:57.219 closed source AI, your GPTs, your clods, as a 00:02:57.219 --> 00:02:59.439 finished meal delivered to your door in a locked 00:02:59.439 --> 00:03:02.240 box. Okay, I follow. You get the meal. It tastes 00:03:02.240 --> 00:03:04.719 great, but you don't know exactly what's in it. 00:03:05.120 --> 00:03:07.780 You can't change the seasoning. And most importantly, 00:03:07.919 --> 00:03:11.000 if the restaurant closes, you starve. That's 00:03:11.000 --> 00:03:14.099 the API model. Precisely. Open source is different. 00:03:14.180 --> 00:03:15.879 They don't give you the meal. They give you the 00:03:15.879 --> 00:03:17.879 recipe, the ingredients, and the keys to the 00:03:17.879 --> 00:03:19.719 industrial kitchen. You own the whole process. 00:03:19.979 --> 00:03:21.680 You own the architecture. You own the training 00:03:21.680 --> 00:03:25.099 code. And critically, you own the weights. Let's 00:03:25.099 --> 00:03:27.900 pause on that word, weights. We hear that term 00:03:27.900 --> 00:03:30.830 constantly. For the non -engineers listening, 00:03:31.110 --> 00:03:33.930 what exactly is a weight in a dozen words or 00:03:33.930 --> 00:03:37.210 less? Think of a weight as the crystallized intuition 00:03:37.210 --> 00:03:40.650 of the model. Crystallized intuition? Yeah. It's 00:03:40.650 --> 00:03:43.110 just numbers representing the relationships between 00:03:43.110 --> 00:03:46.150 words. It's the physical brain file sitting on 00:03:46.150 --> 00:03:47.889 your hard drive. So it's the difference between 00:03:47.889 --> 00:03:50.550 dining out and being the chef. Exactly. But this 00:03:50.550 --> 00:03:52.770 shift didn't just happen gradually. There was 00:03:52.770 --> 00:03:55.490 a specific tipping point. The source calls it 00:03:55.490 --> 00:03:59.219 the deep -seek moment. January 2025. That was 00:03:59.219 --> 00:04:01.539 the turning point. I remember it vividly. Before 00:04:01.539 --> 00:04:05.080 that, open source was, well, it was kind of cute. 00:04:05.219 --> 00:04:08.780 It was for hobbyists making funny poems. Then 00:04:08.780 --> 00:04:11.680 DeepSeek R1 dropped. And the floor just fell 00:04:11.680 --> 00:04:14.419 out from under big tech valuations. It did. Because 00:04:14.419 --> 00:04:16.899 suddenly you had a model that wasn't just good 00:04:16.899 --> 00:04:19.699 for being free. It competed directly with the 00:04:19.699 --> 00:04:23.389 closed giants on pure reasoning. Math. Coding 00:04:23.389 --> 00:04:25.930 logic. It proved you didn't need a trillion dollar 00:04:25.930 --> 00:04:27.930 data center to be smart. You just needed better 00:04:27.930 --> 00:04:30.269 math. Exactly. It changed the psychology of the 00:04:30.269 --> 00:04:32.410 entire industry. So is the main benefit just 00:04:32.410 --> 00:04:35.069 saving the monthly subscription fee? No, it's 00:04:35.069 --> 00:04:37.569 data sovereignty. Yeah. Total immunity to corporate 00:04:37.569 --> 00:04:40.009 pricing and privacy changes. Right. You are the 00:04:40.009 --> 00:04:42.730 captain of the ship. Yeah. But let's talk about 00:04:42.730 --> 00:04:44.709 where that engine is coming from, because we 00:04:44.709 --> 00:04:47.029 really have to address the China factor. We do. 00:04:47.129 --> 00:04:49.839 It's unavoidable. As of mid -2026, if you look 00:04:49.839 --> 00:04:51.779 at the Hugging Face download charts, the top 00:04:51.779 --> 00:04:54.620 slots are dominated by names like Quen, Hunyuan, 00:04:54.860 --> 00:04:57.720 GLM. All Chinese build models. Right. And the 00:04:57.720 --> 00:05:00.759 source quotes a stat from A16s that roughly 80 00:05:00.759 --> 00:05:03.160 % of their portfolio startups are building on 00:05:03.160 --> 00:05:06.199 top of these models. It's a staggering number. 00:05:06.600 --> 00:05:08.800 Western capital building on Eastern code. But 00:05:08.800 --> 00:05:10.959 for a lot of listeners, that triggers a reflex. 00:05:11.279 --> 00:05:13.339 They think, wait, am I sending my data to China? 00:05:13.459 --> 00:05:16.019 Is this a Trojan horse? Right. And this is the 00:05:16.019 --> 00:05:17.939 most critical technical distinction we need to 00:05:17.939 --> 00:05:21.019 make today. Using a Chinese model in open source 00:05:21.019 --> 00:05:23.579 does not mean connecting to a Chinese server. 00:05:23.959 --> 00:05:26.100 Walk us through that mechanism. How can you be 00:05:26.100 --> 00:05:28.759 completely sure? When you use a closed model, 00:05:28.959 --> 00:05:32.439 you send data to a server via an API. They process 00:05:32.439 --> 00:05:34.360 it. They see it. They send it back. That is a 00:05:34.360 --> 00:05:36.779 surveillance risk. Right. But with open source, 00:05:37.100 --> 00:05:39.660 you are downloading a file, the weights we talked 00:05:39.660 --> 00:05:42.540 about, directly to your hard drive. Once that 00:05:42.540 --> 00:05:44.980 download finishes, you can literally pull the 00:05:44.980 --> 00:05:47.079 Ethernet cable out of the wall. The model runs 00:05:47.079 --> 00:05:50.300 purely on your silicon. Exactly. A Chinese -hosted 00:05:50.300 --> 00:05:53.449 API is a surveillance risk. But a Chinese -built 00:05:53.449 --> 00:05:55.649 open source file running on your disconnected 00:05:55.649 --> 00:05:58.569 laptop, that is just math. It cannot phone home 00:05:58.569 --> 00:06:00.410 because there is no home to phone to. You are 00:06:00.410 --> 00:06:02.670 the host. If the models are free and local, does 00:06:02.670 --> 00:06:05.509 that mean the AI arms race is over for the consumer? 00:06:05.910 --> 00:06:08.470 In a way, yes. Western investment accelerated, 00:06:08.569 --> 00:06:11.189 but we just get better tools faster. We are the 00:06:11.189 --> 00:06:13.910 beneficiaries of their war. I like that. But 00:06:13.910 --> 00:06:16.649 let's be real for a second. It's not all sunshine 00:06:16.649 --> 00:06:18.769 and rainbows. No, it's not. The source has a 00:06:18.769 --> 00:06:21.540 section detailing the honest pros and cons. What's 00:06:21.540 --> 00:06:25.139 the catch? The catch is friction. It's not magic. 00:06:25.360 --> 00:06:28.860 The biggest pro is control and privacy. No per 00:06:28.860 --> 00:06:33.980 -token fees. But the con, you are the IT department. 00:06:34.240 --> 00:06:36.740 There is no support desk to call. Exactly. If 00:06:36.740 --> 00:06:39.220 the model hallucinates or your GPU drivers won't 00:06:39.220 --> 00:06:42.120 update, that's on you. You have to be willing 00:06:42.120 --> 00:06:44.319 to tinker. And speaking of tinkering, let's talk 00:06:44.319 --> 00:06:47.019 hardware. Because... Honestly, the hardware specs 00:06:47.019 --> 00:06:49.920 still intimidate me. Really? Yeah, I mean, I 00:06:49.920 --> 00:06:52.279 still wrestle with prompt drift and memory limits 00:06:52.279 --> 00:06:55.180 myself. Whenever I see terms like VRAM and quantization, 00:06:55.560 --> 00:06:57.759 I feel like I'm trying to build a gaming PC in 00:06:57.759 --> 00:07:00.259 the 90s. It is totally valid to feel intimidated, 00:07:00.459 --> 00:07:03.279 but the barrier has lowered massively thanks 00:07:03.279 --> 00:07:06.300 to that exact word, quantization. That's essentially 00:07:06.300 --> 00:07:08.620 compression, right? Like making an MP3 for AI. 00:07:08.839 --> 00:07:11.420 That is a perfect analogy. High -end models used 00:07:11.420 --> 00:07:14.480 to be these massive uncompressed files. Quantization 00:07:14.480 --> 00:07:17.139 is a way of reducing the mathematical precision 00:07:17.139 --> 00:07:19.819 of those weights, making the file dramatically 00:07:19.819 --> 00:07:23.319 smaller. Does compressing a model via quantization 00:07:23.319 --> 00:07:27.180 make it dumber? Surprisingly little. It strips 00:07:27.180 --> 00:07:29.800 file size without gutting the reasoning capabilities. 00:07:30.040 --> 00:07:32.399 That is wild. It really was the big discovery 00:07:32.399 --> 00:07:34.459 the last couple of years. You can shrink a file 00:07:34.459 --> 00:07:37.879 by 70 % and maybe lose 2 % of his logic skills. 00:07:38.120 --> 00:07:40.180 So what are the actual hardware requirements 00:07:40.180 --> 00:07:43.620 now? Lay out the rough guide for 2026. Okay, 00:07:43.660 --> 00:07:46.420 if you want to run a small, fast model, say... 00:07:46.839 --> 00:07:49.579 One to three billion parameters. Yeah. Good for 00:07:49.579 --> 00:07:51.240 quick summaries. You only need four to eight 00:07:51.240 --> 00:07:53.980 gigabytes of RAM. Basically any modern laptop. 00:07:54.180 --> 00:07:56.259 Exactly. Now, if you want strong capabilities, 00:07:56.579 --> 00:08:00.379 coding, complex analysis, that 13 to 34 billion 00:08:00.379 --> 00:08:03.040 parameter range, you need 16 to 32 gigabytes 00:08:03.040 --> 00:08:06.160 of RAM. So a high -end MacBook or a solid gaming 00:08:06.160 --> 00:08:08.379 laptop. Right. If you want the frontier performance, 00:08:08.759 --> 00:08:13.240 the 70 billion parameter beasts, you need 32 00:08:13.240 --> 00:08:15.600 gigabytes of VRAM or more. That's workstation 00:08:15.600 --> 00:08:18.740 territory. It is. But the main takeaway is you 00:08:18.740 --> 00:08:21.259 do not need a supercomputer. Your daily work 00:08:21.259 --> 00:08:23.579 machine is probably enough to start. Okay, we 00:08:23.579 --> 00:08:25.819 have the hardware sorted. Let's actually build 00:08:25.819 --> 00:08:29.339 the stack. The source visualizes this as four 00:08:29.339 --> 00:08:32.080 distinct layers. Let's start at the bottom. Layer 00:08:32.080 --> 00:08:36.000 one, the models. The brain itself. The source 00:08:36.000 --> 00:08:39.870 mentions Quen 3 from Alibaba. And it notes that 00:08:39.870 --> 00:08:44.309 QAN 3 .2 .35b is currently rivaling GPT -5 benchmarks. 00:08:44.730 --> 00:08:47.409 Which is insane for a downloadable file. You 00:08:47.409 --> 00:08:50.190 also have LAMA from Meta, which is basically 00:08:50.190 --> 00:08:52.529 the industry standard for compatibility, and 00:08:52.529 --> 00:08:55.169 Mistral from Europe, which is incredibly efficient. 00:08:55.490 --> 00:08:57.750 You can't just run a raw file. You need a manager. 00:08:58.029 --> 00:09:01.590 Right. That brings us to layer two, Alama. Explain 00:09:01.590 --> 00:09:04.429 Alama. Think of Alama as a package manager for 00:09:04.429 --> 00:09:07.269 brains. It makes running local AI ridiculously 00:09:07.269 --> 00:09:09.870 easy. You download the app, open your terminal, 00:09:09.970 --> 00:09:13.110 and type ulama run when 3. And that's it? That's 00:09:13.110 --> 00:09:14.909 it. It downloads the weights, configures your 00:09:14.909 --> 00:09:16.730 hardware, and drops you into a chat interface. 00:09:17.149 --> 00:09:18.769 You know, I have to just pause on that for a 00:09:18.769 --> 00:09:21.029 second just to marvel at it. Oh, absolutely. 00:09:21.230 --> 00:09:25.370 Whoa. Just imagine it. A billion parameter intelligence 00:09:25.370 --> 00:09:29.210 capable of coding and deep reasoning living silently 00:09:29.210 --> 00:09:31.610 on your laptop hard drive waiting for a command. 00:09:32.029 --> 00:09:34.429 No internet needed. It feels like stealing fire 00:09:34.429 --> 00:09:36.889 from the gods. It really does. If a llama runs 00:09:36.889 --> 00:09:39.330 the brain, how do we make it actually do work? 00:09:39.529 --> 00:09:42.269 We need hands. And that is where the orchestration 00:09:42.269 --> 00:09:45.549 layer comes in. Layer 3. Introduce us to N8M. 00:09:45.669 --> 00:09:48.590 So N8M is for non -coders. It's a visual workflow 00:09:48.590 --> 00:09:51.669 builder. You literally draw lines between boxes 00:09:51.669 --> 00:09:53.889 on your screen. Like connecting flowcharts. Exactly. 00:09:54.070 --> 00:09:57.149 One box is check Google Drive. The next is send 00:09:57.149 --> 00:10:00.039 to a llama. The next is draft an email. It connects 00:10:00.039 --> 00:10:02.580 the local brain to your external tools. And the 00:10:02.580 --> 00:10:04.399 source mentioned something called the self -hosted 00:10:04.399 --> 00:10:07.480 AI starter kit. Yes, this is brilliant. It bundles 00:10:07.480 --> 00:10:10.519 NAN, ALAMA, a vector database called Qtrent for 00:10:10.519 --> 00:10:13.000 memory, and Postgres will all into one single 00:10:13.000 --> 00:10:15.440 Docker container. One download and you have the 00:10:15.440 --> 00:10:17.720 brain, the hands, and the long -term memory. 00:10:17.899 --> 00:10:21.090 Exactly. Briefly, what about layer four? For 00:10:21.090 --> 00:10:22.610 the developers. That's the Python frameworks. 00:10:22.769 --> 00:10:25.350 Tools like the OpenAI agents, SDK, LandGraph, 00:10:25.470 --> 00:10:27.809 Lama Index. Is the logic different when switching 00:10:27.809 --> 00:10:31.529 from OpenAI to local AI? No, the agent logic 00:10:31.529 --> 00:10:34.029 is identical. You just swap the brain component 00:10:34.029 --> 00:10:36.629 in your settings. Meaning you just change the 00:10:36.629 --> 00:10:38.669 web address it points to. Right. You change it 00:10:38.669 --> 00:10:41.570 from the OpenAI server to your local Alama address. 00:10:41.710 --> 00:10:44.629 The code stays exactly the same. That makes the 00:10:44.629 --> 00:10:46.990 transition frictionless. Okay, we've got the 00:10:46.990 --> 00:10:48.929 theory. Let's make this real with a practical 00:10:48.929 --> 00:10:52.149 demo from the source. The private financial analyzer. 00:10:52.210 --> 00:10:55.750 Yes. Walk us through the scenario. Why use open 00:10:55.750 --> 00:10:59.230 source for this specific task? Because of the 00:10:59.230 --> 00:11:03.169 risk. You are analyzing three years of bank statements 00:11:03.169 --> 00:11:05.789 and credit card history. You would never upload 00:11:05.789 --> 00:11:08.549 those PDFs to a public cloud chatbot. Absolutely 00:11:08.549 --> 00:11:10.509 not. It's too sensitive. So you do it offline. 00:11:10.789 --> 00:11:13.389 The workflow is simple. Step one, download Alama 00:11:13.389 --> 00:11:16.649 and N8n. Step two, pull a smart but efficient 00:11:16.649 --> 00:11:20.259 model like Quinn 2 .538b. Got it. Step three, 00:11:20.399 --> 00:11:23.740 import the workflow file. Step four, point the 00:11:23.740 --> 00:11:26.159 read files node to a local folder on your computer 00:11:26.159 --> 00:11:28.460 containing your bank PDFs. And then you just 00:11:28.460 --> 00:11:31.779 run it. You run it. And the output is incredible. 00:11:31.980 --> 00:11:35.139 It categorizes your spending into housing, dining, 00:11:35.360 --> 00:11:37.980 subscriptions. It spots seasonal trends. Like 00:11:37.980 --> 00:11:40.120 noticing you spend more on coffee in the winter. 00:11:40.340 --> 00:11:42.659 Right. And it gives personalized savings advice 00:11:42.659 --> 00:11:45.379 based entirely on your actual habits. And zero 00:11:45.379 --> 00:11:48.659 bytes of that data. Ever left your laptop? Completely 00:11:48.659 --> 00:11:50.980 contained. Oh, and the source has a great technical 00:11:50.980 --> 00:11:54.179 tip here. If your N8n container can't talk to 00:11:54.179 --> 00:11:57.649 your Alama container, don't use localhost. Use 00:11:57.649 --> 00:12:03.629 http .host .docker .internal .11434. Host .docker 00:12:03.629 --> 00:12:06.470 .internal. That is a lifesaver for anyone dealing 00:12:06.470 --> 00:12:09.090 with Docker networking issues. It truly is. Once 00:12:09.090 --> 00:12:11.389 it analyzes the past, can it help with the future? 00:12:11.570 --> 00:12:13.690 Yes, you can extend it to run monthly and alert 00:12:13.690 --> 00:12:15.830 you when you overspend. It goes from a tool to 00:12:15.830 --> 00:12:18.570 a proactive guardian. Exactly. Before we wrap 00:12:18.570 --> 00:12:20.389 up the stack, I want to touch on the coding ecosystem 00:12:20.389 --> 00:12:23.190 because this shift changes how developers build 00:12:23.190 --> 00:12:25.529 software too. Oh, completely. Tools like Cursor, 00:12:25.570 --> 00:12:28.090 Warp, Continuum, and Ader are reshaping development. 00:12:28.470 --> 00:12:30.649 You can have an AI coding assistant that knows 00:12:30.649 --> 00:12:33.409 your entire proprietary code base but runs entirely 00:12:33.409 --> 00:12:35.950 locally. So you aren't leaking company secrets 00:12:35.950 --> 00:12:39.009 to train someone else's model. Right. And this 00:12:39.009 --> 00:12:41.289 brings up the main takeaway from the guide. Open 00:12:41.289 --> 00:12:43.789 source isn't about replacing commercial AI for... 00:12:44.029 --> 00:12:46.110 absolutely everything. It's about using the right 00:12:46.110 --> 00:12:49.190 tool for the job. Exactly. Use commercial cloud 00:12:49.190 --> 00:12:51.690 AI for the absolute highest reasoning on a one 00:12:51.690 --> 00:12:55.509 -off task, but use local open source for privacy, 00:12:55.750 --> 00:12:58.350 for scale, and for building permanent systems. 00:12:58.610 --> 00:13:00.690 What is the only thing holding people back now? 00:13:00.950 --> 00:13:03.929 Honestly, just the habit of convenience. The 00:13:03.929 --> 00:13:06.029 technical barriers are basically gone. It is 00:13:06.029 --> 00:13:08.049 just a matter of changing our default behavior. 00:13:08.649 --> 00:13:11.850 Two sec silence. We are going to take a quick 00:13:11.850 --> 00:13:13.710 break. When we come back, we'll look at the big 00:13:13.710 --> 00:13:16.230 idea behind all of this. Stay with us. You're 00:13:16.230 --> 00:13:19.429 right back. And we're back. Welcome back to the 00:13:19.429 --> 00:13:22.409 Deep Dive. We've broken down the 2026 open source 00:13:22.409 --> 00:13:25.289 stack. We've talked about Alama, Quinn, and Egon. 00:13:25.490 --> 00:13:27.789 But let's zoom out to the core philosophy of 00:13:27.789 --> 00:13:30.610 this entire episode. The big idea. Yeah. Why 00:13:30.610 --> 00:13:33.629 does running this locally actually matter? It 00:13:33.629 --> 00:13:36.009 all comes back to data sovereignty. In a world 00:13:36.009 --> 00:13:38.169 increasingly built on subscriptions and surveillance, 00:13:38.549 --> 00:13:41.429 running your own stack is an act of genuine independence. 00:13:41.870 --> 00:13:44.470 It's opting out of the rental economy. Exactly. 00:13:44.960 --> 00:13:48.080 You completely bypass vendor lock -in. You ensure 00:13:48.080 --> 00:13:50.879 that your business logic survives, even if a 00:13:50.879 --> 00:13:53.440 major AI company changes its pricing, pivots 00:13:53.440 --> 00:13:56.539 its model, or just goes offline entirely. You 00:13:56.539 --> 00:13:59.419 are building an asset you actually own, not renting 00:13:59.419 --> 00:14:02.080 space on someone else's server. Right. So based 00:14:02.080 --> 00:14:04.480 on Maxan's guide, what is the challenge for the 00:14:04.480 --> 00:14:07.360 listener today? The challenge is simple. Go download 00:14:07.360 --> 00:14:11.379 Alama. Open your terminal. Run Alama. Run FWEN3. 00:14:11.870 --> 00:14:14.169 Watch the text stream across your screen and 00:14:14.169 --> 00:14:16.409 just see how it feels to physically own the intelligence. 00:14:16.649 --> 00:14:18.909 Because if you sleep on this now, you are going 00:14:18.909 --> 00:14:21.649 to wake up next year paying rent on a digital 00:14:21.649 --> 00:14:23.809 house you could have easily owned. Seriously. 00:14:24.090 --> 00:14:26.750 And I guarantee, once you see those tokens generating 00:14:26.750 --> 00:14:29.889 entirely offline for the first time, you're going 00:14:29.889 --> 00:14:32.029 to want to check your own server logs just to 00:14:32.029 --> 00:14:33.750 prove yourself it isn't connected to the internet. 00:14:33.870 --> 00:14:36.250 It feels like magic. I still check my logs sometimes, 00:14:36.470 --> 00:14:41.179 just to be sure. Speed. Here is a final provocative 00:14:41.179 --> 00:14:43.940 thought for you to chew on as we wrap up. Let's 00:14:43.940 --> 00:14:47.259 hear it. If you build a completely closed local 00:14:47.259 --> 00:14:51.580 system today, what happens when AI starts communicating 00:14:51.580 --> 00:14:54.360 via high frequency local mesh networks tomorrow? 00:14:54.899 --> 00:14:57.940 Your isolated house suddenly becomes a node in 00:14:57.940 --> 00:15:00.860 a massive decentralized nervous system that no 00:15:00.860 --> 00:15:03.220 corporation controls. Now that is a deep rabbit 00:15:03.220 --> 00:15:05.120 hole. I might need to go spit up another container 00:15:05.120 --> 00:15:07.139 just thinking about it. Have fun in the terminal. 00:15:07.639 --> 00:15:09.220 Thank you all for joining us on this deep dive. 00:15:09.340 --> 00:15:09.879 See you next time.