WEBVTT 00:00.090 --> 00:01.650 -: All right, let's start having a look 00:01.650 --> 00:03.510 at some of the different types of functionality 00:03.510 --> 00:04.800 that we can use in LangGraph. 00:04.800 --> 00:05.910 Just before we get into that, 00:05.910 --> 00:07.530 I just want to make a highlight and say, 00:07.530 --> 00:11.040 if you haven't already, download the GitHub repository link 00:11.040 --> 00:12.360 inside of this course. 00:12.360 --> 00:13.920 We provide lots of different notebooks 00:13.920 --> 00:15.720 for working specifically with LangGraph, 00:15.720 --> 00:18.360 and we've also got some more advanced guides there. 00:18.360 --> 00:20.220 Okay, so the first one we're gonna do 00:20.220 --> 00:21.780 is install a couple of different packages. 00:21.780 --> 00:23.566 So we'll be installing LangGraph, LangSmith, 00:23.566 --> 00:25.080 and LangChain OpenAI. 00:25.080 --> 00:27.090 And we've got some functionality here, 00:27.090 --> 00:31.410 which just checks to see, do you have a OpenAI API Key. 00:31.410 --> 00:33.180 So I'm obviously gonna run all these. 00:33.180 --> 00:36.360 And the important point here is you can use 00:36.360 --> 00:39.540 one of LangChain's products called LangSmith, 00:39.540 --> 00:42.210 which will give you some of the observability 00:42.210 --> 00:43.470 to see what's happening inside 00:43.470 --> 00:45.360 of these LangGraph state machines. 00:45.360 --> 00:47.700 I'm probably not gonna do it in this scenario, 00:47.700 --> 00:49.860 but I'll leave it by size and point. 00:49.860 --> 00:51.750 Now we're gonna build this chatbot over time. 00:51.750 --> 00:54.510 So we'll start by just building a very simple chatbot. 00:54.510 --> 00:56.820 And then over the next couple of coming lessons, 00:56.820 --> 00:59.250 we'll start adding more and more functionality 00:59.250 --> 01:01.200 into this state machine. 01:01.200 --> 01:02.340 The first thing we're gonna do 01:02.340 --> 01:05.190 is we're gonna just have a look at some very simple things 01:05.190 --> 01:06.090 inside of LangGraph. 01:06.090 --> 01:09.390 So we've got this idea of a StateGraph, 01:09.390 --> 01:12.420 and we are using something called a TypedDict. 01:12.420 --> 01:14.070 Now TypedDict is quite interesting 01:14.070 --> 01:17.910 because it allows us to have some type, typing functionality 01:17.910 --> 01:20.940 over, specifically over a variable. 01:20.940 --> 01:22.920 And what we have in this scenario is 01:22.920 --> 01:27.210 we have a list of messages, which is an annotated list, 01:27.210 --> 01:30.210 and we have this add messages annotation. 01:30.210 --> 01:34.170 And in this annotation it is actually how 01:34.170 --> 01:36.000 the update happens to that. 01:36.000 --> 01:39.840 And so you can see LangGraph provides a add messages one, 01:39.840 --> 01:42.750 which will basically merge two lists of messages, 01:42.750 --> 01:46.860 and it updates the existing messages by a specific ID. 01:46.860 --> 01:48.600 So what's happening is, 01:48.600 --> 01:50.580 imagine you've got these two messages, 01:50.580 --> 01:53.670 like messages and messages. 01:53.670 --> 01:55.380 The add messages will make sure 01:55.380 --> 01:57.753 that these two lists are merged, right? 01:58.980 --> 02:01.320 So we've got our StateGraph, 02:01.320 --> 02:03.570 and you can have other keys in here, 02:03.570 --> 02:04.620 which we'll see later on. 02:04.620 --> 02:06.840 So you could say like, is human, right? 02:06.840 --> 02:08.130 Which is the type of string. 02:08.130 --> 02:10.170 Or we could say like, you know, 02:10.170 --> 02:13.440 has completed workflow, right? 02:13.440 --> 02:14.670 We can have a Boolean. 02:14.670 --> 02:17.640 So I want you to think about these as types 02:17.640 --> 02:19.890 which will be used inside of your state machine 02:19.890 --> 02:22.020 to hold data, okay? 02:22.020 --> 02:25.533 For now, we're just gonna use the messages key. 02:26.940 --> 02:29.340 We set up our graph with the graph builder. 02:29.340 --> 02:30.960 Now if we actually have a look at what's happening 02:30.960 --> 02:33.810 inside of this graph, it has a StateGraph, 02:33.810 --> 02:35.370 and it has a lot of different functionalities. 02:35.370 --> 02:39.060 So you can have these nodes, we can have edges, 02:39.060 --> 02:41.040 we can have branches, 02:41.040 --> 02:44.730 we can also do things like set the entry point, 02:44.730 --> 02:46.560 set the finish point. 02:46.560 --> 02:48.390 We have various different things, 02:48.390 --> 02:51.810 and we've also got some for persistence as well. 02:51.810 --> 02:54.300 Now, the most important ones that you'll use quite a lot 02:54.300 --> 02:57.870 is the add node, the add edge, the add conditional edges. 02:57.870 --> 02:59.910 So we have our piece of state, 02:59.910 --> 03:01.260 which we can see as a TypedDict 03:01.260 --> 03:03.030 with that single messages key. 03:03.030 --> 03:06.360 And it's annotated with the ad messages function. 03:06.360 --> 03:07.950 And then now what we need to do 03:07.950 --> 03:09.990 is we need to have every node, 03:09.990 --> 03:13.470 which will basically receive the current state as the input 03:13.470 --> 03:16.140 and return a value that updates that state. 03:16.140 --> 03:17.130 The second thing we're gonna do 03:17.130 --> 03:19.620 is the messages will be appended to the current list 03:19.620 --> 03:21.420 rather than directly overwritten. 03:21.420 --> 03:23.310 And that's kind of how we're using that 03:23.310 --> 03:25.320 with this add messages function. 03:25.320 --> 03:27.900 So we're gonna load up the chat OpenAI model. 03:27.900 --> 03:29.640 We're gonna use GPT-4o. 03:29.640 --> 03:31.380 And then we have a node here, 03:31.380 --> 03:32.880 which we're gonna make into a node 03:32.880 --> 03:35.850 where we take in the state, right? 03:35.850 --> 03:38.100 And on the state what we're gonna say is 03:38.100 --> 03:40.950 return messages, right? 03:40.950 --> 03:43.920 And also then we've got the llm.invoke, 03:43.920 --> 03:45.570 which takes in the messages, 03:45.570 --> 03:48.360 and it invokes the chat model with those messages. 03:48.360 --> 03:50.490 And notice how we're wrapping this in the list 03:50.490 --> 03:53.190 so that it actually gets added on back. 03:53.190 --> 03:56.100 Now, once you've got your Python function set up 03:56.100 --> 03:58.350 to take in the LangGraph state, 03:58.350 --> 04:01.590 and to manipulate that state, you can then add the node. 04:01.590 --> 04:02.820 And so that's what you've got here 04:02.820 --> 04:04.140 where we've got that graph builder, 04:04.140 --> 04:06.030 and we're adding the node of chatbot 04:06.030 --> 04:07.413 to this Python function. 04:09.390 --> 04:11.280 Now we also have to decide 04:11.280 --> 04:13.590 where do we start inside of the graph? 04:13.590 --> 04:15.360 Where do we start in our state machine? 04:15.360 --> 04:17.610 So as we've only got one node in the graph, 04:17.610 --> 04:19.953 we're gonna set the entry point to chatbot. 04:20.790 --> 04:21.960 And also as well as that, 04:21.960 --> 04:24.360 we can set the finish point, right? 04:24.360 --> 04:26.610 So the finish point can be chatbot. 04:26.610 --> 04:28.860 Now we can compile the graph, 04:28.860 --> 04:31.200 and this will create a compile graph type. 04:31.200 --> 04:32.880 So you'll see if we actually look at this, 04:32.880 --> 04:34.800 it's a compiled StateGraph. 04:34.800 --> 04:36.060 And we can look at things like, 04:36.060 --> 04:37.710 we can look at the nodes that there are. 04:37.710 --> 04:39.660 You can see this is the starting node, 04:39.660 --> 04:41.460 and we've got other ones as well. 04:41.460 --> 04:46.460 So what we can now do is we can use the get graph function 04:46.740 --> 04:49.170 on the compiled graph to actually see 04:49.170 --> 04:51.420 what that state machine looks like. 04:51.420 --> 04:52.950 So you can see we've got the start, 04:52.950 --> 04:55.440 which goes into chatbot, which then ends. 04:55.440 --> 04:56.400 So let's try and run it. 04:56.400 --> 04:58.590 So we've got a while loop here. 04:58.590 --> 05:02.070 And while this is running, we are getting the user's input. 05:02.070 --> 05:06.150 And if the input is not quit, exit, or q, 05:06.150 --> 05:07.980 then we're gonna run this bit of code. 05:07.980 --> 05:11.130 But if it is, then we're gonna run this blocker code here. 05:11.130 --> 05:12.600 So that'll be printing goodbye 05:12.600 --> 05:14.280 and breaking out the while loop. 05:14.280 --> 05:15.113 So let's have a look. 05:15.113 --> 05:16.920 So I can say, hello. 05:16.920 --> 05:19.140 My name is James. 05:19.140 --> 05:20.160 I'm just typing this. 05:20.160 --> 05:22.774 And then it says, how can I assist you today? 05:22.774 --> 05:24.783 I'll say, what is going on. 05:25.920 --> 05:26.753 Right? 05:27.990 --> 05:29.460 And then it's giving me a response 05:29.460 --> 05:32.103 and I can say, let's say quit. 05:32.970 --> 05:35.850 Now what's going on here is we've just quit out of the loop, 05:35.850 --> 05:37.740 but you can see how we call this. 05:37.740 --> 05:40.080 So we do a graph.stream, 05:40.080 --> 05:43.470 and the input is the first argument, okay? 05:43.470 --> 05:48.210 And the input here is our Python dictionary 05:48.210 --> 05:49.200 with the state. 05:49.200 --> 05:52.170 So we're almost saying to the state, 05:52.170 --> 05:53.850 if we scroll back up here 05:53.850 --> 05:56.730 and we have a look, we have messages here. 05:56.730 --> 06:00.120 Well, we can actually set the inputs for that 06:00.120 --> 06:04.020 specifically when we do the .stream functionality. 06:04.020 --> 06:06.120 The final thing I just want to make you aware of 06:06.120 --> 06:09.330 is that you can also call a graph using other metrics 06:09.330 --> 06:14.310 and functions such as the .invoke function. 06:14.310 --> 06:16.650 So if we don't want to stream the nodes 06:16.650 --> 06:19.560 and the events that are happening inside of the graph, 06:19.560 --> 06:21.450 we could just call the .invoke. 06:21.450 --> 06:23.730 And so you can see here we've got the messages, 06:23.730 --> 06:25.380 and we've said, hello world, 06:25.380 --> 06:29.430 and the AI has responded with, How can I assist you today? 06:29.430 --> 06:30.900 In the next lesson, what we're gonna look at 06:30.900 --> 06:32.880 is how you can add memory 06:32.880 --> 06:35.970 to your specific LangGraph state machine, 06:35.970 --> 06:40.350 and also how you can save that state machine to disc 06:40.350 --> 06:43.020 and then load that state machine as well. 06:43.020 --> 06:44.913 So we'll be looking at persistence.