WEBVTT 00:00.200 --> 00:02.800 And welcome to week five, day four. 00:02.960 --> 00:05.120 We're going to talk more about Autogen core. 00:05.680 --> 00:11.040 You remember Autogen core, the fundamental the bottom layer of the stack for Autogen. 00:11.320 --> 00:13.280 And it's an interaction framework. 00:13.280 --> 00:17.320 It's responsible for worrying about how agents play together. 00:17.360 --> 00:22.360 It doesn't care about how the agents are implemented, although it works well with with agent chat. 00:22.800 --> 00:30.120 It's somewhat analogous to landgraaf, but except rather than focusing on repeatable workflows, it's 00:30.120 --> 00:33.760 focused on the interactions between diverse agents. 00:34.840 --> 00:38.640 And in particular, we talked about two different types of runtime the standalone. 00:38.680 --> 00:39.680 The distributed. 00:40.040 --> 00:45.080 Last time we talked about standalone today we're looking at distributed runtimes. 00:45.120 --> 00:47.560 It's going to be even higher level than last time. 00:47.760 --> 00:49.960 And it's just to give you that flavor. 00:49.960 --> 00:55.480 And I should also point out that Microsoft says that that the distributed runtime is still experimental 00:55.480 --> 00:58.320 and the APIs are liable to change at any point. 00:58.320 --> 00:59.520 So it should be taken in that light. 00:59.520 --> 01:05.280 It's not actually ready for a production system yet, but it's more of an architecture and idea and 01:05.280 --> 01:12.120 an exciting, uh, idea in terms of future possibilities and the distributed runtime itself that we're 01:12.120 --> 01:19.840 going to look at right now, uh, is really it's described as something which handles processes, handles 01:19.840 --> 01:22.960 the messaging across process boundaries. 01:23.120 --> 01:24.240 That's the idea. 01:24.240 --> 01:27.680 It's no longer a single threaded thing running on our machine. 01:27.680 --> 01:32.840 It's something that can run across different processes that might not be Python processes. 01:32.840 --> 01:33.880 They could be anything. 01:34.160 --> 01:39.160 And it consists of two different things, two different components to it. 01:39.360 --> 01:46.080 One of them is called the host service, and that is the sort of container that runs this. 01:46.200 --> 01:51.880 And it connects to a worker runtime or potentially many worker runtimes. 01:51.880 --> 01:58.080 And it handles the delivery and it handles sessions for direct messages, direct messages that are going 01:58.120 --> 02:01.590 to be handled by gRPC if you're familiar with that technology. 02:01.790 --> 02:05.750 Remote procedure calls and it's going to handle the session management around that. 02:05.750 --> 02:11.350 All of the nuts and bolts of the infrastructure around the complicated business of sending a message 02:11.350 --> 02:16.990 remotely from one computer to another, from one process to another, that'll be taken care of by the 02:16.990 --> 02:17.790 framework. 02:18.190 --> 02:23.990 And then the other concept here is a worker runtime, that is the runtime that we'll be able to treat 02:23.990 --> 02:27.270 much as we treated the runtime in the single threaded case. 02:27.390 --> 02:29.870 And it will be able to manage different agents. 02:29.870 --> 02:33.030 It will have different agents that are that are registered with it. 02:33.070 --> 02:36.870 It will advertise the agents it's got to its host servers. 02:36.870 --> 02:43.350 So the host service knows what it's got and it will actually handle, of course executing the code the 02:43.390 --> 02:48.910 worker runtime has the agents, which themselves are delegates for something that that does something, 02:49.550 --> 02:56.630 and that will be handled by the worker runtime so that that's how it fits together. 02:56.630 --> 02:59.950 But it's going to be more concrete when I show it to you, which I'll do right now.