WEBVTT 00:01.000 --> 00:05.040 So let me now go and accept all of the changes. 00:07.080 --> 00:10.640 And let me run everything in debug here. 00:10.800 --> 00:16.880 So I want to debug this to show you exactly what is going to happen with the intermediate results. 00:17.320 --> 00:20.440 So we can see first things first, the call to the LM. 00:20.480 --> 00:23.600 The first call is what is the length of the word dog. 00:23.600 --> 00:25.720 And I put a breakpoint here. 00:25.800 --> 00:30.960 So once the AI will return a message we can go and check it. 00:33.680 --> 00:34.120 Cool. 00:34.120 --> 00:35.480 So we got a response. 00:35.480 --> 00:41.000 Let me show you all the local variables and the variable AI message. 00:41.200 --> 00:42.960 Let's take a look at all the fields. 00:43.120 --> 00:47.360 And we're looking for the attribute in the field of tool calls. 00:47.800 --> 00:50.840 We can see it right over here in the middle. 00:51.720 --> 00:52.920 Let me go and open it. 00:52.920 --> 00:56.680 And we can see this is a list containing one element. 00:57.000 --> 01:01.600 And in that element we have all of the information about the tool call. 01:01.640 --> 01:05.720 So the function we want to invoke is get text length. 01:06.040 --> 01:12.730 The arguments is going to be dictionary containing the keyword text and the input dog. 01:12.890 --> 01:15.210 We can also see the tool called id here. 01:15.530 --> 01:15.890 Cool. 01:15.890 --> 01:16.650 So let's continue. 01:16.690 --> 01:18.410 Now the execution step by step. 01:18.690 --> 01:22.210 So the number of tool calls is greater than one. 01:22.210 --> 01:25.650 So we are going to execute this if statement. 01:25.810 --> 01:31.610 And first thing is we want to append to the messages list the AI message with all of the information 01:31.610 --> 01:32.650 about the tool call. 01:32.690 --> 01:35.290 So this is going to be our agent trajectory. 01:35.290 --> 01:39.010 And now we want to iterate over the tool calls which we have only one. 01:39.210 --> 01:41.490 And we want to extract from that tool. 01:41.490 --> 01:44.610 Call the name the arguments and the ID. 01:44.890 --> 01:49.210 So let me show you again a the structure of the tool call. 01:49.410 --> 01:50.850 So we're going to do that. 01:51.050 --> 01:56.490 And once we have that we'll find all the variables we need in order to invoke the tool. 01:56.850 --> 02:01.690 So now we want to get the actual Python function to invoke. 02:01.690 --> 02:05.050 So we simply use the helper function we wrote earlier. 02:05.090 --> 02:06.290 Find tool by name. 02:06.610 --> 02:13.810 And the tool to use variable is going to contain the structured invokable in class. 02:14.010 --> 02:16.650 So we can simply use the invoke method on it. 02:17.900 --> 02:24.460 And then we simply go and invoke that with the input and we get back an observable. 02:24.700 --> 02:29.500 So let me go and put a breakpoint in in the tool that we wrote. 02:29.500 --> 02:33.460 So let me go up here and this is here. 02:33.500 --> 02:38.860 Let me put a breakpoint here and let me continue the run and let me continue the run. 02:42.380 --> 02:42.900 All right. 02:42.900 --> 02:45.220 So we can see we got here. 02:45.220 --> 02:46.940 Let's go and run everything. 02:46.940 --> 02:52.140 And the result we get is the length of this text here. 02:52.420 --> 02:54.700 So let's go and continue. 02:54.860 --> 02:56.380 And the result should be three. 02:57.620 --> 02:59.980 And we can see here in the observation. 02:59.980 --> 03:02.220 Let's go and check out this observation here. 03:02.220 --> 03:03.500 This observation is three. 03:03.740 --> 03:04.260 Cool. 03:04.260 --> 03:08.980 So now we want to append the observation the result of the tool execution. 03:09.020 --> 03:11.180 We want to append it to the message list. 03:11.580 --> 03:17.580 And now we want to go and to continue the iteration on the tool calls which we don't have anymore. 03:17.580 --> 03:20.860 So we're going to finish the iteration in this for loop over here. 03:20.860 --> 03:25.500 And we're going to hit this continue and continue statement. 03:25.740 --> 03:28.470 So let's go and click continue. 03:28.750 --> 03:29.310 Right. 03:29.310 --> 03:30.590 So now we've finished. 03:30.630 --> 03:32.950 Now the iteration over the tool calls. 03:32.990 --> 03:39.110 Now we're left to do another iteration in the while loop with the message history containing the agent 03:39.110 --> 03:41.710 trajectory and all the intermediate results. 03:42.270 --> 03:47.790 So let me show you the messages here so we can see we have a list of three elements. 03:47.790 --> 03:49.790 The first one is the human message. 03:49.990 --> 03:55.150 The second one is the AI message with all the tool calling information. 03:55.150 --> 04:00.110 And the third message is a tool message containing the result of the tool execution. 04:00.470 --> 04:00.750 All right. 04:00.750 --> 04:02.550 So let's invoke the LM call. 04:03.030 --> 04:04.830 Now we're waiting for a response. 04:05.070 --> 04:09.350 And now the response should have zero two calls when we get it back. 04:09.630 --> 04:11.830 So let's take a look at the AI message. 04:11.830 --> 04:12.870 We get back here. 04:13.190 --> 04:15.950 And now let's take a look at the tool calls field. 04:15.950 --> 04:18.430 And we can see it's actually an empty list. 04:18.630 --> 04:24.110 And that's why we are going to print the answer and break and finish the run here. 04:24.790 --> 04:30.790 And we can see that the final answer of the LM is the length of the word dog is three characters. 04:30.950 --> 04:35.390 So we have our react agent working with function calling.