WEBVTT 00:00.440 --> 00:07.560 In this tutorial, I'm going to show you how to integrate large language models like Deepsee R1 using 00:07.600 --> 00:09.560 Grog's fast AI inference. 00:10.040 --> 00:16.840 In the case of Deepsee Karen, it's actually faster than using their official API platform and even 00:17.240 --> 00:24.920 supports tool usage, which is pretty cool because Typekit R1 is designed as a reasoning model, so 00:24.920 --> 00:28.400 it was built mainly for complex problem solving tasks. 00:28.800 --> 00:37.080 When you are using it through their official API platform, or by using operator or even self-hosting 00:37.120 --> 00:46.560 it on your own machine, you will likely run into errors when trying to use tools with your agents. 00:46.680 --> 00:49.920 I would not recommend using it to call tools directly. 00:50.120 --> 00:57.960 You can still get the best out of it by combining it with models that do support tools like GPT four 00:58.360 --> 00:59.120 or cloth. 00:59.690 --> 01:07.090 Think of Leipzig or one as the brains of the operation is great at planning making decisions, while 01:07.130 --> 01:13.530 another model takes care of two based tasks like API calls or data lookups. 01:14.290 --> 01:18.730 So just for the sake of this video, I've created two simple agents. 01:19.410 --> 01:27.570 In the first one, I've connected Leipzig, or one through grok, which is a fast AI inference platform. 01:27.810 --> 01:32.090 I will explain how it works and how you can connect to it in just a sec. 01:32.850 --> 01:37.210 And in the second agent, I'm using deep seeker one through open rotor. 01:38.570 --> 01:41.250 So I'm using the exact same model. 01:44.170 --> 01:49.530 So it's called Leipzig or one distilled llama 70 b. 01:52.290 --> 01:56.850 And the same in the second agent as our chat model. 01:58.610 --> 01:59.810 In both agents. 02:00.250 --> 02:10.090 I have connected two simple tools, one to fetch emails from my Gmail inbox, another to send emails. 02:11.090 --> 02:12.370 Okay, now let's test them. 02:12.370 --> 02:15.690 I'm going to prompt both agents with the same task. 02:16.450 --> 02:18.370 So let's ask the first agent. 02:21.530 --> 02:25.890 Give me our unread emails from today. 02:39.090 --> 02:43.290 So this agent makes API calls to Leipzig or one through grok. 02:50.650 --> 02:55.210 And as you can see, it successfully fetched all my unread emails. 02:55.210 --> 02:55.300 Nice. 02:57.660 --> 03:03.220 Now let's test the second agent which connects to Deepsee R1 to open rotor. 03:04.820 --> 03:11.100 I prepare the query inside the when clicking Test Workflow trigger node. 03:13.780 --> 03:15.140 So the query is the same. 03:16.820 --> 03:18.540 Give me all unread emails from today. 03:22.140 --> 03:23.540 So let's test our workflow. 03:29.100 --> 03:31.580 And as you can see we've got an error. 03:32.100 --> 03:39.940 So if you are using Deepsee or won by making API calls to the platform to operate on locally, and wondering 03:39.940 --> 03:42.820 why your tools are not working is not a bug. 03:43.300 --> 03:46.540 The model just doesn't support tool functions. 03:48.900 --> 03:53.300 So as you can see, no endpoints found that support tool use. 03:54.180 --> 03:56.940 Now how to use Deepsee R1 to grok. 03:57.740 --> 04:03.660 So to get started with grok, go to console.com and you will land on this page. 04:05.020 --> 04:11.220 So it's a super fast AI hosting platform around ten times faster than most services. 04:11.700 --> 04:13.820 They host a lot of models themselves. 04:15.820 --> 04:22.660 So click on the console to access the playground sign up. 04:24.380 --> 04:26.300 Head over to API keys. 04:29.460 --> 04:30.940 Create a new API key. 04:32.660 --> 04:33.500 Give it a name. 04:39.580 --> 04:40.420 Copy it. 04:43.700 --> 04:51.340 Now if you want to learn more about these services and what models they offer under documentation, 04:53.030 --> 04:53.990 Click on models. 04:56.110 --> 04:59.030 And you can see all the models they they host. 05:01.910 --> 05:03.510 So we are interested in this one. 05:07.830 --> 05:13.110 You can even use their API for free for some time on the free plan which is great. 05:13.950 --> 05:15.430 Now go back to n810. 05:19.190 --> 05:22.070 Let's remove this node so I can show you from scratch. 05:26.150 --> 05:27.310 Ah, the new chat model. 05:30.790 --> 05:31.830 Search for grog. 05:33.390 --> 05:33.710 Grog. 05:33.710 --> 05:34.630 Chat model node. 05:41.150 --> 05:46.350 Create new credential and paste your API key right here. 05:48.150 --> 05:48.830 It's safe. 05:58.750 --> 06:01.150 Now you'll have access to all the models. 06:02.150 --> 06:09.030 So just select Leipzig or one distill llama 70. 06:11.430 --> 06:12.590 And you are all set. 06:13.230 --> 06:20.910 Here is a quick comparison of the deep sea or 170 model when hosted on different platforms, including 06:20.910 --> 06:25.990 grok Operator and Deep Six official API. 06:26.630 --> 06:29.990 And for reference, OpenAI's own model. 06:30.550 --> 06:42.950 When you look at the relative cost grok operator and Deep Six official API are about 25 to 27 times 06:42.950 --> 06:44.950 cheaper than OpenAI one. 06:45.430 --> 06:46.870 So that's a massive difference. 06:48.070 --> 06:49.630 And that's pretty much it.