1 00:00:00,000 --> 00:00:04,350 Welcome to the deep dive today. We're really getting into something exciting flow 2 00:00:04,350 --> 00:00:07,000 wise. It's an open source platform and honestly 3 00:00:07,000 --> 00:00:10,700 It's changing the game for building AI agents our mission today to show you how 4 00:00:10,700 --> 00:00:13,880 something that sounds super complex is actually becoming well 5 00:00:13,880 --> 00:00:18,240 Surprisingly accessible even if you're not, you know a hardcore AI coder 6 00:00:18,240 --> 00:00:22,910 But hang on before we dive right in a quick shout out to our supporter safe server 7 00:00:22,910 --> 00:00:23,320 dot DE 8 00:00:23,320 --> 00:00:26,060 They're fantastic when it comes to digital transformation 9 00:00:26,060 --> 00:00:30,560 They actually host the software that helps us put together these deep dives for you 10 00:00:30,560 --> 00:00:35,660 So if you're curious about boosting your own digital setup, check them out at www.safeserver.de 11 00:00:35,660 --> 00:00:40,920 Right. So flow wise as I said open source and it lets you build AI agents visually 12 00:00:40,920 --> 00:00:42,880 now when you hear AI agents 13 00:00:42,880 --> 00:00:46,220 Maybe you picture I don't know teams of developers deep and complex code, right? 14 00:00:46,220 --> 00:00:50,440 But the real magic here what flow wise does is it kind of demystifies all that it 15 00:00:50,440 --> 00:00:52,520 makes building these things intuitive really powerful 16 00:00:52,520 --> 00:00:56,140 And honestly something a much wider range of people can actually do that's such a 17 00:00:56,140 --> 00:00:57,440 key point. I mean traditionally 18 00:00:57,440 --> 00:01:01,290 Yeah building AI agents meant you needed serious technical chops coding machine 19 00:01:01,290 --> 00:01:03,100 learning frameworks the whole deal 20 00:01:03,100 --> 00:01:08,760 What's so fascinating about flow wise is how it just bypasses that whole 21 00:01:08,760 --> 00:01:10,160 intimidating entry barrier 22 00:01:10,160 --> 00:01:14,860 It brings this really advanced capability into a visual space, you know makes it 23 00:01:14,860 --> 00:01:17,320 approachable. It's not just simplifying things 24 00:01:17,320 --> 00:01:21,760 It's genuinely like democratizing access to these powerful AI tools, which is 25 00:01:21,760 --> 00:01:23,560 frankly a massive step forward 26 00:01:23,560 --> 00:01:26,320 Okay, so let's unpack that a bit for anyone who listening is maybe, you know 27 00:01:26,320 --> 00:01:28,800 tangled with complex AI stuff before flow 28 00:01:28,800 --> 00:01:31,370 Why it sounds like a completely different approach. What exactly is it 29 00:01:31,370 --> 00:01:32,240 fundamentally? 30 00:01:32,240 --> 00:01:37,670 And how does it pull off making something so sophisticated feel well easier? Yeah, 31 00:01:37,670 --> 00:01:38,160 good question 32 00:01:38,160 --> 00:01:44,280 So at its heart flow wise is an open source platform specifically for developing 33 00:01:44,280 --> 00:01:45,420 agentic systems 34 00:01:45,920 --> 00:01:51,040 The revolutionary bit is the visual building it uses these modular building blocks 35 00:01:51,040 --> 00:01:56,260 Think of it like like building with high-tech Legos instead of forging metal parts 36 00:01:56,260 --> 00:01:56,900 from scratch 37 00:01:56,900 --> 00:02:01,530 You snap together these pre-built functional modules on a screen to define how your 38 00:02:01,530 --> 00:02:02,120 AI works 39 00:02:02,120 --> 00:02:06,920 It's logic its flow all visual this massively cuts down on the you know 40 00:02:06,920 --> 00:02:10,270 The mental load and the amount of boilerplate code you'd normally have to write and 41 00:02:10,270 --> 00:02:12,120 just to give a sense of its credibility 42 00:02:12,120 --> 00:02:16,710 It's backed by Y Combinator and already being used by teams all over the world. 43 00:02:16,710 --> 00:02:17,720 That says a lot, right? 44 00:02:17,720 --> 00:02:21,320 So it's not just a pretty interface. It's changing how you actually build the thing 45 00:02:21,320 --> 00:02:24,740 Yeah sounds almost like those drag-and-drop website builders that totally changed 46 00:02:24,740 --> 00:02:26,240 web design, but this is for AI 47 00:02:26,240 --> 00:02:30,480 You're arranging pieces to map out complex AI behavior instead of typing endless 48 00:02:30,480 --> 00:02:31,100 lines of code 49 00:02:31,100 --> 00:02:34,280 That definitely sounds like it lets you as developers say iterate fast 50 00:02:34,280 --> 00:02:38,710 Does this visual approach ever feel limiting or can flow wise really handle the 51 00:02:38,710 --> 00:02:40,560 complex stuff? You'd normally code 52 00:02:40,640 --> 00:02:45,150 That's a really fair question and it gets right to the core of how flow wise is 53 00:02:45,150 --> 00:02:47,120 designed. It simplifies 54 00:02:47,120 --> 00:02:50,600 Yes, but crucially it doesn't really sacrifice power or flexibility 55 00:02:50,600 --> 00:02:56,580 The core promise is that you can build everything from say simple workflows where 56 00:02:56,580 --> 00:02:59,380 an AI follows set steps all the way up to fully 57 00:02:59,380 --> 00:03:02,160 Autonomous agents and all through that visual interface 58 00:03:02,160 --> 00:03:05,880 So you're focusing purely on the logic and the flow how the AI should behave 59 00:03:05,880 --> 00:03:08,840 without getting bogged down in the underlying code implementation 60 00:03:08,840 --> 00:03:10,880 And for anyone wondering about getting started 61 00:03:10,880 --> 00:03:16,120 It's incredibly simple like a quick NPM install dash g flow wise and then npx flow 62 00:03:16,120 --> 00:03:17,040 wise start and boom 63 00:03:17,040 --> 00:03:21,460 You're basically up and running that ease of setup is key for rapid prototyping and 64 00:03:21,460 --> 00:03:24,080 the modular design means it's robust enough for complex 65 00:03:24,080 --> 00:03:28,440 Scenarios to it bridges that gap between no code simplicity and code heavy power, 66 00:03:28,440 --> 00:03:29,160 you know 67 00:03:29,160 --> 00:03:33,480 Okay, that makes the barrier to entry almost nothing. So let's get practical. What 68 00:03:33,480 --> 00:03:34,480 can you actually build with this? 69 00:03:34,480 --> 00:03:37,810 What kind of tangible things could our listeners start thinking about for their own 70 00:03:37,810 --> 00:03:38,560 work or projects? 71 00:03:38,960 --> 00:03:42,450 Great. This is where it gets really interesting. You can basically break down what 72 00:03:42,450 --> 00:03:44,120 you build into two main categories 73 00:03:44,120 --> 00:03:46,560 First up, you've got chat assistance 74 00:03:46,560 --> 00:03:50,200 These are probably what most people think of as an AI chatbot 75 00:03:50,200 --> 00:03:54,930 Single agent systems designed for specific interactions, but they're super 76 00:03:54,930 --> 00:03:55,640 versatile 77 00:03:55,640 --> 00:04:00,210 They support things like tool calling which means the AI isn't just talking it can 78 00:04:00,210 --> 00:04:04,280 actually do things like interact with other systems trigger actions 79 00:04:04,280 --> 00:04:08,900 Oh interesting like pull up customer data or book something exactly pull order 80 00:04:08,900 --> 00:04:11,040 history from a CRM book an appointment 81 00:04:11,040 --> 00:04:14,440 maybe even process a simple transaction all based on the chat and 82 00:04:14,440 --> 00:04:18,990 Critically these assistants are great at knowledge retrieval often using this 83 00:04:18,990 --> 00:04:22,360 technique called rag retrieval augmented generation 84 00:04:22,360 --> 00:04:25,640 This means the AI isn't just using its general knowledge 85 00:04:25,640 --> 00:04:30,590 It can intelligently search and pull specific up-to-date info from your own private 86 00:04:30,590 --> 00:04:32,640 data sources before it answers 87 00:04:32,640 --> 00:04:37,280 Ah, so it's like giving it a super focused research assistant for your specific 88 00:04:37,280 --> 00:04:38,500 stuff precisely 89 00:04:38,500 --> 00:04:41,440 It makes the answers not just accurate in general 90 00:04:41,440 --> 00:04:45,750 But highly relevant to your context your company docs latest market reports product 91 00:04:45,750 --> 00:04:46,240 manuals 92 00:04:46,240 --> 00:04:49,640 Whatever and when I say it supports a wide array of data sources 93 00:04:49,640 --> 00:04:52,720 I really mean it it handles almost everything you'd find in a business 94 00:04:52,720 --> 00:04:59,320 Txt files PDFs rtf doc HTML CSS JSON XML CSV markdown 95 00:05:00,040 --> 00:05:04,080 Even direct sequel database chats Wow. Yeah, it's comprehensive 96 00:05:04,080 --> 00:05:07,600 You can feed it pretty much all your relevant info turning all that scatter data 97 00:05:07,600 --> 00:05:09,600 into something that AI can use intelligently 98 00:05:09,600 --> 00:05:12,840 Then you step up to the next level multi agent systems 99 00:05:12,840 --> 00:05:16,530 This is where it gets really powerful and honestly quite transformative instead of 100 00:05:16,530 --> 00:05:19,640 one AI doing everything you orchestrate workflows across multiple 101 00:05:19,640 --> 00:05:23,240 Coordinated agents think of it like building a specialized AI 102 00:05:23,240 --> 00:05:27,480 Okay, how does that work? Well each agent might have a specific role 103 00:05:27,480 --> 00:05:32,120 maybe one agent is great at research another excels at summarizing information a 104 00:05:32,120 --> 00:05:37,170 Third drafts content and maybe a fourth reviews at all. They communicate they 105 00:05:37,170 --> 00:05:38,160 delegate tasks 106 00:05:38,160 --> 00:05:42,100 They collaborate to tackle problems far more complex than any single agent could 107 00:05:42,100 --> 00:05:43,000 manage on its own 108 00:05:43,000 --> 00:05:48,760 This opens the door to automating entire business processes that used to need a lot 109 00:05:48,760 --> 00:05:49,900 of human coordination 110 00:05:49,900 --> 00:05:55,040 Or really complex software integrations that distinction really helps visualize it 111 00:05:55,040 --> 00:05:58,170 Yeah, so just to make sure I'm getting this you could literally build say a 112 00:05:58,170 --> 00:06:02,440 customer service chatbot that answers questions using your entire library of PDF 113 00:06:02,440 --> 00:06:07,130 Manuals and word docs always giving the right info and on the multi-agent side you 114 00:06:07,130 --> 00:06:08,320 could build a team of AIs 115 00:06:08,320 --> 00:06:12,840 Maybe one does market research another analyzes social media sentiment a third 116 00:06:12,840 --> 00:06:16,000 writes a summary report and a fourth even drafts some marketing ideas 117 00:06:16,000 --> 00:06:19,240 That's yeah, that's a whole different level of automation. Exactly, right? 118 00:06:19,240 --> 00:06:22,800 The potential scope is huge really just limited by the workflow you design 119 00:06:22,800 --> 00:06:25,400 It moves beyond simple task automation towards 120 00:06:25,400 --> 00:06:28,680 Intelligent orchestration. Okay, that sounds 121 00:06:28,680 --> 00:06:34,560 Incredibly powerful almost maybe a little scary like how do you maintain control? 122 00:06:34,560 --> 00:06:38,130 How does flow wise make sure things are transparent and you can have human 123 00:06:38,130 --> 00:06:38,720 oversight? 124 00:06:38,720 --> 00:06:42,750 You definitely don't want a black box AI running wild especially for important 125 00:06:42,750 --> 00:06:43,280 stuff, right? 126 00:06:43,280 --> 00:06:48,080 That's a critical point and yeah a very valid concern flow wise tackles this head-on 127 00:06:48,080 --> 00:06:49,720 with a few key features 128 00:06:49,840 --> 00:06:54,840 Focused on control and transparency first off. There's human in the loop or 129 00:06:54,840 --> 00:07:00,430 HITL this is huge. It lets humans review tasks that agents perform during the 130 00:07:00,430 --> 00:07:02,300 process within the feedback loop 131 00:07:02,300 --> 00:07:05,960 Okay, so you can step in. Absolutely. Yeah, imagine your AI support agent is about 132 00:07:05,960 --> 00:07:09,040 to give slightly outdated advice with HITL 133 00:07:09,040 --> 00:07:13,740 A human can catch that specific output before it goes out correct it and this is 134 00:07:13,740 --> 00:07:13,820 key 135 00:07:13,820 --> 00:07:18,000 The AI learns from that correction instantly. It's not just fixing one mistake 136 00:07:18,000 --> 00:07:22,580 It's continuously refining the agents performance based on real-time human guidance 137 00:07:22,580 --> 00:07:25,820 that helps bridge the gap between AI efficiency and human nuance 138 00:07:25,820 --> 00:07:30,440 Then beyond stepping in flow eyes offer strong observability through what they call 139 00:07:30,440 --> 00:07:33,700 execution traces execution traces like a log file 140 00:07:33,700 --> 00:07:38,180 Sort of but much more detailed. It's like a full step-by-step recording of the AI's 141 00:07:38,180 --> 00:07:39,740 decision-making process 142 00:07:39,740 --> 00:07:44,050 You see exactly how it reached an answer what data it looked at which tools it used 143 00:07:44,050 --> 00:07:45,980 the sequence of its reasoning 144 00:07:46,140 --> 00:07:51,300 This is invaluable for debugging for optimizing performance and for auditing making 145 00:07:51,300 --> 00:07:52,420 sure it's doing what you expect 146 00:07:52,420 --> 00:07:56,220 It even integrates with standard tools like Prometheus and open telemetry for 147 00:07:56,220 --> 00:07:58,400 deeper system monitoring 148 00:07:58,400 --> 00:08:02,620 Which is essential for enterprise use and finally, even though it's visual and easy 149 00:08:02,620 --> 00:08:05,580 to use flow wise is also super developer friendly 150 00:08:05,580 --> 00:08:10,090 It's not some closed little garden the agents you design visually you can extend 151 00:08:10,090 --> 00:08:13,240 them and integrate them smoothly into your existing apps and systems 152 00:08:13,240 --> 00:08:17,600 Oh, okay. So plays nice with other things. Definitely. It has robust API's 153 00:08:17,600 --> 00:08:22,520 Software development kits SDKs for both TypeScript and Python and even embeddable 154 00:08:22,520 --> 00:08:23,400 chat widgets 155 00:08:23,400 --> 00:08:27,870 You can drop onto a website so you can design the core AI logic visually and flow 156 00:08:27,870 --> 00:08:28,160 wise 157 00:08:28,160 --> 00:08:32,060 Which is fast and then plug that intelligence into your website your internal tools 158 00:08:32,060 --> 00:08:32,840 your mobile app 159 00:08:32,840 --> 00:08:36,360 Whatever you're building it gives developers the best of both worlds visual speed 160 00:08:36,360 --> 00:08:38,080 plus coding flexibility 161 00:08:38,080 --> 00:08:42,170 So yeah, definitely not a pleck box. You can see inside guide it and plug it into 162 00:08:42,170 --> 00:08:43,340 pretty much anything 163 00:08:43,340 --> 00:08:47,300 You're already using that makes it way more practical for actual businesses, which 164 00:08:47,300 --> 00:08:48,160 leads me to ask 165 00:08:48,160 --> 00:08:53,070 How are businesses using flow wise right now? What kind of real-world impact are we 166 00:08:53,070 --> 00:08:54,120 seeing out there? 167 00:08:54,120 --> 00:08:58,680 Oh, the impact is definitely tangible and it really highlights those benefits speed 168 00:08:58,680 --> 00:09:00,720 efficiency opening up new AI 169 00:09:00,720 --> 00:09:05,520 Possibilities their customer stories show this well, for example uneq. They do 170 00:09:05,520 --> 00:09:07,960 these amazing digital human experiences 171 00:09:08,120 --> 00:09:12,660 They significantly cut down the resources needed to deploy their AI brains by using 172 00:09:12,660 --> 00:09:13,280 flow eyes 173 00:09:13,280 --> 00:09:15,960 That means faster launches and lower costs for them 174 00:09:15,960 --> 00:09:20,830 Another one kumit guitar iFleet used flow eyes to seriously boost the quality and 175 00:09:20,830 --> 00:09:24,080 efficiency of a new co-pilot feature in their fleet 176 00:09:24,080 --> 00:09:27,520 Management product so it's great for enhancing existing things to right now just 177 00:09:27,520 --> 00:09:28,480 building from scratch 178 00:09:28,480 --> 00:09:33,450 Exactly and BTS digital they sped up their whole initiative around build your own 179 00:09:33,450 --> 00:09:33,680 AI 180 00:09:33,880 --> 00:09:38,290 Assistance easily showing how flow wise helps companies become more self-sufficient 181 00:09:38,290 --> 00:09:38,720 with AI 182 00:09:38,720 --> 00:09:42,720 There's great quote from David Mikado a senior director of DX and AI 183 00:09:42,720 --> 00:09:46,280 He said flow eyes has truly changed how we approach AI 184 00:09:46,280 --> 00:09:50,700 It's simple enough to prototype an idea in minutes yet powerful enough to take all 185 00:09:50,700 --> 00:09:51,960 the way to production 186 00:09:51,960 --> 00:09:55,400 That's a pretty strong statement, you know, yeah that sums it up perfectly 187 00:09:55,400 --> 00:09:59,720 Prototype fast scale up and it's not just the official stories 188 00:09:59,720 --> 00:10:03,850 The open source community around flow wise is buzzing on Twitter and places like 189 00:10:03,850 --> 00:10:04,160 that 190 00:10:04,160 --> 00:10:08,530 You see people doing amazing things like creating super specific AI teaching 191 00:10:08,530 --> 00:10:09,480 personas building 192 00:10:09,480 --> 00:10:14,440 Multimodal chatbots that handle both text and images making telegram bots for real-time 193 00:10:14,440 --> 00:10:16,080 bus info using live data 194 00:10:16,080 --> 00:10:21,280 Even no code sequel chatbots. So business users can query databases just by asking 195 00:10:21,280 --> 00:10:22,720 questions in plain English 196 00:10:22,720 --> 00:10:23,800 Wow, that's diverse 197 00:10:23,800 --> 00:10:27,680 It really is it shows the platform is enabling all sorts of innovation from 198 00:10:27,680 --> 00:10:29,680 education to transport to business 199 00:10:30,000 --> 00:10:33,580 Intelligence, it really does sound like a game-changer then speeding things up 200 00:10:33,580 --> 00:10:37,500 making AI more efficient and unlocking these capabilities for people who aren't 201 00:10:37,500 --> 00:10:38,280 necessarily 202 00:10:38,280 --> 00:10:42,720 AI-phd's it lets you test ideas super fast and then actually build them out 203 00:10:42,720 --> 00:10:45,480 properly and having such a strong community 204 00:10:45,480 --> 00:10:49,960 What was it over 42,000 stars on github? That's huge for an open source project 205 00:10:49,960 --> 00:10:52,060 shows. It's alive and well-supported 206 00:10:52,060 --> 00:10:56,600 Absolutely, that community is vital and importantly flow eyes is also built to work 207 00:10:56,600 --> 00:11:00,890 Well with other popular AI tools and frameworks that you know, many listeners might 208 00:11:00,890 --> 00:11:04,040 already be using we're talking smooth integrations with big LLM providers 209 00:11:04,040 --> 00:11:06,880 Like AWS bedrock open a eyes GPT models 210 00:11:06,880 --> 00:11:10,300 frameworks like Lang chain for managing complex AI steps and 211 00:11:10,300 --> 00:11:13,990 Essential vector databases like lemma index or milvis for that smart knowledge 212 00:11:13,990 --> 00:11:14,560 retrieval 213 00:11:14,560 --> 00:11:18,830 It even works with multi-agent systems frameworks like autogen so flow eyes fits 214 00:11:18,830 --> 00:11:20,920 neatly into the bigger AI ecosystem 215 00:11:20,920 --> 00:11:25,130 It doesn't force you into some isolated silo. You can leverage your existing tech 216 00:11:25,130 --> 00:11:25,280 stack 217 00:11:25,280 --> 00:11:28,680 okay, this all sounds amazing incredibly powerful, but 218 00:11:28,680 --> 00:11:33,800 Often tools like this, especially the ones ready for big companies come with a 219 00:11:33,800 --> 00:11:35,120 pretty hefty price tag 220 00:11:35,120 --> 00:11:38,800 How accessible is flow wise really for someone just starting out? 221 00:11:38,800 --> 00:11:43,520 Maybe a small team wanting to experiment that's another area where flow eyes really 222 00:11:43,520 --> 00:11:45,400 scores points accessibility and scalability 223 00:11:45,400 --> 00:11:50,390 You can literally as they say start building for free their free tier gives you 224 00:11:50,390 --> 00:11:51,640 enough to really get going 225 00:11:51,640 --> 00:11:55,820 Two flows in assistance 100 predictions a month a bit of storage plus access to all 226 00:11:55,820 --> 00:11:57,000 the community support 227 00:11:57,000 --> 00:12:01,000 It's perfect for just trying things out learning the ropes building a proof of 228 00:12:01,000 --> 00:12:02,480 concept all with zero cost 229 00:12:02,480 --> 00:12:04,600 Okay, that's genuinely free to start 230 00:12:04,600 --> 00:12:05,200 Yeah 231 00:12:05,200 --> 00:12:08,960 And then if you need more they have starter and pro tiers that are really 232 00:12:08,960 --> 00:12:11,360 reasonably priced for individuals or smaller teams 233 00:12:11,360 --> 00:12:15,890 It scales up affordably, but crucially it's also built to be enterprise ready for 234 00:12:15,890 --> 00:12:17,580 bigger organizations 235 00:12:17,580 --> 00:12:21,750 It has all the features you'd expect support for like over a hundred different 236 00:12:21,750 --> 00:12:23,000 large language models 237 00:12:23,000 --> 00:12:27,680 Tons of embedding options various vector databases plus robust deployment options 238 00:12:27,680 --> 00:12:29,640 on your own servers or in the cloud 239 00:12:29,640 --> 00:12:34,140 And it's designed for horizontal scaling so it can handle massive workloads and 240 00:12:34,140 --> 00:12:34,840 stay reliable 241 00:12:34,840 --> 00:12:39,140 It truly spans from you know a hobby project right up to industrial scale AI 242 00:12:39,140 --> 00:12:44,240 So you can genuinely start for nothing play around learn build something cool 243 00:12:44,240 --> 00:12:48,720 And if it takes off you have a clear path to scale it up massively that level of 244 00:12:48,720 --> 00:12:50,040 flexibility is incredible 245 00:12:50,040 --> 00:12:53,420 It really empowers everyone from beginners to big enterprises 246 00:12:53,420 --> 00:12:56,920 Wow what an incredible deep dive into flow eyes today 247 00:12:56,920 --> 00:13:00,960 It really does seem to stand out by making this complex world of AI agent building 248 00:13:00,960 --> 00:13:02,900 so much more accessible so visual 249 00:13:02,900 --> 00:13:07,160 It's clearly empowering people and organizations to innovate much faster whether 250 00:13:07,160 --> 00:13:08,200 they're seasoned developers 251 00:13:08,200 --> 00:13:12,070 Or you know just starting to explore what AI can do and that really brings up an 252 00:13:12,070 --> 00:13:14,040 interesting question for everyone listening 253 00:13:14,040 --> 00:13:17,240 Doesn't it thinking about your own work your own life? 254 00:13:17,240 --> 00:13:22,150 What task or challenge could you transform with an AI agent especially one that you 255 00:13:22,150 --> 00:13:24,300 could actually build and control yourself? 256 00:13:24,300 --> 00:13:28,610 Maybe without needing to become a coding expert first just imagine the 257 00:13:28,610 --> 00:13:31,760 possibilities flow wise could unlock for you personally 258 00:13:31,760 --> 00:13:35,860 That's a fantastic thought to leave people with thank you so much for joining us on 259 00:13:35,860 --> 00:13:37,520 this deep dive into flow wise today 260 00:13:37,520 --> 00:13:41,300 We really hope you learn something useful and maybe feel inspired to check out 261 00:13:41,300 --> 00:13:43,960 visual AI building and before we go a huge thing 262 00:13:44,040 --> 00:13:48,870 Thank you once again to our sponsor safe server.de for helping make these deep dives 263 00:13:48,870 --> 00:13:50,480 happen do check them out at 264 00:13:50,480 --> 00:13:55,940 www.safeserver.de to see how they can support your digital transformation until 265 00:13:55,940 --> 00:13:56,640 next time