WEBVTT 00:00.080 --> 00:05.760 Well, this is a familiar screen, and it's nice that it looks pretty similar to what we saw before. 00:05.880 --> 00:06.960 I'll just show you a few things. 00:06.960 --> 00:11.760 If you come down to settings here, then you can go to like usage and Plan. 00:11.800 --> 00:15.840 This is where you can see, uh, what kind of things that you're working on. 00:16.040 --> 00:20.760 When you get that email with the key, you press Enter Activation Key here to activate or to basically 00:20.760 --> 00:22.120 unlock a few extra features. 00:22.360 --> 00:27.840 And down here it will tell you the version number that you're running, uh, for, for the pros, the 00:27.880 --> 00:29.040 know about this kind of stuff. 00:29.040 --> 00:33.480 When you specify the Docker container, you can put the container name colon. 00:33.520 --> 00:40.920 Next, if you want to be on the sort of latest upcoming version of which, which right now is 2.2. something 00:41.160 --> 00:46.720 for me which but 2.1.3 is, is the sort of the latest official version. 00:47.040 --> 00:50.720 And there's various settings here that that might turn out to be useful for you. 00:50.720 --> 00:53.080 So it's worth knowing about this. 00:53.320 --> 00:58.760 Uh, and if we go back here we are again on this first screen you'll see overview. 00:58.800 --> 01:05.160 There's also a new beta chat version that allows you to chat with the AI to to construct, to work your 01:05.160 --> 01:08.400 way around in, rather than having AI within your workflow. 01:08.720 --> 01:10.640 And there's the same templates. 01:10.640 --> 01:15.120 If you remember that this is a great place to start should you wish to have one that's kind of canned, 01:15.120 --> 01:19.440 that you can use to explore things that other people have built, and we're building from the ground 01:19.440 --> 01:25.520 up in hours just so that you, you really get to understand, um, the plumbing of it and, uh, then, 01:25.520 --> 01:28.360 then then there's the main sections here. 01:28.360 --> 01:33.920 The overview is where we see all of our workflows and where we can get to begin our first workflow, 01:33.920 --> 01:34.880 which is what we'll do now. 01:34.920 --> 01:35.440 Okay. 01:35.640 --> 01:37.040 I pressed start from scratch. 01:37.280 --> 01:38.080 Here we go. 01:38.480 --> 01:39.040 We are. 01:39.080 --> 01:39.840 Let's get rid of that. 01:39.840 --> 01:42.440 There we are looking at a very familiar screen. 01:42.440 --> 01:43.160 It looks the same. 01:43.160 --> 01:44.320 This is the canvas. 01:44.360 --> 01:47.520 Let's see if pressing the plus and minus expands and contracts. 01:47.520 --> 01:48.680 For sure it does. 01:48.960 --> 01:51.240 And see if pressing Tab gets us this. 01:51.280 --> 01:52.320 Of course it does. 01:52.360 --> 01:56.840 We can use our favorite on chat message and then press escape to go back. 01:56.840 --> 01:59.110 Click and press escape to come back here. 01:59.190 --> 02:02.870 We can, uh, press here to to to resize. 02:02.910 --> 02:03.870 Clean everything up. 02:03.910 --> 02:06.030 When a chat message is received, what do we want to do? 02:06.030 --> 02:08.630 Well, press tab to bring up the screen AI. 02:09.190 --> 02:11.830 AI agent and escape. 02:12.030 --> 02:15.870 And we are faced with the usual screen that we know so well. 02:16.110 --> 02:20.710 And it's time for our first chat model when running running, self-hosted. 02:20.710 --> 02:26.270 So I'm going to press here and I'm going to recommend that we go with Open Router, which is what we 02:26.310 --> 02:32.070 set up at the very start, if you remember, because we we can call an open source model that's running 02:32.070 --> 02:32.790 on the cloud. 02:32.790 --> 02:36.670 And since we're feeling like we're we're running this ourselves, open source feels more legit. 02:36.830 --> 02:42.750 But you could equally you could use a llama and run something actually on your computer should you wish. 02:42.750 --> 02:48.030 And if you're familiar with Aloha, maybe you've taken my AI engineering courses, in which case you 02:48.030 --> 02:50.150 can just look at the llama chat model. 02:50.190 --> 02:53.990 Now you probably need to know enough about Docker to to to mess around. 02:54.030 --> 02:55.270 You'll need to run you. 02:55.270 --> 02:56.630 Can you have two choices. 02:56.630 --> 03:02.470 You could run a llama locally on your host, and then just map the port so that you can see that from 03:02.470 --> 03:04.110 within the Docker container. 03:04.350 --> 03:11.110 Or you can use an approach that they recommend on the website, which is to use one of their ready made 03:11.110 --> 03:18.430 Docker compose files, which then will give you a container for for a llama and a container for an and 03:18.430 --> 03:19.670 map between them. 03:19.670 --> 03:25.150 Now the disadvantage of that second approach, whilst it is a very effective approach, a downside is 03:25.150 --> 03:29.030 if you're on a mac, you wouldn't be able to take advantage of your Mac's GPU. 03:29.230 --> 03:34.950 To do that, you would need to run a llama on your host computer, but if you're on a PC, then you 03:34.950 --> 03:35.150 could. 03:35.150 --> 03:37.630 You could do it that way should you wish to do it the technical way. 03:37.630 --> 03:45.070 But for now, we are going to focus on the open router model, because we're going to make a connection 03:45.070 --> 03:47.630 to open router and run open source models in the cloud. 03:47.630 --> 03:52.430 And here we are looking at the credential to connect with. 03:52.710 --> 03:54.230 And we know these credentials well. 03:54.230 --> 03:56.100 We say create new credential. 03:56.140 --> 03:56.660 Up it comes. 03:56.660 --> 03:58.020 It needs an API key. 03:58.140 --> 03:59.060 And this is where I go. 03:59.060 --> 04:01.820 And I copy my API key from Open Router. 04:01.820 --> 04:03.180 And I paste it in here. 04:03.180 --> 04:08.740 And I know to be careful I think it starts src o dash v1 I think. 04:08.860 --> 04:09.900 And I've got it right. 04:09.900 --> 04:13.580 I pasted it in and I'm careful so that I'm not going to be in trouble. 04:13.620 --> 04:17.620 And when I press testing we get the connection was tested successfully. 04:17.620 --> 04:18.420 So did you. 04:18.620 --> 04:21.020 And so now we come back here. 04:21.060 --> 04:21.580 No thank you. 04:21.620 --> 04:22.260 LastPass. 04:22.340 --> 04:25.060 Uh, we have set up our chat model. 04:25.060 --> 04:27.940 We're also going to add in a simple memory. 04:28.260 --> 04:31.820 There we have it and we can have our first chat. 04:31.980 --> 04:32.420 Uh oh. 04:32.420 --> 04:33.820 But we have to select the model first. 04:33.820 --> 04:34.940 I didn't select the model. 04:35.100 --> 04:38.060 Uh it has GPT four or mini but we don't want that at all. 04:38.060 --> 04:40.220 We want a model like deep seq. 04:40.980 --> 04:49.740 Uh, we will choose V3 point two, which for me is the latest and greatest version of the deep seq model 04:49.740 --> 04:52.140 from Chinese startup deep seq AI. 04:52.140 --> 04:52.460 I. 04:52.820 --> 04:55.100 All right, so back we go. 04:55.140 --> 04:57.220 It's time for us to chat with Deep Seek. 04:57.260 --> 05:04.900 Okay, I press open chat and I say hi there and we'll get back something. 05:04.940 --> 05:05.500 Hello? 05:05.540 --> 05:06.700 How can I help you today? 05:07.460 --> 05:12.420 And of course, the great thing about this is that it's all running on the computer. 05:12.660 --> 05:17.500 Deep within the computer, within a Docker container, it's all running beautifully. 05:17.540 --> 05:20.420 It wasn't that easy to set up, but it also wasn't that hard. 05:20.420 --> 05:26.140 And when you think about the enormity of what we're doing, we're running the whole platform on our 05:26.140 --> 05:26.700 computer. 05:26.700 --> 05:31.740 And it's the same if you're on a PC or on a mac, and we're talking through to through APIs to deep 05:31.780 --> 05:35.620 seq and Open Router, and it's all being hanging together on our computer. 05:35.660 --> 05:36.660 That's pretty cool. 05:36.820 --> 05:41.140 Uh, and of course, as I say, you could just deploy this to the cloud yourself, running on your computer 05:41.140 --> 05:43.020 in the same way anywhere. 05:43.180 --> 05:45.580 And it's all working just great. 05:45.660 --> 05:48.140 That is our first AI agent. 05:48.260 --> 05:49.300 Running. 05:49.540 --> 05:53.050 Running locally, self-hosted on our computers. 05:53.050 --> 05:57.250 And of course, it wouldn't really be a agent if we didn't give it a tool. 05:57.730 --> 06:05.050 Uh, because an agent is an LLM in a loop equipped with tools to achieve a goal. 06:05.370 --> 06:06.650 Uh, you remember that? 06:06.650 --> 06:10.290 So we are going to use our old friend market stack. 06:10.290 --> 06:11.210 Remember that? 06:11.250 --> 06:12.730 A nice one to start with. 06:12.770 --> 06:15.170 We have to connect it with a credential. 06:15.170 --> 06:17.490 We say create new credential. 06:17.490 --> 06:19.930 We paste in an API key and you have to get it right. 06:19.970 --> 06:22.010 You have to copy that API key, press save. 06:22.450 --> 06:24.170 We don't get a green thing with this one. 06:24.170 --> 06:26.010 But but I know that I got it right. 06:26.170 --> 06:27.410 Uh, okay. 06:27.450 --> 06:29.450 And now you remember there's some fiddly stuff here. 06:29.610 --> 06:34.850 We want to say that the model can choose which ticker to use. 06:35.090 --> 06:37.050 And then we have to add a filter. 06:37.050 --> 06:38.090 Latest. 06:38.090 --> 06:40.130 And I always forget to turn that on. 06:40.130 --> 06:42.210 You have to turn on the latest filter. 06:42.250 --> 06:43.410 Otherwise it doesn't work. 06:43.570 --> 06:45.330 So that is set up. 06:45.330 --> 06:48.490 And now we have something new to test. 06:48.650 --> 06:49.610 Let's try it. 06:50.050 --> 06:54.730 And I should probably mention that deep seek and open router aren't super reliable at tool calling. 06:54.730 --> 06:58.490 Something funny sometimes happens, so you have to sometimes try a few times, but we'll see. 06:58.690 --> 07:03.090 Uh, I don't know if it's a deep sea problem or an open router problem or a combo between the two. 07:03.170 --> 07:04.570 But anyway, let's give this a try. 07:04.610 --> 07:06.170 So I'm going to say, uh, hi there. 07:07.970 --> 07:08.770 Off it goes. 07:08.770 --> 07:09.970 Stuff happens. 07:10.010 --> 07:10.730 Thinking. 07:13.570 --> 07:14.050 Hello. 07:14.090 --> 07:16.450 I'm here to help you with financial data. 07:17.610 --> 07:20.170 Uh, it's already so it's looked at the tool. 07:20.170 --> 07:21.770 It's found out what it can do. 07:21.930 --> 07:25.330 Uh, when I say it's looked at the tool that was in the system prompt already, the description of the 07:25.330 --> 07:25.690 tool. 07:25.850 --> 07:28.570 And so it's telling us that that's what it can do. 07:28.690 --> 07:30.610 Uh, what stock would you like information about? 07:30.730 --> 07:36.210 Please tell me the price of Apple. 07:37.850 --> 07:38.730 Let's let it do its thing. 07:38.770 --> 07:40.290 See if it calls the tool here. 07:40.290 --> 07:41.370 It didn't call the tool. 07:41.370 --> 07:42.890 You see, you get this thing. 07:43.090 --> 07:47.850 Uh, it's it's responding with, um, uh, function invoke. 07:48.320 --> 07:49.120 Let's try that. 07:49.120 --> 07:55.880 Just simply a please tell me the stock price of Apple. 07:56.000 --> 07:57.200 Let's just try it again. 08:00.480 --> 08:02.760 Look to see whether that tool is getting called or not. 08:04.000 --> 08:04.360 There we go. 08:04.400 --> 08:05.840 I just saw the tool got called. 08:10.880 --> 08:13.360 And it's thinking maybe a long answer here. 08:16.080 --> 08:18.920 Still thinking, but you can see it's got that information right there. 08:21.000 --> 08:21.680 There we go. 08:22.240 --> 08:24.720 Uh, maybe I need to make this a bit bigger for you. 08:24.760 --> 08:30.320 Hang on and expand this up based on the latest available data. 08:30.520 --> 08:32.320 And there is the information. 08:32.320 --> 08:35.280 So for whatever reason, we had to ask the question twice. 08:35.400 --> 08:40.600 Uh, if you wanted to, you could put in, like, an if, uh, node there and put something in to make 08:40.640 --> 08:42.840 sure that it repeats until you get back a good answer. 08:42.960 --> 08:48.600 Uh, if you want to get fancy, but you should experiment with different models, you'll almost certainly 08:48.600 --> 08:55.080 find that it's just just a problem using deep seq 3.2 through Open Router, that it's not always reliably 08:55.080 --> 08:59.480 calling the tool, and there'll be other models that will do it more reliably. 08:59.480 --> 09:03.760 But this is showing you that we've been able to use an open source model. 09:03.840 --> 09:10.320 And then using driven from our self-hosted instance of Nate and using a tool with an agent. 09:10.320 --> 09:11.720 And it's all work great. 09:11.720 --> 09:17.640 And just so we can go with one that's bound to be, uh, completely bulletproof, let's go into our 09:17.880 --> 09:29.720 model here and let's just change this to be open AI, GPT oss 120 be, uh, OpenAI's top open source 09:29.720 --> 09:33.600 model running through open router so it won't be connecting with OpenAI at all. 09:33.640 --> 09:40.280 It's their published weights, uh, being run by but on open routers or one of open routers providers 09:40.280 --> 09:41.160 infrastructure. 09:41.360 --> 09:43.360 Um, okay, let's go back. 09:44.320 --> 09:45.990 Let's, uh, save that. 09:45.990 --> 09:46.910 Go back here. 09:46.910 --> 09:49.990 I'm going to clear this out, and then we will try this one more time. 09:49.990 --> 09:51.710 And so I say hi there. 09:53.990 --> 09:54.550 Hello. 09:54.550 --> 09:55.670 How can I help you today? 09:55.710 --> 09:56.670 It's familiar, isn't it? 09:56.710 --> 10:02.070 It sounds very like, uh, GPT, uh, I'd like to know. 10:04.590 --> 10:05.750 The closing price. 10:05.910 --> 10:06.910 I just say the price. 10:07.230 --> 10:09.910 The price of Apple stock. 10:10.910 --> 10:11.710 Let's see what happens. 10:11.710 --> 10:12.470 Off it goes. 10:12.470 --> 10:14.150 You can see the tool is called. 10:14.310 --> 10:18.390 And it says the most recent closing price for Apple in the data we have is 270. 10:18.630 --> 10:23.670 Uh and so of course OSS super reliable, very bulletproof. 10:23.830 --> 10:27.830 Uh, so you could try this out a few times and then you have another tool, a jankiness. 10:27.870 --> 10:29.990 It should all work very nicely indeed. 10:30.230 --> 10:34.750 And so give this a try and experiment with, with other things you can do with it. 10:34.750 --> 10:39.790 Use some of the things from earlier in the course and confirm that with the exception of webhooks and 10:39.830 --> 10:43.430 OAuth, uh, things will be basically identical. 10:43.430 --> 10:44.270 Give it a try.