WEBVTT 00:00.200 --> 00:06.120 Okay, so I checked out this branch, which is called project slash react search agent. 00:06.320 --> 00:12.000 And just a quick note, if you want to check out the code for this section, check out this branch over 00:12.000 --> 00:12.520 here. 00:12.520 --> 00:18.360 And for technical reasons, I committed the code to this branch over here and not the branch on the 00:18.360 --> 00:19.320 bottom left here. 00:19.960 --> 00:23.480 Alright, let's start by initializing our Python project. 00:23.480 --> 00:24.840 Let's write init. 00:25.560 --> 00:28.520 And now let's go install the dependencies. 00:28.520 --> 00:30.200 Let's add link chain. 00:30.520 --> 00:32.480 We want to use link chain OpenAI. 00:33.080 --> 00:36.520 And we discussed those packages in previous videos. 00:37.040 --> 00:41.640 So now we want to add a new package which is called link Chain. 00:42.480 --> 00:49.720 And Tivoli is a third party which is going to help us connect our agent to the web and give the agent 00:49.760 --> 00:54.360 searching capabilities so it will act as a search engine for our agent. 00:54.360 --> 01:01.480 And it also has other very useful APIs like Tivoli Crawl, Tivoli Map, Tivoli Extract, which are very 01:01.480 --> 01:08.610 useful services for AI agents, and we'll be using those capabilities in direct section of the course, 01:08.610 --> 01:10.050 so we'll cover them as well. 01:10.250 --> 01:17.290 And by the way, David is actually the most popular choice for integrating a web search into an agent. 01:17.570 --> 01:24.490 And they're even featured in the official documentation as the default service for a search engine. 01:24.610 --> 01:26.490 And when implementing agents. 01:26.690 --> 01:29.010 So they really go to example. 01:29.290 --> 01:33.690 And I think the reason for that is that they first of all have an excellent API. 01:33.730 --> 01:34.850 It scales very well. 01:34.890 --> 01:36.090 It's very easy to use. 01:36.330 --> 01:40.450 And I think they were actually the first ones to go and integrate this. 01:40.450 --> 01:44.370 So to connect an agent A with searching capabilities. 01:44.410 --> 01:49.130 Now it's also important to note that they have a very generous free tier. 01:49.330 --> 01:55.450 And you can see it's 1000 API requests per month, which is going to be more than enough for this course. 01:56.250 --> 01:56.570 All right. 01:56.570 --> 02:00.450 Let me just show you a quick demo of the capabilities that we can integrate with it. 02:01.250 --> 02:08.740 So in the API playground let me write a query like what are the latest anthropic models? 02:09.020 --> 02:12.540 And we can see in the response we get here the URL of the source. 02:12.740 --> 02:18.340 We also get the content you can see here it's talking about Cloud Outpost 4.1. 02:18.740 --> 02:23.460 And in this search result it talks about Cloud Ops and Sonet four. 02:24.620 --> 02:26.780 And the search can be much more advanced. 02:26.820 --> 02:30.620 We can have some special filtering and advanced capabilities. 02:30.620 --> 02:34.060 However in this section we're only going to stick with the basics. 02:34.220 --> 02:34.580 All right. 02:34.580 --> 02:36.140 So that's enough about me. 02:36.260 --> 02:38.900 And I'm going to link to Vili in the videos. 02:38.900 --> 02:39.620 Resources. 02:40.620 --> 02:41.220 Alrighty. 02:41.220 --> 02:47.900 Let's go back to the code here and link to Vili is the link chain integration. 02:48.180 --> 02:54.540 And here the provider is the Vili a in contrast to OpenAI in the previous example. 02:54.780 --> 03:00.980 So a provider of link chain doesn't have to be a vendor which offers large language models. 03:00.980 --> 03:06.980 It can also be a vendor which offer services that are exposed via API's and clients. 03:06.980 --> 03:12.840 So in this case Turville is going to expose their services, their searching capabilities, their scraping 03:12.840 --> 03:19.600 capabilities, and the LinkedIn integration here is going to make our lives very, very easy when we 03:19.640 --> 03:26.600 want to integrate those capabilities into our link chain graph applications here, because the team 03:26.640 --> 03:31.200 they wrote now, this integration, they connected to all the link chain objects, which we're going 03:31.200 --> 03:31.560 to use. 03:31.560 --> 03:32.560 And I'll give you a quick hint. 03:32.560 --> 03:34.280 It's going to be link chain tools. 03:34.280 --> 03:37.480 And this is another example of a link chain provider. 03:37.840 --> 03:38.320 All right. 03:38.320 --> 03:44.840 Let's go and also install Python dot env to load environment variables like we did before. 03:45.160 --> 03:46.760 And let me also install black. 03:46.760 --> 03:51.760 And I sought for some formatting and uh yeah let's go and install everything. 03:51.760 --> 03:53.280 So let me go and click enter. 03:55.120 --> 03:59.240 We can check out the packages in the pyproject.toml file over here. 03:59.240 --> 04:02.920 And if you want the exact packages we can check out the dot log. 04:03.280 --> 04:07.680 And let's go to our main.py file the boilerplate code. 04:07.680 --> 04:08.680 So let's try it. 04:08.680 --> 04:11.450 And let's run this code and make sure that it's working. 04:12.090 --> 04:12.610 All right. 04:12.610 --> 04:13.970 So we can see it's working. 04:14.290 --> 04:17.530 And now let me show you what I have in my env file. 04:17.810 --> 04:24.410 So like in the previous section I have my open API key I have Lang Smith. 04:24.450 --> 04:25.890 Tracing equals true. 04:25.890 --> 04:30.050 So we want to trace everything and we want to check out the traces I have here. 04:30.050 --> 04:31.530 My Lang Smith API key. 04:31.570 --> 04:37.530 Like previous sections I have my Lang Smith project which I changed in called Search Agent. 04:37.730 --> 04:44.450 And I also have here a API key which I got from the UI. 04:44.770 --> 04:45.210 All right. 04:45.210 --> 04:47.010 Let me show you how you can get it. 04:47.010 --> 04:49.890 You simply need to go to the village to log in. 04:53.250 --> 04:55.450 Let me now go and log with Google. 04:57.930 --> 05:00.170 And right here I have my environment. 05:00.170 --> 05:01.770 You can see all my API keys. 05:02.090 --> 05:03.810 Let me go and generate an API key. 05:03.850 --> 05:06.130 For example I'm using the free plan by the way. 05:06.570 --> 05:08.770 Let me go and click here this plus button. 05:10.770 --> 05:12.810 Now let's name the API key. 05:13.900 --> 05:15.260 let's call it API key. 05:15.820 --> 05:18.140 I'm going to limit my monthly usage. 05:18.420 --> 05:20.660 And yeah let's hit create. 05:20.820 --> 05:24.300 And here you can simply go you can check out you can review the API key. 05:24.340 --> 05:25.580 You can copy it. 05:25.580 --> 05:29.260 And this is the value that you should put in your um dot env file. 05:30.060 --> 05:32.620 So let me go back now to the env file. 05:33.020 --> 05:39.620 And it's also important to note that the name of the environment variable needs to be underscore API 05:39.620 --> 05:44.740 key, because that's where link chain is going to be looking for the credentials to authenticate to 05:44.780 --> 05:46.060 the services. 05:46.060 --> 05:53.500 So it's working very similarly to the OpenAI API key here, where it's going to look for that API key 05:53.540 --> 05:55.060 in the environment variables. 05:55.260 --> 05:58.820 Alrighty, let's go now to our main.py file. 05:58.820 --> 06:01.140 And let's start with the imports. 06:01.420 --> 06:07.580 And first let's go and let's load all the environment variables from the env file like we did in the 06:07.580 --> 06:08.620 previous videos. 06:08.620 --> 06:11.620 So let's import the load env function. 06:11.620 --> 06:16.670 And I want to call it in the beginning of the file here Hey there. 06:16.710 --> 06:17.470 Ethan here. 06:17.590 --> 06:19.430 Um, just a quick disclaimer. 06:19.430 --> 06:27.070 So this is being added in October 20th, 2025, and I just want to have a quick note before we jump 06:27.070 --> 06:27.430 in. 06:27.630 --> 06:30.510 So LinkedIn recently released version one. 06:30.950 --> 06:35.070 And it does have some breaking changes, especially with imports. 06:35.230 --> 06:39.470 And I'm working really hard to make sure everything in this course is compatible for you. 06:39.710 --> 06:46.590 But right now, because LinkedIn v1 has a few mismatches and bugs between the documentation and the 06:46.590 --> 06:52.670 code, I recommend you using for the imports link chain Dash classic package. 06:52.870 --> 06:59.350 That package keeps the old import stable and I'll show you the updated imports alongside the original 06:59.350 --> 07:05.070 ones in the code, so you can follow along easily in case you installed link chain version one, which 07:05.070 --> 07:07.990 is currently now the rolled out default. 07:08.110 --> 07:09.830 So thanks for bearing with me. 07:09.830 --> 07:14.510 I'm making sure everything is going to be as smooth as possible for you, and I will be refilming, 07:14.510 --> 07:18.400 re-editing and adjusting the videos for everything to be compatible. 07:18.400 --> 07:22.280 But right now, because this is so new, this is a transition period. 07:22.600 --> 07:26.960 So let me now import from LinkedIn the hub object. 07:26.960 --> 07:30.320 And the election hub is within the LinkedIn ecosystem. 07:30.320 --> 07:37.120 And it's designed for sharing and exploring prompts, chains and agents created by the community. 07:37.400 --> 07:43.520 So it serves also as a resource where developers can find inspiration, can reuse components. 07:43.720 --> 07:49.680 And we're going to use the hub in order to download a very special prompt, which is called a react 07:49.680 --> 07:50.200 prompt. 07:50.640 --> 07:51.200 All right. 07:51.200 --> 07:57.440 And we want to import the create react agent function and the agent executor class. 07:57.840 --> 07:58.280 Hey there. 07:58.320 --> 07:59.840 This is hidden here popping up. 08:00.200 --> 08:04.000 Now I want to give you a quick heads up in this section. 08:04.000 --> 08:08.200 I want us to treat the react agent as a black box. 08:08.200 --> 08:13.760 So it's going to receive the tools that we want our agent to have, and it's going to run it, and it's 08:13.760 --> 08:16.840 going to give us some awesome capabilities like we saw in the demo. 08:17.240 --> 08:21.490 Now, I am not expecting you to know how it's working under the hood. 08:21.650 --> 08:27.770 In fact, the next section is entirely dedicated for us to figure out how this magic is working. 08:28.010 --> 08:32.330 So in this section, I am not expecting you to understand how this is working. 08:32.330 --> 08:33.290 And what's the magic? 08:33.290 --> 08:35.170 This is going to be in the next section. 08:35.210 --> 08:39.650 I am expecting you just to learn the interface of how to use those objects. 08:40.090 --> 08:40.370 Right. 08:40.410 --> 08:43.650 So back to the code and let's explain those imports. 08:44.010 --> 08:53.810 So the Create react agent is a built in special function that is going to create us a runnable a link 08:53.810 --> 08:58.650 chain chain which is referred to as the react agent. 08:59.010 --> 09:02.130 Now this agent is not really an agent. 09:02.170 --> 09:03.450 It's a link chain chain. 09:03.490 --> 09:05.210 It's a runnable object. 09:05.210 --> 09:09.890 And it's going to be our reasoning engine of our react agent. 09:10.050 --> 09:13.330 And the create react agent receives an LLM. 09:13.330 --> 09:17.250 So this is going to be the LLM which is going to be our reasoning engine. 09:17.250 --> 09:19.170 It's going to receive a list of tools. 09:19.170 --> 09:21.770 So here in our example it's going to be a search tool. 09:21.970 --> 09:24.150 And it's going to receive a prompt. 09:24.310 --> 09:30.390 And the prompt we're going to give it is going to be the react prompt, which we're going to cover in 09:30.390 --> 09:30.750 depth. 09:30.750 --> 09:31.350 Trust me. 09:31.550 --> 09:38.350 And everything together is going to be a very special prompt that is being sent to the LM, and it's 09:38.350 --> 09:43.070 going to make our LM become the reasoning engine, the brain of our agent. 09:43.070 --> 09:48.270 So it's going to decide what we want to do, maybe to invoke something to call a tool, or maybe to 09:48.310 --> 09:49.350 return the answer. 09:49.390 --> 09:50.790 This is going to be the brain. 09:50.830 --> 09:53.670 Now again, you do not need to fully understand it, okay? 09:53.670 --> 09:58.590 You just need to know that it's going to receive the LM, the tools and the special prompt. 09:59.190 --> 09:59.630 All right. 09:59.670 --> 10:02.070 Now let's talk about the agent executor. 10:02.510 --> 10:07.470 And the agent executor is going to be the runtime of the agent. 10:07.950 --> 10:14.470 If the reasoning engine is going to tell us what we need to run, if we need to call the search tool 10:14.470 --> 10:19.430 with which arguments, then the agent executor is going to make the actual calls. 10:19.430 --> 10:20.870 So it's going to be the runtime. 10:20.870 --> 10:24.190 So it's going to be the things that is going to execute everything. 10:24.390 --> 10:26.240 And I'll give you a quick spoiler. 10:26.640 --> 10:31.120 The agent executor is not that complicated as it may sound. 10:31.160 --> 10:35.000 It has a very fancy name, but in the end it's a for loop. 10:35.160 --> 10:37.160 And you're going to see this in the next section. 10:37.560 --> 10:40.720 All right, so I know I introduced a lot of terminology right now. 10:40.880 --> 10:42.840 I hope I didn't freak you out too much. 10:42.880 --> 10:48.560 I remind you, the goal of this section is just to review the interface and to review the capabilities. 10:49.040 --> 10:49.480 Cool. 10:49.480 --> 10:51.440 So let's continue to other imports. 10:51.680 --> 10:55.960 We want to import chat OpenAI because we're going to be using GPT. 10:56.360 --> 11:03.120 And let's go and also import from let's go and import the Tavileh search. 11:03.560 --> 11:09.400 And Tavileh search is a link chain tool that we're going to connect to our LLM, which is going to give 11:09.400 --> 11:14.280 our agent the capability to make searches over the internet. 11:14.480 --> 11:21.720 And Tavileh search is the Tavileh API that the Tavileh team wrapped as a learning tool. 11:21.920 --> 11:27.640 And now it's compatible with the link chain ecosystem, and we can connect it to our link chain agents.