WEBVTT 00:00.160 --> 00:03.480 And so with that, I'll just move on to talk about what Autogen actually is. 00:03.880 --> 00:06.640 Let's remind you, these are the frameworks we've been going through. 00:06.840 --> 00:12.720 Autogen the last of the frameworks, uh, before, because MCP isn't really a framework. 00:13.000 --> 00:13.880 Uh, okay. 00:13.920 --> 00:15.560 So what is autogen? 00:15.600 --> 00:23.680 So Autogen, as with the others, it's a bunch of different things all wrapped under one autogen umbrella 00:23.720 --> 00:24.320 name. 00:24.680 --> 00:27.920 It's first of all, something called Autogen core. 00:28.400 --> 00:34.760 And Autogen core is something which is like, uh, it's agnostic to the, to the framework that the 00:34.760 --> 00:37.200 actual agents and llms that are being used. 00:37.200 --> 00:45.440 It is like a general generic framework for building scalable multi-agent systems, something that manages 00:45.440 --> 00:51.160 things like messaging between agents, even if they're distributed in different places. 00:51.280 --> 00:54.440 So it's like a fabric for agents to, to run within. 00:54.440 --> 00:55.320 It's something different. 00:55.320 --> 00:59.800 It's got some stuff in common with, with an idea like Landgraaf, but much, much simpler. 01:00.000 --> 01:06.840 It's got uh, but but basically it's like a runtime, an agent runtime for running agents together. 01:06.840 --> 01:10.850 That is autogen core autogen agent chat. 01:10.890 --> 01:11.770 It's quite a mouthful. 01:11.770 --> 01:13.250 Autogen agent chat. 01:13.530 --> 01:21.250 That is a framework that is going to be very familiar to you, because it's very similar to OpenAI agents 01:21.290 --> 01:23.010 SDK and to crew. 01:23.170 --> 01:30.410 It is a lightweight, simple abstraction for putting together llms in an agent construct to allow them 01:30.410 --> 01:36.450 to use tools to allow them to interact with each other, that is, Autogen agent chat and then built 01:36.450 --> 01:38.290 on top of Autogen agent chat. 01:38.330 --> 01:40.450 Agent chat itself is built on top of core. 01:40.690 --> 01:43.730 Built on top of agent chat are a couple of offerings. 01:43.730 --> 01:50.010 There's something called studio, which we've heard that name a few times now that is Autogen low code, 01:50.050 --> 01:53.050 no code app for sort of constructing things visually. 01:53.290 --> 01:57.090 And there's also something called Magentic one, which is like a product. 01:57.090 --> 01:58.810 It is a console. 01:59.090 --> 02:02.970 It's something you use in the command line, which can take care of things for you and do them. 02:02.970 --> 02:04.130 So it's like an application. 02:04.170 --> 02:11.290 It's a it's a application to run an agent framework that's out of the box that's canned, that runs 02:11.290 --> 02:12.970 from the command line. 02:13.170 --> 02:17.170 So these are the various bits and pieces that make up Autogen. 02:17.490 --> 02:21.690 And uh, all of this is part of open source. 02:21.690 --> 02:26.130 It's part of it's sort of managed by Microsoft Research, but it's available. 02:26.170 --> 02:28.930 There are people who contribute from all all over the place. 02:29.250 --> 02:35.410 And it's an interesting point that whilst when we talked about crew and and Lang Chain, we saw that 02:35.410 --> 02:40.690 there was perhaps the possibility that that monetisation commercialisation was driving some of their 02:40.690 --> 02:41.650 product roadmap. 02:42.010 --> 02:44.250 There's a different angle here with Microsoft. 02:44.290 --> 02:48.130 This is all a Microsoft research for the community kind of thing. 02:48.290 --> 02:55.050 And so you don't necessarily see that things like Studio and Magenta X1, Magenta studio is considered 02:55.090 --> 02:57.170 like a research environment. 02:57.210 --> 03:00.050 It's not it's not considered ready for production. 03:00.050 --> 03:02.850 And they state that very clearly all over it. 03:02.850 --> 03:04.570 So it is a different positioning. 03:04.570 --> 03:08.730 This is a big open source contribution, uh, through and through. 03:09.610 --> 03:15.810 And of course, no surprise what we're going to be focusing on is going to be the, uh, agent, uh, 03:15.930 --> 03:20.650 Autogen core and the Autogen agent chat, those two parts of the puzzle. 03:20.690 --> 03:22.170 We're less interested in the low code. 03:22.170 --> 03:23.570 No code, because we're coders. 03:23.610 --> 03:24.410 That's what we do. 03:24.650 --> 03:28.450 Uh, and, uh, magenta one, it seems, seems like a cool thing to try out, but it's. 03:28.490 --> 03:32.820 I guess it's a bit like, uh, someone building what we've already built. 03:33.020 --> 03:40.380 Uh, in the end of last week, uh, with with our co-worker that we have our sidekick. 03:40.420 --> 03:41.660 Our personal sidekick. 03:42.180 --> 03:47.980 Uh, so one more point to make here is that we're going to be focusing mostly on Autogen agent chat. 03:48.140 --> 03:55.860 Uh, because that is the sort of direct comparable with Crewe and with OpenAI agents SDK and with the 03:55.900 --> 03:58.500 sort of agent interactions part of Landgraaf. 03:58.540 --> 04:03.580 We will also look at Autogen cause some pretty cool things about it, but but more to give you some 04:03.580 --> 04:06.380 interest in it and to experiment a bit. 04:06.620 --> 04:08.580 Um, not as much as agent chat. 04:08.580 --> 04:10.860 And even when we do, we'll be using agent chat as well. 04:11.300 --> 04:12.820 So that's the lay of the land. 04:13.260 --> 04:19.260 And to talk about the concepts behind, uh, the, the Autogen framework and particularly behind Agent 04:19.260 --> 04:21.060 Chat that we're now going to talk about. 04:21.340 --> 04:23.420 What are the building blocks here? 04:23.460 --> 04:25.220 Well, they're going to be very familiar to you. 04:25.260 --> 04:28.380 They are straightforward and things that we've already met before. 04:28.620 --> 04:31.540 So they have a concept called models. 04:31.540 --> 04:36.460 And their concept called models is similar to Llms that we've seen in other platforms. 04:36.860 --> 04:41.500 Messages I guess it's new that they identify this as a core concept. 04:41.500 --> 04:49.220 Messages that can represent messages between agents, or it can represent events that happen within 04:49.220 --> 04:52.940 an agent's interactions, which which really is referring to, like calling tools. 04:52.940 --> 04:55.140 Those are also considered messages. 04:55.140 --> 04:58.860 And so we'll see some some stuff about messages that can be from a user to the model. 04:58.860 --> 05:00.460 It can be between agents. 05:00.460 --> 05:03.380 It can be within an agent when it's calling tools. 05:03.380 --> 05:06.860 But those are called messages and agent. 05:06.860 --> 05:08.300 It's going to be what we're used to. 05:08.340 --> 05:14.500 It's something which has a model behind it and something which is able to carry out a series of tasks 05:14.860 --> 05:20.940 on behalf of, of a request, on behalf of a user or another agent and then teams again. 05:20.980 --> 05:21.940 That's familiar. 05:21.940 --> 05:23.740 This is like a crew from crew. 05:23.900 --> 05:28.540 Uh, it'd be a group of agents that can interact to achieve some goal. 05:28.860 --> 05:32.460 So these are the sort of core building blocks and there are others. 05:32.500 --> 05:34.940 But but I think these are probably the ones to introduce. 05:34.940 --> 05:40.820 And for today we're going to look at the first three models messages and agents very briefly very quickly 05:40.820 --> 05:42.060 to set up a quick example. 05:42.060 --> 05:42.980 But we'll do something new. 05:43.020 --> 05:47.580 We'll involve some SQL in it because I know that's that's going to be useful for some people. 05:47.580 --> 05:48.700 So let's do that right now.