WEBVTT 00:00.630 --> 00:01.920 Eden: Hey, there, Eden here. 00:01.920 --> 00:04.320 And in this video, we're going to cover, 00:04.320 --> 00:06.630 from the theoretical point of view, 00:06.630 --> 00:09.210 the core components of LangGraph. 00:09.210 --> 00:10.800 And this is super important 00:10.800 --> 00:12.720 because right after this video, we're going 00:12.720 --> 00:16.623 to start implementing project using those components. 00:18.870 --> 00:20.700 Alright, so with LangGraph, we're going 00:20.700 --> 00:23.700 to be implementing a well-defined control flow 00:23.700 --> 00:25.740 or a state machine or a graph. 00:25.740 --> 00:27.570 You can call it whatever you want. 00:27.570 --> 00:30.960 But the basic idea here is that within this well-defined 00:30.960 --> 00:33.240 and well scoped control flow, we're going 00:33.240 --> 00:34.890 to be leveraging LLMs 00:34.890 --> 00:37.440 and they're going to have a very crucial role. 00:37.440 --> 00:41.010 So LLMs are either going to define in the control flow 00:41.010 --> 00:42.510 where we're going to go next. 00:42.510 --> 00:45.030 So this is going to be non-deterministic, 00:45.030 --> 00:47.400 or we can even, in this control flow, 00:47.400 --> 00:50.310 execute LLM calls or agent calls. 00:50.310 --> 00:53.940 And to get started, we need three LangGraph core components: 00:53.940 --> 00:57.000 nodes, edges, and conditional edges. 00:57.000 --> 00:59.580 Nodes are literally python functions. 00:59.580 --> 01:02.190 So you can put here any code you want. 01:02.190 --> 01:05.550 This code can be deterministic code, regular Python code, 01:05.550 --> 01:08.190 it can be Python code that calls an LLM, 01:08.190 --> 01:10.860 or it can even be an LLM agent. 01:10.860 --> 01:14.130 But you have full flexibility of what you want to do 01:14.130 --> 01:15.750 inside the nodes. 01:15.750 --> 01:19.290 Edges connect those nodes within the graphs execution 01:19.290 --> 01:22.230 and conditional edges are helped to make decisions 01:22.230 --> 01:24.960 whether to go to node A or to node B, 01:24.960 --> 01:28.230 and this is dynamic and can be very, very flexible. 01:28.230 --> 01:30.930 And this is the power of LangGraph. 01:30.930 --> 01:33.780 So we have full control of how we're going 01:33.780 --> 01:36.513 to navigate in this graph execution. 01:38.640 --> 01:42.330 Cool. Two important built-in LangGraph nodes 01:42.330 --> 01:45.180 are the start node and the end node. 01:45.180 --> 01:47.580 And the start node is the entry point 01:47.580 --> 01:49.470 for our graph execution. 01:49.470 --> 01:51.720 So we're going to start from there 01:51.720 --> 01:54.300 and the end node is going to be the last node 01:54.300 --> 01:56.370 that is going to be executed. 01:56.370 --> 01:59.190 And both of those nodes don't really do anything. 01:59.190 --> 02:02.373 So you can think about them as no operation. 02:06.420 --> 02:09.240 One of the most important concepts in LangGraph 02:09.240 --> 02:11.850 is the state or the agent state. 02:11.850 --> 02:14.820 And the state is simply a dictionary 02:14.820 --> 02:16.890 that is going to have information 02:16.890 --> 02:19.740 that is important for us to keep tracking the graph. 02:19.740 --> 02:22.230 Maybe some nodes execution results, 02:22.230 --> 02:26.430 maybe some temporary results or even our chat history. 02:26.430 --> 02:29.910 It can be fairly simple like only storing the chat history, 02:29.910 --> 02:33.300 so the results of the LLMs every time, 02:33.300 --> 02:35.100 or it could be very complex 02:35.100 --> 02:37.953 and we can customize it to anything we want. 02:39.240 --> 02:42.450 It is local to the graph that is it is available 02:42.450 --> 02:46.140 for every node to access within our graph's execution. 02:46.140 --> 02:51.120 And it is also available on each edge and it can be local. 02:51.120 --> 02:54.450 So within our graph execution in our runtime. 02:54.450 --> 02:58.530 However, we can also persist it into persistent storage. 02:58.530 --> 03:02.310 So if you want to stop for some reason, the flow execution 03:02.310 --> 03:05.280 of the graph and we want to continue from that same point 03:05.280 --> 03:08.400 with that given state, we can do it with persistence 03:08.400 --> 03:10.830 and we're going to show you that in the course. 03:10.830 --> 03:13.170 Alright, so remember that I told you 03:13.170 --> 03:16.500 that every node in LangGraph is a function 03:16.500 --> 03:18.870 and we can write there whatever we want. 03:18.870 --> 03:20.160 This is still the case, 03:20.160 --> 03:22.590 but nodes are special functions 03:22.590 --> 03:24.840 that always receive as an input, 03:24.840 --> 03:27.180 the current graph state, which is going 03:27.180 --> 03:29.790 to have all the information that the node needs 03:29.790 --> 03:32.310 in order to do its work. 03:32.310 --> 03:35.040 And what this function returns the node function, 03:35.040 --> 03:39.360 it's updating the state, so it always returns a dictionary 03:39.360 --> 03:43.233 with the keys of what do we want to update in our state. 03:44.820 --> 03:48.420 And the result of this is that in LangGraph, 03:48.420 --> 03:50.460 every node is going to update the state. 03:50.460 --> 03:53.730 So the state object is going to change over time 03:53.730 --> 03:56.430 and the conditional edges and the edges 03:56.430 --> 03:58.050 are going to rely on the states 03:58.050 --> 04:01.260 whether to go to node A or to node B. 04:01.260 --> 04:04.050 And this is how our software is going to look like. 04:04.050 --> 04:06.270 So with these core components, 04:06.270 --> 04:08.793 we can build very advanced things. 04:10.710 --> 04:14.460 Alright, so more concepts that are important for the rest 04:14.460 --> 04:17.820 of the course is, first, a cyclic graph. 04:17.820 --> 04:21.030 So with LangGraph, we can also implement loops 04:21.030 --> 04:25.080 and this is very powerful, which is something that 04:25.080 --> 04:27.210 with LangChain was super hard to do, 04:27.210 --> 04:30.720 and we're going to later discuss this in the course. 04:30.720 --> 04:32.220 Also human in the loop. 04:32.220 --> 04:35.220 So if we want to get human feedback 04:35.220 --> 04:38.160 and this will decide where do we need to go 04:38.160 --> 04:42.180 in the graph's execution, so to node A or to node B, 04:42.180 --> 04:43.650 then this is also something 04:43.650 --> 04:47.430 that LangGraph also helps us to easily implement. 04:47.430 --> 04:50.940 And the last thing is persistence, which I mentioned it. 04:50.940 --> 04:54.600 And LangGraph comes with very nice 04:54.600 --> 04:56.850 and simple built-in functions 04:56.850 --> 05:00.120 that will help us persist our state of the graph. 05:00.120 --> 05:04.710 And this will help us also make our software more robust 05:04.710 --> 05:06.480 and more fault-tolerant, 05:06.480 --> 05:09.150 but also implement some very cool logic 05:09.150 --> 05:12.420 that will show you in the course that is going to result us 05:12.420 --> 05:15.333 with very nice user experience.