WEBVTT 00:00.400 --> 00:03.840 Okay, so now we get to the meat of Autogen core. 00:04.240 --> 00:11.360 We are going to create a runtime, a single threaded agent runtime it's called, which is a standalone 00:11.360 --> 00:17.640 runtime running on my computer, which as it says, will handle agents in a single threaded way. 00:18.160 --> 00:22.640 And the first thing that we're going to do is register. 00:22.640 --> 00:28.920 We call register on the agent itself to say, I want you to register yourself with this runtime. 00:29.040 --> 00:31.560 Now, this isn't creating an agent. 00:31.600 --> 00:34.440 This is just saying you are a type of agent. 00:34.440 --> 00:40.680 And I want you to tell the runtime that you are a type of agent that can be spawned, you can be created, 00:40.920 --> 00:45.560 and you are a type of agent of type simple agent that's going to be your type. 00:45.920 --> 00:52.920 And this, this thing here, this is a function which can generate new versions of you. 00:52.920 --> 00:54.240 It can instantiate you. 00:54.280 --> 00:55.360 It is a factory. 00:55.640 --> 00:57.600 And you pass that in as well. 00:57.840 --> 00:58.960 And so I will do this. 00:59.000 --> 01:03.660 It's now created a runtime and it's created a type of agent called a simple agent. 01:03.700 --> 01:05.980 We haven't actually yet built any of these. 01:05.980 --> 01:07.300 We haven't instantiated one. 01:07.300 --> 01:12.660 But the as a type of agent is now a known thing to our runtime. 01:12.700 --> 01:15.500 Our runtime knows that there are such things as simple agents. 01:15.740 --> 01:18.380 And we're now going to start our runtime. 01:18.380 --> 01:20.220 It is now running. 01:20.220 --> 01:21.980 It is a it is a proper runtime. 01:22.460 --> 01:31.460 And now we are going to uh, we're going to first of all create an agent ID object to identify an agent. 01:32.100 --> 01:34.340 And we'll we'll do this properly. 01:34.380 --> 01:41.100 We'll say agent ID equals that, that is an ID that would identify this. 01:41.100 --> 01:45.980 And we are going to say that we want the default agent. 01:47.100 --> 01:52.860 Uh, because, because that, that will then, uh, we'll make sure that we have an agent created, 01:53.140 --> 01:59.860 and then we're going to send a message to, we're going to send a message to the, to the agent, which 01:59.860 --> 02:04.630 is called simple agent, The default ID and we're going to send a message. 02:04.630 --> 02:06.550 Well hi there to it. 02:06.750 --> 02:09.990 And then we will print whatever comes back. 02:10.470 --> 02:12.790 So any ideas what's going to come back I wonder. 02:14.590 --> 02:18.950 Uh, so uh it replied this is simple agent default. 02:18.950 --> 02:24.830 So it's simple agent is its type, which is the same one we registered and it default is the, uh, 02:24.990 --> 02:29.150 the, the id uh, and it said, you said, well hi there. 02:29.190 --> 02:30.310 And I disagree. 02:31.510 --> 02:32.190 Charming. 02:32.630 --> 02:34.990 Uh, and that is uh, that's it. 02:34.990 --> 02:39.870 So there is a demo of auto gen core and that's what auto gen core does. 02:39.910 --> 02:44.950 And putting in agent functionality like having code in there that calls LMS. 02:44.990 --> 02:46.510 That's your job or my job. 02:46.510 --> 02:48.950 That's not auto gen core's job. 02:49.070 --> 02:54.630 What it's there to do is to handle the passing of messages around by looking things up based on their 02:54.630 --> 02:57.230 type and their ID, and that's what it does. 02:57.510 --> 03:03.770 So now I'll show you one which has an LM behind it, but it shouldn't be any surprise to you because 03:03.770 --> 03:05.370 it's just the same infrastructure. 03:05.370 --> 03:13.370 The fact that we're calling an LLM is is almost of no consequence to, to, uh, the Autogen core framework. 03:13.530 --> 03:19.050 So what we're going to do is we're going to make a slightly more interesting one called my LM agent, 03:19.090 --> 03:21.210 still a subclass of rooted agent. 03:21.450 --> 03:24.050 And we're going to use a proper LM. 03:24.050 --> 03:30.210 And we could just call OpenAI directly using OpenAI dot create. 03:30.250 --> 03:32.490 We could just call the Python client library. 03:32.690 --> 03:39.490 But we can also use Autogen agent chat and we might as well since it's autogen week. 03:39.810 --> 03:46.090 So this is the same kind of thing, a subclass of rooted agent that's going to be two functions again 03:46.130 --> 03:49.330 init and handle my message. 03:49.570 --> 03:53.450 So the init just just calls the superclass. 03:53.770 --> 03:56.610 But it then creates a model client. 03:56.890 --> 03:58.050 But this could be doing anything. 03:58.050 --> 04:00.650 But this is the Autogen agent chat way. 04:00.690 --> 04:05.070 We're going to create one of these things, a GPT four mini model client. 04:05.070 --> 04:10.190 And we're going to set a field called underscore delegate, which we're just sort of holding on to. 04:10.230 --> 04:12.470 This is this is the the underlying. 04:12.510 --> 04:18.390 This is what our agent object is going to delegate to when it actually needs some code, some, some 04:18.430 --> 04:19.550 an LLM to run. 04:19.550 --> 04:21.110 And we could have anything we want. 04:21.110 --> 04:26.670 We could, we could do we could reply back with random things or we could create an instance of assistant 04:26.670 --> 04:27.030 agent. 04:27.030 --> 04:28.310 And that's what we're going to do. 04:28.550 --> 04:29.870 And that's where we put in. 04:29.910 --> 04:33.750 If you remember, when you create an assistant agent, there's the name of the agent and there's the 04:33.750 --> 04:35.350 model client, and there it is. 04:35.550 --> 04:43.190 And as a result in underscore delegate underscore often used to show sort of like private secret, uh, 04:43.510 --> 04:47.230 variables that other people should know about, like, like private in the, in the Java world. 04:47.470 --> 04:52.110 Um, so the this is being set to an assistant agent. 04:53.270 --> 04:57.790 So, uh, now we just have to write our message handler. 04:58.070 --> 05:00.350 And as I say, it can be called anything you like. 05:00.510 --> 05:02.640 I'm calling it handle my message type. 05:03.040 --> 05:08.680 Um, and that's really to sort of draw your attention to the fact that what matters is what type of 05:08.680 --> 05:16.800 object comes in the message field, because Autogen core will automatically route messages to agents 05:16.800 --> 05:24.680 that should receive them based on the the finding a handler that that looks for a message of that type. 05:24.840 --> 05:25.520 That's how it works. 05:25.520 --> 05:27.320 That's the clever routing that it does. 05:27.360 --> 05:31.200 And by the way, you can send a message direct to an agent, which is what we're doing right now. 05:31.200 --> 05:37.120 But it also has a whole kind of pub sub thing there where you can subscribe to topics and you can publish 05:37.120 --> 05:41.200 out messages to lots of agents that are all interested in one particular topic. 05:41.360 --> 05:48.160 So the whole process of dispatching messages to the right agent and then calling the right handler, 05:48.320 --> 05:49.560 that's the clever stuff. 05:49.560 --> 05:50.960 That's what it does well. 05:51.000 --> 05:55.120 And why why we use a framework like this rather than coding it ourselves. 05:55.360 --> 05:57.520 But for some of these examples, it's pretty trivial. 05:57.520 --> 05:59.320 So we could equally code it ourselves. 05:59.640 --> 06:03.820 But you can imagine that this is done in a way that can be scaled and distributed. 06:04.100 --> 06:06.740 So anyways, handle my message type. 06:07.100 --> 06:14.100 Uh, as long as, uh, message is sent to this agent with the ID and and type, and that the object 06:14.100 --> 06:17.580 that's sent is of this type, it will arrive at this function. 06:17.780 --> 06:21.820 And I'm going to say that I received a message and print the content. 06:22.340 --> 06:24.700 Then I'm going to now now this is confusing. 06:25.300 --> 06:31.020 This object here text message looks subtly different to this object here. 06:31.020 --> 06:31.780 Message. 06:32.180 --> 06:34.300 Uh, do you know what's going on here? 06:34.500 --> 06:39.180 This text message is text message from the, um, autogen. 06:39.500 --> 06:40.460 It's this thing here. 06:40.460 --> 06:42.060 It's the agent chat messages. 06:42.060 --> 06:43.020 The text message. 06:43.020 --> 06:50.020 The same thing we were using yesterday and the day before when we were interacting with agent chat agents. 06:50.020 --> 06:55.260 So this is a autogen agent chat, uh, message, which can be a bit confusing. 06:55.260 --> 06:56.020 And you might. 06:56.220 --> 06:59.900 Yeah, there are ways you could you could get around that by, by giving it a different you can say 06:59.900 --> 07:08.960 like as agent chat text message if you wanted to to like make it so the code is a little bit clearer 07:08.960 --> 07:13.560 and you're distinguishing between the Autogen core and the Autogen agent chat. 07:13.880 --> 07:14.520 All right. 07:14.960 --> 07:16.080 Anyways, so sorry. 07:16.080 --> 07:16.760 Back we go. 07:16.920 --> 07:24.920 So we create the sort of text message that you need for Autogen agent chat passing in the content of 07:24.920 --> 07:27.000 our message and the source is user. 07:27.240 --> 07:35.080 And this is the onmessage that we call in agent Chat land with this text message and the cancellation 07:35.080 --> 07:36.520 token if you remember that thing. 07:37.160 --> 07:39.840 And then we get back the content. 07:40.120 --> 07:41.880 We say that that was the reply. 07:41.880 --> 07:49.160 And we return a new instance of message of our message, with the content being that reply. 07:50.120 --> 07:53.280 I hope you followed all that, or at least got enough of an idea of it. 07:53.320 --> 07:54.760 Let's give this a shot. 07:54.920 --> 08:00.320 Let's create a new runtime, a single threaded agent runtime, standalone runtime. 08:00.320 --> 08:06.490 We're going to register two agents the simple agent and the LM agent. 08:06.530 --> 08:10.610 The new one, Simple Agent, isn't really an LM at all. 08:10.610 --> 08:12.770 It just disagrees with whatever it's told. 08:12.770 --> 08:21.250 And LM agent is actually going to dispatch through this underscore delegate down to a true gpt4 mini. 08:21.730 --> 08:23.850 So we're going to start the runtime. 08:24.330 --> 08:27.130 Then we are going to send a message. 08:27.130 --> 08:30.410 Hi there to the real agent. 08:30.610 --> 08:33.610 And with whatever it replies we're going to send that. 08:33.610 --> 08:39.890 We're going to forward that on to the other agent, to the simple agent and see what comes out there. 08:42.210 --> 08:44.850 So the LM agent received hi there. 08:44.850 --> 08:46.930 And it said hello, how can I assist you today? 08:46.970 --> 08:49.170 A classic GPT four minion response. 08:49.570 --> 08:53.770 And, uh, the simple agent said, this is simple agent default. 08:53.770 --> 08:55.690 You said, hello, how can I assist you today? 08:55.690 --> 08:59.610 And I disagree, and there you go. 09:00.050 --> 09:04.330 Um, and with that, we will we will stop this pantomime.