WEBVTT 00:00.160 --> 00:06.200 In this tutorial, I'm going to teach you how to build a team of AI agents without writing a single 00:06.200 --> 00:07.000 line of code. 00:07.160 --> 00:11.080 Let's do a quick demo and see what this team of agents can do for you. 00:11.800 --> 00:13.400 Let me show you how this works. 00:14.000 --> 00:21.040 So we will communicate just with our supervisor agent via telegram, because it acts as the coordinator 00:21.040 --> 00:25.040 and delegates the tasks to three specialized sub agents. 00:25.640 --> 00:29.800 So the research agent, email agent, and calendar agent. 00:31.080 --> 00:33.000 We can use both voice or text. 00:33.040 --> 00:35.760 You can give your agent multiple tasks at once. 00:36.440 --> 00:38.160 So I've prepared a prompt. 00:38.520 --> 00:40.440 So I'm just gonna copy and paste it. 00:41.200 --> 00:48.120 So let's test it and ask to check what's on my schedule today etc.. 00:48.120 --> 00:51.080 So I'm gonna click Test Workflow. 00:53.400 --> 00:54.560 Send this message. 00:57.400 --> 01:00.600 So I want it to check what's on my schedule today. 01:00.640 --> 01:04.400 Send an email to Mark Johnson to cancel our meeting and let him know. 01:04.880 --> 01:06.640 I will contact him later at two. 01:06.680 --> 01:09.800 Reschedule and give me news about Leipzig from today. 01:11.000 --> 01:17.240 So our supervisor agent calls the calendar and research agents to fulfill these tasks. 01:18.960 --> 01:26.480 The calendar agent fetches the details from my Google calendar and sends them back to the main agent 01:27.240 --> 01:28.960 to create a response for me. 01:28.960 --> 01:35.400 And the research agent scrapes the internet to find the news articles about Dopesick. 01:40.200 --> 01:41.240 It takes a moment. 01:47.680 --> 01:49.800 So now we got a response. 01:52.360 --> 01:54.200 Here is a summary of your requests. 01:54.720 --> 02:01.120 So today's schedule Projects meeting client brainstorming with Mark and Jim. 02:01.120 --> 02:01.800 Upper body. 02:04.040 --> 02:06.200 So now let's check my calendar. 02:08.400 --> 02:09.520 And this is correct. 02:11.120 --> 02:13.120 So now let's check my inbox. 02:18.680 --> 02:23.960 And yes, we got an email about the meeting cancellation. 02:24.600 --> 02:28.480 Hi Mark, I hope this message finds you well, etc.. 02:30.880 --> 02:32.760 Now our agent also. 02:35.040 --> 02:39.520 Wants the latest news about Leipzig. 02:41.240 --> 02:42.320 So let's have a look. 02:43.880 --> 02:45.600 So recent news about Leipzig. 02:48.480 --> 02:53.680 And we can see all the headlines here along with links to the original articles. 02:56.830 --> 02:58.110 So let's check one of them. 03:01.870 --> 03:02.630 Investigates. 03:02.630 --> 03:04.430 Deep six use of its models. 03:05.710 --> 03:06.350 Open it. 03:09.910 --> 03:12.310 And yep, this one was posted today. 03:14.710 --> 03:18.110 We can also send a voice message to our agent. 03:19.070 --> 03:24.710 I will show you two ways to generate a voice for other agents, so you will have plenty of voices to 03:24.750 --> 03:25.390 choose from. 03:25.830 --> 03:29.870 You can even generate or clone your own voice or someone else's. 03:30.430 --> 03:37.550 You can also provide our agent with a custom knowledge base like contact details, frequently ask questions, 03:37.550 --> 03:40.910 training materials, product lists. 03:41.110 --> 03:43.710 In this setup, we have contact details. 03:44.390 --> 03:52.710 So we store it in Google Sheets in Google Drive, but you can use any storage like any storage format 03:52.710 --> 03:55.430 Not like PDFs, docs, etc.. 03:55.830 --> 03:57.990 You can also implement persistent memory. 03:58.430 --> 03:58.830 All right. 03:58.830 --> 04:00.030 So this is your canvas. 04:01.150 --> 04:09.230 We'll start by renaming the workflow from my workflow to the team of agents. 04:17.990 --> 04:21.270 And press enter just to confirm it. 04:23.310 --> 04:29.430 Next I strongly recommend changing your time zone so the times are correct for all tasks. 04:30.510 --> 04:32.350 So click on these three dots. 04:33.990 --> 04:35.070 Click on settings. 04:37.990 --> 04:39.390 And select your time zone. 04:49.870 --> 04:50.630 Save it. 04:57.110 --> 05:00.070 So our first step is creating the supervisor agent. 05:00.670 --> 05:02.670 So let's add our first node. 05:03.510 --> 05:05.110 So click on this plus button. 05:07.310 --> 05:08.630 Type a agent. 05:10.950 --> 05:11.670 Click on it. 05:13.870 --> 05:14.830 Close it for now. 05:18.910 --> 05:26.350 So basically nodes represent actions like fetching data, sending emails or making API calls. 05:26.910 --> 05:31.270 So let's rename our agent to. 05:33.670 --> 05:35.270 Supervisor or agent. 05:41.070 --> 05:42.510 Click rename to confirm it. 05:44.910 --> 05:49.030 And now we have to set up a trigger which starts the workflow. 05:50.350 --> 05:57.550 So usually it's the communication channel you want to communicate with your agent. 05:58.110 --> 06:01.190 So every workflow needs a trigger to be activated. 06:01.630 --> 06:04.230 And by default we have this chat on message. 06:05.230 --> 06:08.750 So let's get rid of it. 06:13.870 --> 06:15.750 So click on this plus button again. 06:17.670 --> 06:26.910 And you can activate the workflow manually on App events which will use in this setup. 06:27.350 --> 06:35.470 But you can also run your workflow on a schedule like every day, every hour, etc. you can use webhook 06:35.470 --> 06:38.470 call or on the form submission. 06:39.830 --> 06:42.830 So now click on App Event. 06:44.790 --> 06:49.230 And now we have to select the platform We want to communicate with. 06:50.070 --> 06:51.310 You can choose WhatsApp. 06:53.950 --> 06:54.670 Telegram. 06:56.430 --> 06:57.030 Slack. 06:59.590 --> 07:00.790 And other platforms. 07:03.310 --> 07:06.270 So we will use telegram because it's very easy to set up. 07:09.910 --> 07:11.950 So click on this telegram node. 07:14.390 --> 07:19.990 I've got a tutorial on how to set up WhatsApp in six minutes, so check it out if you prefer WhatsApp. 07:21.630 --> 07:24.150 So the trigger we need is on message. 07:25.590 --> 07:30.870 So this means the workflow will start when we send a message to our agent on telegram. 07:31.910 --> 07:32.870 So click on that. 07:34.550 --> 07:40.630 First we need to set up our telegram credentials to connect our agents to telegram. 07:41.950 --> 07:49.580 So click on this drop down and select Create New Credential. 07:51.580 --> 07:53.740 Now we need to provide the access token. 07:54.020 --> 07:56.540 And edit N makes it easy. 07:57.180 --> 07:59.780 So you can click on Open Docs. 08:00.380 --> 08:06.140 If you need help filling out these fields to follow the instructions yourself. 08:08.300 --> 08:09.540 So it looks like that. 08:14.460 --> 08:18.620 Or just follow along with me right now. 08:18.620 --> 08:21.860 If we send a message on telegram, nothing will happen. 08:22.220 --> 08:28.060 That's because N810 is not listening for events yet and our workflow is not activated. 08:29.420 --> 08:30.500 So let's close it. 08:35.300 --> 08:41.420 And at this point, we will not get answer from our agent because it's not connected to a large language 08:41.420 --> 08:46.380 model, which essentially acts as the agent's brain. 08:46.820 --> 08:50.860 So to add a brain to our agent, click on this plus button. 08:52.420 --> 08:54.820 And here are all the available options. 08:55.500 --> 09:04.940 We can integrate all of OpenAI's models like GPT four, GPT four, mini, etc. you can use cloud 3.5 09:04.980 --> 09:07.980 sonnet by using this node. 09:08.700 --> 09:11.500 We can use grok here. 09:11.500 --> 09:17.180 You can choose the model you would like to use as a brain for your AI agent. 09:17.740 --> 09:23.460 For this tutorial, we'll be using GPT four mini because it's very fast and cost effective. 09:24.180 --> 09:28.460 And I prepared a well detailed prompts to guide it so it's good enough. 09:28.460 --> 09:33.540 But you can choose the other models like GPT four. 09:37.700 --> 09:39.740 Now let's go back to the canvas. 09:41.780 --> 09:43.300 Now let's move this trigger. 09:43.300 --> 09:45.860 Note the left side. 09:47.460 --> 09:49.660 Connect it to a supervisor agent. 09:52.300 --> 09:54.620 And now we need to add a memory node. 09:54.660 --> 09:55.580 As you can see. 09:56.580 --> 10:01.900 So click on this plus button and choose Window Buffer memory. 10:05.820 --> 10:07.660 We need to retrieve this session ID. 10:10.020 --> 10:15.380 Because the memory node allows the agent to remember past messages from our conversation. 10:16.300 --> 10:21.180 So change the session ID to define below. 10:24.740 --> 10:25.860 Close it for now. 10:27.620 --> 10:29.020 Click on Test Workflow. 10:31.140 --> 10:33.660 And send any message on on telegram. 10:34.140 --> 10:35.340 I'm just gonna say hi. 10:42.140 --> 10:43.580 Now open this note again. 10:46.100 --> 10:52.060 And you can get a conversation ID from a trigger note by going to mapping. 10:54.620 --> 10:55.900 Just grab this ID. 10:58.300 --> 10:59.420 And drop it over here. 11:02.340 --> 11:07.660 So now the context window length sets how many pass messages to include. 11:07.900 --> 11:13.140 So for example let's say we want our agent to remember ten previous messages. 11:16.500 --> 11:17.660 Let's close it for now. 11:20.460 --> 11:21.540 Let's hit save. 11:28.500 --> 11:29.460 This workflow again. 11:32.180 --> 11:33.060 Set message. 11:34.700 --> 11:35.340 Anything. 11:40.580 --> 11:41.820 Now we got another error. 11:42.420 --> 11:45.300 So let's open our agent. 11:48.780 --> 11:51.420 And as you can see there is no prompt specified. 11:51.860 --> 11:59.300 So we need to set up the prompt to give instructions for our agent what to do and how to behave. 12:00.900 --> 12:09.900 Now under prompt click drop down, select define below and what you can do. 12:10.140 --> 12:16.660 You can just drag the message we received from telegram and drop it over here. 12:18.900 --> 12:21.460 So this is our message we sent on telegram. 12:21.900 --> 12:27.940 Now under options add a new option and select System Message. 12:33.100 --> 12:38.570 Switch to expression Then expand the view. 12:41.010 --> 12:48.650 So the simplest way to structure the prompt is to assign the role objective and provide additional information. 12:49.970 --> 12:52.330 So I'm just going to paste the prompt. 12:54.850 --> 12:58.010 So it can be as simple as you are a professional assistant. 12:58.330 --> 13:04.210 Your primary goal is to manage contacts, email and calendar events and perform research tasks for the 13:04.210 --> 13:04.690 user. 13:05.090 --> 13:06.690 And now additional information. 13:09.570 --> 13:12.210 So you have access to three sub agents to help you. 13:12.650 --> 13:16.170 Email agent, calendar agent, and research agent. 13:16.490 --> 13:20.850 So I will provide you with a template so you can just copy and paste it. 13:21.290 --> 13:29.170 So make sure to do this before using your team of agents, as I have added some comments to avoid errors 13:29.970 --> 13:34.010 because an optimized prompt is crucial for building effective agents. 13:35.210 --> 13:35.690 All right. 13:35.690 --> 13:37.490 So let's test our agent again. 13:41.930 --> 13:45.130 Click this workflow and send a simple message on telegram. 13:52.650 --> 13:53.250 Great. 13:53.970 --> 13:56.290 So this time we successfully received a response. 13:57.770 --> 14:00.130 When you open this now again you can see the output is. 14:00.130 --> 14:00.450 Hello. 14:00.450 --> 14:01.890 How can I assist you today? 14:02.690 --> 14:05.330 Just need to send this message back to telegram. 14:06.170 --> 14:13.210 So back in the canvas add a new note and search for telegram. 14:17.930 --> 14:20.610 Scroll through the options and select. 14:22.970 --> 14:24.490 Send a text message. 14:30.010 --> 14:30.850 Which is right here. 14:32.530 --> 14:34.410 Next, choose the credentials. 14:36.850 --> 14:39.690 Next, choose the credentials you set up earlier. 14:42.530 --> 14:46.090 Then we have to grab the chat ID from the telegram trigger. 14:48.010 --> 14:49.010 Just expand it. 14:51.690 --> 14:54.890 And drag and drop it over here to chat. 14:55.050 --> 14:56.250 To chat ID field. 14:58.450 --> 15:05.810 Now for the text field we need to assign the A agents output. 15:07.010 --> 15:10.050 So drag and drop output over here. 15:13.650 --> 15:17.730 So let's run a test and have a look. 15:22.970 --> 15:23.290 Yes. 15:23.290 --> 15:23.770 That's great. 15:23.770 --> 15:29.370 So we just received a message from our agent. 15:29.890 --> 15:33.570 But we got some extra text, which is redundant. 15:34.090 --> 15:35.170 So let's fix it. 15:37.130 --> 15:38.290 Click on add field. 15:41.210 --> 15:43.050 Append an item attribution. 15:45.850 --> 15:46.930 And switch it off. 15:50.290 --> 15:51.650 Let's back to our canvas. 15:53.530 --> 15:57.130 And now we can explore how to handle audio messages. 15:57.450 --> 15:59.130 First, run a test workflow. 16:01.970 --> 16:03.130 Switch to telegram. 16:07.490 --> 16:10.010 And we'll press and hold the record button. 16:12.370 --> 16:15.890 Say something and then release to send it. 16:19.130 --> 16:23.170 Now back in our workflow, we got an error. 16:26.970 --> 16:31.920 And when we check the trigger note and the schema and scroll down. 16:33.520 --> 16:41.080 It looks different this time because instead of text, we now have a voice note that contains a file 16:41.080 --> 16:41.520 ID. 16:42.800 --> 16:46.440 This means we need to handle messages differently based on their type. 16:47.240 --> 16:53.520 So to do this, we will insert a switch node between the diagram trigger node and the agent node. 16:54.480 --> 16:55.800 So go back to canvas. 16:56.400 --> 16:58.160 Let's close this error message. 16:59.880 --> 17:03.880 Zoom out a bit and let's search for each node. 17:10.360 --> 17:11.920 Now let's remove this connection. 17:15.520 --> 17:16.840 And move it over here. 17:21.080 --> 17:24.920 So the switch node allows us to create separate processing paths. 17:25.760 --> 17:30.040 Connect the telegram signal to the switch node. 17:35.880 --> 17:41.960 Let's open the switch node and for the first condition, we will check if the message contains a file 17:41.960 --> 17:44.000 ID inside the voice object. 17:44.800 --> 17:45.880 So scroll down. 17:46.280 --> 17:51.560 Let's find our file within the voice object. 17:53.360 --> 17:56.440 Grab it and drop over here. 17:57.400 --> 18:07.400 We assign this to the value field, then select string and exist. 18:08.080 --> 18:15.400 It simply means if a voice file id is detected, we route the workflow through the audio processing 18:15.400 --> 18:15.880 path. 18:16.920 --> 18:22.040 So to keep it organized, let's just rename it to audio. 18:27.200 --> 18:28.120 And let's close it. 18:28.720 --> 18:30.600 And now we can see the audio label. 18:30.840 --> 18:35.760 Since we also need to handle text messages let's make that adjustment. 18:37.280 --> 18:39.320 So I'm going to run test Workflow again. 18:42.440 --> 18:44.640 And send the basic text message. 18:47.920 --> 18:50.680 And now when we open the switch node. 18:53.320 --> 18:57.440 It appears red because the expected field is not present. 18:57.960 --> 19:02.760 So instead we have a text property from the telegram trigger. 19:03.160 --> 19:06.280 So to fix this we have to add a new route. 19:07.120 --> 19:09.040 So add a new routing rule. 19:11.600 --> 19:14.320 Use text field as the value. 19:20.640 --> 19:24.280 Again string Exist. 19:26.440 --> 19:28.400 And let's rename it to text. 19:39.080 --> 19:40.000 Let's close it. 19:40.920 --> 19:44.000 And now our switch node has two different routes. 19:44.520 --> 19:46.880 One for audio and one for text. 19:47.640 --> 19:52.760 So via text message comes in it will be directed to the agent for processing. 19:53.400 --> 19:55.040 When we connect these two nodes. 19:56.120 --> 19:59.880 But if we receive an audio message we need to handle it differently. 20:00.400 --> 20:04.800 So first let's add a new node and send another audio file into the flow. 20:05.360 --> 20:07.720 And now we can see the audio message. 20:08.080 --> 20:09.440 Trigger the audio route. 20:10.400 --> 20:12.040 Let's go back to our canvas. 20:13.440 --> 20:15.880 And now we have to download our voice message. 20:16.760 --> 20:18.800 So again click on Add node. 20:20.480 --> 20:21.680 Search for telegram. 20:26.640 --> 20:27.840 Select the get file. 20:29.640 --> 20:35.720 Now pass the file ID from the voice object into this field. 20:39.600 --> 20:41.040 And let's test this node. 20:48.600 --> 20:50.600 Here the file is successfully downloaded. 20:51.800 --> 20:53.280 I will rename this node to. 20:55.480 --> 20:57.080 Node file. 21:02.520 --> 21:03.360 Confirm it. 21:07.800 --> 21:08.680 And hit save. 21:10.640 --> 21:13.960 The next step is to transcribe the audio to text. 21:14.480 --> 21:15.920 So click add node. 21:17.760 --> 21:18.830 Search for OpenAI. 21:24.470 --> 21:25.150 Scroll down. 21:29.150 --> 21:30.470 Transcribe a recording. 21:31.990 --> 21:33.710 This is the action we want to achieve. 21:36.150 --> 21:38.470 We don't need to change anything in this note. 21:39.950 --> 21:41.110 So run a test. 21:49.110 --> 21:49.750 Great. 21:51.390 --> 21:57.310 Now that we have the transcribed text, we need to make sure our agent processes it correctly. 21:57.670 --> 21:59.510 Now let's make it more organized. 22:00.430 --> 22:01.630 I'm going to move it here. 22:04.670 --> 22:09.350 And we can connect the transcribed text to our AI agent. 22:10.790 --> 22:19.030 Now when we open the supervisor AI agent node there is a small issue we need to fix, because when we 22:19.030 --> 22:28.550 receive a text message from telegram, it comes inside an object called message, which contains a text 22:28.550 --> 22:29.070 field. 22:30.310 --> 22:35.110 However, when we got a transcription from an audio message. 22:37.310 --> 22:38.510 So let's expand it. 22:40.990 --> 22:49.470 The text is stored directly in the field called text without being wrapped in in a message object. 22:53.110 --> 22:59.950 Now when we open the supervisory agent node, there is a small issue we need to fix. 23:00.310 --> 23:07.790 Because when we receive a standard text message from telegram, it comes inside an object called message, 23:09.030 --> 23:11.630 which contains a text field. 23:12.910 --> 23:17.670 However, when we got a transcription from an audio message. 23:18.670 --> 23:19.990 So let's expand it. 23:30.070 --> 23:38.030 The text is stored directly in a field called text without being wrapped in a message object. 23:40.830 --> 23:47.590 We can quickly fix it and simplify the text format before sending it to the agent. 23:49.030 --> 23:50.430 So go back to canvas. 23:53.230 --> 23:56.910 Now we have to add a set node right after the switch node. 23:57.590 --> 23:59.950 So click on this small plus button. 24:01.190 --> 24:03.150 Search for set node. 24:07.110 --> 24:11.070 Click Add field and name it text. 24:16.750 --> 24:17.750 Let's scroll down. 24:20.470 --> 24:26.390 Let's grab text and assign it to the value called text. 24:31.230 --> 24:33.190 Now let's go back to our canvas. 24:36.790 --> 24:43.550 And if the text comes from telegram, it will extract text from the message object. 24:44.950 --> 24:49.830 And if it comes from the transcription it will use the text directly. 24:52.030 --> 24:56.230 So let's connect this node to our agent. 24:57.470 --> 25:05.390 And our agent will always receive a standardized text field no matter where the input comes from. 25:06.790 --> 25:10.070 So I will rename this node to set text. 25:19.430 --> 25:19.830 Film. 25:22.790 --> 25:30.230 Now, if we send a voice message, we might want the agent to reply with the voice too, not just text. 25:30.790 --> 25:37.630 So what we can do is add a new node between our agent and the telegram response node. 25:38.630 --> 25:44.470 So just by clicking on this plus button let's search for if node. 25:46.710 --> 25:50.030 What we will do here is add an if condition. 25:50.950 --> 25:58.070 So what we want to say is let's go to our telegram trigger node. 26:00.950 --> 26:05.950 And we want to check if the voice ID exists. 26:09.700 --> 26:11.340 So let's grab this file ID. 26:13.380 --> 26:14.540 And drop over here. 26:16.100 --> 26:25.900 If you can see this voice object that's mean you need to add some data to to the flow. 26:26.580 --> 26:33.020 So just simply test a workflow and send a voice message on telegram. 26:35.340 --> 26:44.180 So now you want to click on drop down select string and exist. 26:48.180 --> 26:54.500 So basically now we have two branches true and false. 26:55.500 --> 27:00.500 So if true we want to respond with audio. 27:01.540 --> 27:03.300 So we have to remove this connection. 27:05.860 --> 27:09.260 And if First, we will simply return text. 27:09.940 --> 27:12.100 So you have to connect these two nodes. 27:13.820 --> 27:17.220 And now we want to generate audio instead. 27:19.140 --> 27:21.020 So let's click on Add nodes. 27:24.060 --> 27:25.220 Search for open AI. 27:31.180 --> 27:37.220 Go to Audio Actions and generate audio. 27:41.060 --> 27:42.580 Select your credentials. 27:46.220 --> 27:50.660 And here you can select a voice. 27:53.500 --> 27:55.060 So I go with echo. 27:58.060 --> 27:59.900 And you can leave this node as it is. 28:00.700 --> 28:02.100 So now you have a setup. 28:04.220 --> 28:10.260 And the two conditions if the file ID exists. 28:13.220 --> 28:14.860 So the condition is true. 28:15.820 --> 28:19.500 Generate an audio if doesn't exist. 28:20.500 --> 28:22.260 Return a text message. 28:22.700 --> 28:27.220 Now let's rename this node to generate audio. 28:38.820 --> 28:40.380 Let's test our workflow. 28:41.780 --> 28:46.100 I'm just gonna say a short voice message on telegram. 28:48.140 --> 28:48.340 Hi. 28:48.380 --> 28:49.100 How are you doing? 29:02.020 --> 29:05.700 So we should Just get a response in a second. 29:06.300 --> 29:06.540 Yes. 29:06.540 --> 29:07.380 You got an error? 29:07.380 --> 29:07.740 Which is? 29:07.780 --> 29:08.980 Which was expected. 29:09.740 --> 29:10.860 Let's open this node. 29:12.660 --> 29:16.860 Now we got some data in the flow so we can. 29:18.900 --> 29:26.420 Expand our supervisor agent node and grab this output property. 29:28.820 --> 29:29.780 And drop it here. 29:35.220 --> 29:37.140 Let's test our workflow again. 29:37.980 --> 29:43.780 Just before that I've got a great trick for you in every trigger node. 29:44.740 --> 29:51.740 Doesn't matter if it's a telegram, WhatsApp or any other platform, there is an option to pin data. 29:52.660 --> 29:56.900 So there is a pin icon in the top right corner. 29:57.300 --> 30:03.660 So when you click on it, you'll pin data within your workflow and you can simply unpin it. 30:06.060 --> 30:16.220 So it means that the message, voice message or text message you send area will be saved in this node, 30:17.620 --> 30:21.860 so you don't need to send it over and over again. 30:22.620 --> 30:28.940 Now you just need to click on Test Workflow and it will use the previous message. 30:39.180 --> 30:39.660 Great. 30:39.660 --> 30:45.740 So now we generated the ice voice response. 30:45.740 --> 30:48.100 So we need to send it back to telegram. 30:49.620 --> 30:51.260 So add the telegram node. 30:55.580 --> 30:56.460 Scroll down. 30:58.970 --> 30:59.370 And. 31:03.930 --> 31:05.010 We are looking for. 31:05.050 --> 31:06.850 Send an audio file. 31:12.170 --> 31:13.090 Which is right here. 31:17.250 --> 31:19.370 Now switch to the schema. 31:21.690 --> 31:26.810 And we need to select the chat ID from the telegram trigger node. 31:26.810 --> 31:27.810 So expand it. 31:32.010 --> 31:33.930 Grab this chat ID. 31:36.650 --> 31:37.770 And move over here. 31:39.530 --> 31:40.730 Also we need to. 31:44.090 --> 31:51.170 Enable binary file so it knows to send the audio instead of text okay that's it. 31:51.170 --> 31:52.490 And now let's test it out. 31:55.530 --> 31:57.730 I'm just gonna click on Test Workflow. 32:11.890 --> 32:12.170 Great. 32:12.170 --> 32:14.890 I just received an audio file. 32:15.170 --> 32:16.810 Let's listen to it. 32:20.570 --> 32:22.690 I'm here and ready to assist you. 32:22.730 --> 32:24.090 How can I help you today? 32:24.170 --> 32:24.570 All right. 32:24.570 --> 32:25.090 Great. 32:28.170 --> 32:34.410 Now let's check if our agent will respond to a text message properly. 32:35.050 --> 32:37.570 So I'm just gonna unpin data. 32:40.650 --> 32:41.610 Test workflow. 32:44.810 --> 32:48.410 And send short text message on telegram. 32:49.290 --> 32:52.290 I'm just gonna say hi. 32:52.970 --> 32:54.330 Can you assist me, please? 32:59.410 --> 32:59.890 Great. 32:59.890 --> 33:04.810 Now we have received a text respond back to this branch. 33:06.690 --> 33:07.930 So let's have a look. 33:13.450 --> 33:14.090 Absolutely. 33:14.130 --> 33:15.290 How can I assist you today? 33:15.290 --> 33:17.290 Please let me know what you need help with. 33:18.490 --> 33:19.210 Alright, great. 33:19.250 --> 33:26.130 Now what I like to do, I like to summarize the message before generating the voice. 33:26.130 --> 33:32.410 So this makes the voice message more to the point and also reduces token cost. 33:32.810 --> 33:36.930 And obviously we don't want to listen to long messages. 33:36.930 --> 33:45.490 So what we are going to do now is we are going to grab an open AI message model. 33:45.490 --> 33:49.210 So let's search for open AI. 33:52.690 --> 33:58.570 Just bit between the teeth and generate audio notes. 33:59.010 --> 34:00.810 So let's click on this plus button. 34:02.810 --> 34:03.730 Search for open AI. 34:05.890 --> 34:09.570 And under Text actions select Message Model. 34:12.250 --> 34:18.610 Connect your credentials and select the large language model you want to use. 34:19.490 --> 34:22.050 So I want to use GPT for me. 34:23.650 --> 34:27.770 Now you want to get some data into this node. 34:29.290 --> 34:32.290 So I'm just gonna say test the workflow. 34:34.050 --> 34:36.170 Send a short voice message on telegram. 34:38.010 --> 34:38.210 Hi. 34:38.210 --> 34:39.210 Can you help me, please? 34:52.970 --> 34:53.650 All right. 34:57.330 --> 35:01.970 So now we want to give it give a prompt to our model. 35:03.450 --> 35:04.250 So let's. 35:06.890 --> 35:08.530 Ask it to summarize. 35:12.490 --> 35:13.290 The message. 35:14.850 --> 35:18.490 So I'm going to provide you with the prompt. 35:19.530 --> 35:21.890 So I'm going to copy and paste it now. 35:24.090 --> 35:24.690 There we go. 35:25.570 --> 35:28.970 I want this model to summarize this incoming information. 35:29.810 --> 35:32.810 So the message from the previous node. 35:34.690 --> 35:38.770 So I'm going to expand it to show you how it looks like. 35:39.650 --> 35:46.370 So summarize this incoming information which is in this format is basically our previous message for 35:46.370 --> 35:51.120 the calendar events provides only the name of the event and time for or you may provide the title of 35:51.160 --> 35:52.640 the email and summary of the body. 35:52.680 --> 35:55.000 No time and output, only plain text. 35:55.040 --> 35:58.000 There should be no markdown formatting or any symbols. 35:59.360 --> 36:00.480 So let's close it. 36:03.880 --> 36:11.240 We want the summary in plain text, so we'll make sure there are no symbols or special characters because 36:11.240 --> 36:14.240 the AI generated voice will read everything. 36:15.240 --> 36:15.760 That's it. 36:15.760 --> 36:16.880 Let's test it out. 36:19.800 --> 36:25.080 It's going to ping data again and test the workflow. 36:43.920 --> 36:45.400 Let's listen to the response. 36:49.720 --> 36:50.520 Of course. 36:50.520 --> 36:51.640 I'm here to help. 36:51.680 --> 36:53.880 Please let me know what you need assistance with. 36:54.960 --> 36:55.520 Perfect. 36:55.520 --> 36:58.080 So now your agent responds with voice two. 36:58.680 --> 37:03.680 And if the voice is too long, it will be summarized before sending back to you. 37:03.840 --> 37:10.040 What if you want a different voice because you don't like any of the OpenAI's options? 37:11.720 --> 37:13.760 So there is a solution for that. 37:14.560 --> 37:21.920 Also, if you want to clone your own voice or someone else's voice, you can use voices from 11 labs, 37:21.920 --> 37:24.880 which is the best AI voice generation platform. 37:25.280 --> 37:29.720 So this is optional, but I'm going to show you how to set this up. 37:30.320 --> 37:30.760 All right. 37:30.760 --> 37:38.560 So now that we have our summarized text, we are going to send it to 11 labs to generate the actual 37:38.600 --> 37:44.520 voice file instead of sending it to OpenAI voice models. 37:44.560 --> 37:49.760 So let's remove this connection and this one too. 37:53.040 --> 38:01.680 We want to feed this summarized text into post requests, and we'll generate audio file using a voice 38:01.680 --> 38:03.320 from platform. 38:03.720 --> 38:12.440 We need to set up the HTTP request in order to use a Post request to send our text to 11 labs. 38:12.840 --> 38:16.200 So let's search for http request node. 38:20.760 --> 38:24.960 Now we need to change the method from get to post. 38:27.400 --> 38:34.520 And we need the URL as some headers and JSON body. 38:35.320 --> 38:36.680 Now go to 11 labs. 38:36.680 --> 38:38.720 You can find a link in the description. 38:39.280 --> 38:39.920 Sign up. 38:40.120 --> 38:41.080 Create an account. 38:41.400 --> 38:48.080 Now when I was testing this, I ran into some credential errors on the free plan, so I recommend you 38:48.080 --> 38:54.480 to upgrade to the starter plan for $5 a month, which gives you 40,000 credits. 38:54.920 --> 39:01.240 That should last quite a while unless you are generating really long voice files in 11 laps. 39:01.280 --> 39:05.360 Go to the text to speech section. 39:05.680 --> 39:14.680 So create speech and now you can copy and paste values from this code snippet. 39:15.840 --> 39:18.120 So first we need this URL. 39:18.400 --> 39:19.520 So let's copy it. 39:25.640 --> 39:27.960 Go back to 2010. 39:30.120 --> 39:31.360 And paste it right here. 39:37.760 --> 39:39.920 Right now we need some headers. 39:39.920 --> 39:42.790 So enable send headers. 39:46.190 --> 39:48.830 Go back to the platform. 39:53.430 --> 39:58.350 Copy this one of the two headers. 39:59.470 --> 40:01.030 Go back to N810. 40:04.230 --> 40:05.590 Pass it as a first name. 40:08.390 --> 40:11.270 Now as a value, we need to generate our API keys. 40:15.830 --> 40:18.110 In your account. 40:19.950 --> 40:25.750 Go to settings API keys. 40:30.430 --> 40:31.630 Create API key. 40:33.870 --> 40:34.750 Give it a name. 40:37.670 --> 40:38.350 Create. 40:44.430 --> 40:45.190 Copy it. 40:46.870 --> 40:48.070 Go back to it again. 40:50.350 --> 40:51.430 And paste it here. 40:55.230 --> 40:58.510 Now we need to add the second parameter. 40:59.510 --> 41:00.430 So add new one. 41:02.110 --> 41:05.270 Go to API documentation. 41:09.430 --> 41:10.790 Copy this line. 41:18.230 --> 41:19.470 So this is our value. 41:27.950 --> 41:30.830 With that all right. 41:30.830 --> 41:32.870 So now we need to send a body. 41:33.590 --> 41:35.230 So basically JSON file. 41:37.030 --> 41:38.070 So enable it. 41:38.070 --> 41:38.110 Hybrid. 41:45.230 --> 41:47.110 Switch to using JSON. 41:49.590 --> 41:50.950 Switch to expression. 41:53.670 --> 41:55.110 Now from the previous node. 41:56.630 --> 42:02.710 Grab this property content and drop over here. 42:05.190 --> 42:16.030 And now we need to convert this JSON file into JSON string and clean it up a bit because it will return 42:16.030 --> 42:19.190 some symbols we don't want in our message. 42:22.870 --> 42:25.670 So I'm going to provide you with a code snippet. 42:26.150 --> 42:28.510 Now I'm going to paste it here. 42:32.910 --> 42:36.710 So as you can see JSON dot Stringify. 42:36.750 --> 42:45.150 So it's going to convert it to JSON string and remove all the special characters. 42:45.990 --> 42:47.150 So it should work now. 42:48.910 --> 42:50.590 Now let's scroll up. 42:51.790 --> 43:00.790 And now in our URL after text to speech forward slash, we can paste the ID of the voice we want to 43:00.830 --> 43:01.230 use. 43:02.070 --> 43:05.110 So let's go back to our Level Labs account. 43:13.430 --> 43:16.150 And in the left menu click on voices. 43:19.270 --> 43:22.270 On this page you can choose the voice you like. 43:23.270 --> 43:24.510 So you can test. 43:25.030 --> 43:29.750 I believe there are over 3000 voices to choose from. 43:33.190 --> 43:34.750 So I'm going to pick this one. 43:34.990 --> 43:36.350 Just trust yourself. 43:36.390 --> 43:38.190 Then you will know how to live. 43:39.310 --> 43:46.910 Now click on view and right here in the right bottom corner you can see an ID. 43:47.790 --> 43:52.830 Just copy the ID go back to n810. 43:59.830 --> 44:06.310 And after the text to speech forward slash paste this ID. 44:06.670 --> 44:13.830 Now in the telegram response node we need to provide a name of the field. 44:14.630 --> 44:19.230 So under additional fields click on Add field. 44:20.550 --> 44:29.310 Select File Name and give it a name data dot. 44:29.550 --> 44:33.060 And it needs to be this format this file extension. 44:35.100 --> 44:36.300 To copy and paste it. 44:37.980 --> 44:39.060 Now let's close it. 44:41.020 --> 44:41.780 Hit save. 44:46.380 --> 44:51.620 I'm going to unpin data in the telegram trigger node. 44:52.900 --> 44:54.660 So simply unpin. 44:58.420 --> 44:59.140 Save again. 45:00.660 --> 45:01.900 Now let's test it out. 45:01.900 --> 45:05.820 So I'm going to send a short voice message on telegram. 45:13.380 --> 45:13.740 Hi. 45:13.740 --> 45:18.500 Can you please help me write a viral LinkedIn post? 45:38.340 --> 45:39.820 All right, we got a response. 45:39.860 --> 45:40.980 Let's listen to it. 45:46.060 --> 45:52.380 To help write a viral LinkedIn post, I need one topic or message to convey two personal experiences 45:52.380 --> 45:54.020 or insights to share three. 45:54.220 --> 45:56.900 Target audience for specific call to action. 45:57.060 --> 45:57.980 That sounds good. 46:00.980 --> 46:01.780 That's awesome. 46:01.820 --> 46:07.540 Our agent can already read and listen to our messages and respond with text or audio. 46:08.020 --> 46:13.300 But what makes an agent powerful is its ability to interact with the world. 46:13.620 --> 46:18.620 In other words, connecting with external systems, apps, and data sources. 46:19.020 --> 46:26.900 So this means it can, for example, read and reply to emails, manage calendar, search the web for 46:27.180 --> 46:31.620 real time information Update spreadsheets or databases. 46:32.580 --> 46:33.660 Fashion process. 46:33.660 --> 46:34.940 Customer requests. 46:36.260 --> 46:40.660 So you could implement these functionalities by adding tools. 46:43.540 --> 46:55.100 So simply by clicking on this plus button as N810 has direct integrations with many platforms such as 46:55.140 --> 46:58.380 Google Apps, CRMs, Wikipedia. 46:59.940 --> 47:05.660 Then set up credentials specify in the prompt how to use these tools. 47:06.340 --> 47:13.300 In general, there are over 1000 apps and services you can connect to. 47:14.180 --> 47:20.940 And if there is not a direct integration for what you are looking for, you can always connect via HTTP 47:20.940 --> 47:26.380 requests and API calls in Agentic workflows. 47:26.500 --> 47:34.540 It's not recommended to give your agent too many tools, because the agent may might struggle to decide 47:34.540 --> 47:35.700 which tool to use. 47:36.300 --> 47:40.780 More tools mean more chances for conflicts and failures. 47:41.340 --> 47:46.260 Managing multiple tools in a single agent can slow down responses. 47:46.620 --> 47:54.340 So instead, it's better to build separate subagents, each responsible for a specific task. 47:55.540 --> 48:02.980 Like in this setup, we will create email agent handles, emails and responses. 48:03.460 --> 48:07.060 Calendar agent manages our calendar. 48:09.140 --> 48:13.660 And research agent does research for us. 48:14.500 --> 48:16.380 So this is a basic example. 48:16.660 --> 48:25.100 But what if you wanted your supervisor agent to do more tasks such as managing social media, Making 48:25.100 --> 48:34.100 calls on your behalf or nurturing leads, you would need a swarm of agents, which, by the way. 48:36.180 --> 48:38.820 Can have their own subagents and tools. 48:39.220 --> 48:42.220 And this is what we call a multi-agent system. 48:42.980 --> 48:46.020 Think of it like a multi-layered organization. 48:46.780 --> 48:52.420 So at the top you have an executive agent overseeing everything. 48:52.540 --> 48:59.020 Below that, directors to handle major responsibilities. 48:59.580 --> 49:04.100 And each director has a manager a agent. 49:06.260 --> 49:07.780 Who coordinates tasks. 49:08.660 --> 49:10.060 They have supervisors. 49:11.020 --> 49:18.260 And finally employs agents to execute specific actions. 49:19.660 --> 49:22.500 And each employee has access to tools. 49:22.500 --> 49:22.570 Rules. 49:23.250 --> 49:30.730 So instead of overloading one agent with too many tools, the best practice is to build a network of 49:30.730 --> 49:34.530 specialized agents that communicate and work together. 49:35.210 --> 49:45.890 Each agent focuses on a specific task, making the system more efficient, scalable, and reliable. 49:46.810 --> 49:53.490 By the way, I'm currently working on a system with over 20 agents in N810, so stay tuned. 49:57.090 --> 50:02.130 Now, the only tool you want to add in this setup is a contact database. 50:02.490 --> 50:11.250 So our agent can access email addresses, phone numbers, calendar events, etc. for example, I've 50:11.250 --> 50:15.130 got a spreadsheet with one person's contact details. 50:15.170 --> 50:24.090 For this demo, for simplicity, we'll use Google Sheets, but you could also use CRM like Airtable 50:24.090 --> 50:31.970 for structured data, or pinecone, a vector database for large contact repositories. 50:33.210 --> 50:34.690 So click on add tool. 50:38.410 --> 50:40.170 Search for Google Sheets. 50:44.090 --> 50:45.490 Select Google Sheets tool. 50:52.210 --> 50:54.570 Now you need to connect your Google account. 50:55.130 --> 50:59.530 Once you set up your credentials and connect it to Google account. 50:59.890 --> 51:00.650 Let's move on. 51:01.930 --> 51:05.170 We don't need to change anything within this field. 51:06.650 --> 51:17.090 As a resource, we want to retrieve sheet within document as operation. 51:17.330 --> 51:19.050 Let's leave controls. 51:19.050 --> 51:19.130 Patrols. 51:20.170 --> 51:21.570 And now we can choose. 51:24.170 --> 51:28.090 Specific document we want to retrieve from our Google Sheets. 51:28.810 --> 51:32.650 So from the list let's choose. 51:35.490 --> 51:40.050 And now we can choose the specific sheets. 51:41.690 --> 51:42.730 Leave it as it is. 51:44.330 --> 51:45.210 Let's close it. 51:46.810 --> 51:48.650 And now let's hit save. 51:51.010 --> 51:52.410 And let's test our workflow. 51:54.010 --> 51:57.530 Retrieve the email address of Mark Johnson. 51:58.650 --> 52:00.770 So I'm going to send a voice message. 52:04.090 --> 52:07.090 So we should receive a voice message from our agent. 52:08.730 --> 52:09.770 Let's listen to it. 52:13.770 --> 52:20.890 I found a contact named Mark Johnson with the email Nate at gmail.com, but I couldn't locate specific 52:20.890 --> 52:22.610 information for Marques Johnson. 52:22.610 --> 52:24.890 Let me know if you need help with anything else. 52:25.810 --> 52:26.090 All right. 52:26.090 --> 52:26.610 Great. 52:28.410 --> 52:34.810 Now our agent is able to pull rows of contact information from this sheet. 52:37.970 --> 52:43.530 So to keep things organized, I will rename this tool to Contact Database. 52:57.570 --> 52:59.130 The foundation is now set up. 52:59.370 --> 53:02.050 So let's continue by adding the Subagents. 53:03.730 --> 53:10.130 So now we have to create a completely new workflow and make calls. 53:10.730 --> 53:15.970 So let's open a new tab and go back to The N810. 53:16.010 --> 53:16.730 Home screen. 53:23.850 --> 53:25.130 Make sure to save it. 53:35.850 --> 53:36.850 Click that workflow. 53:43.490 --> 53:45.850 And name it to Email Agent. 53:52.770 --> 53:53.530 Press enter. 53:59.090 --> 54:06.370 So this will act as a dedicated sub agent to handle email related tasks. 54:07.370 --> 54:09.570 So add a trigger node. 54:11.770 --> 54:14.160 Set to when called by another workflow. 54:18.960 --> 54:22.920 This means our personal assistant will call this agent when we ask. 54:24.120 --> 54:27.160 For example, can you send an email or can you check my emails? 54:28.520 --> 54:29.400 Let's close it. 54:31.920 --> 54:32.360 Now. 54:33.160 --> 54:34.880 Add an agent node. 54:49.880 --> 54:54.240 We don't need this trigger, so let's remove it. 54:59.080 --> 55:00.240 Let's add a chat model. 55:03.240 --> 55:06.920 I'm going to use OpenAI GPT four mini. 55:12.080 --> 55:14.440 So no memory is needed for this agent. 55:18.920 --> 55:20.200 Now we need to add tools. 55:22.520 --> 55:23.160 Perfect. 55:23.200 --> 55:25.880 Now we need to add tools to our email agent. 55:26.680 --> 55:31.560 So each tool represents a specific task we want our agent to handle. 55:32.000 --> 55:38.520 So in this example we want it to perform two main actions right. 55:38.520 --> 55:42.480 And send emails and fetch emails from your inbox. 55:43.280 --> 55:44.880 So let's search for Gmail. 55:54.600 --> 55:57.840 In Gmail we have a ton of different actions we can take. 55:58.720 --> 56:03.600 We are going to keep this agent simple, but very useful for functionalities. 56:04.320 --> 56:05.920 So what you can what it can do. 56:06.280 --> 56:08.160 So sending and receiving emails. 56:09.960 --> 56:16.040 Let's rename this tool to send email to make it clear what it does. 56:25.200 --> 56:30.240 I want this tool to set automatically as a resource. 56:30.240 --> 56:32.560 We want to have a message. 56:33.840 --> 56:38.440 We want to send an email, but we need to configure these fields. 56:40.200 --> 56:46.560 Now after I recorded the original version of this tutorial, the Na ten team released this super helpful 56:46.560 --> 56:54.360 feature so you can now use AI to fill in these fields automatically for you. 56:55.680 --> 56:59.320 So first I'm going to show you the simplest way to complete these fields. 56:59.440 --> 57:00.960 So using this new feature. 57:01.960 --> 57:07.200 So just click this little purple I button next to the input. 57:08.120 --> 57:15.640 and the model will intelligently figure out things like the recipient's email address, the subject, 57:15.640 --> 57:19.840 and the message content based on your email context. 57:20.840 --> 57:22.240 So it's really cool. 57:22.240 --> 57:29.280 And it it interprets the user's intent and maps the details directly into the right fields. 57:29.720 --> 57:32.880 So it makes your Gmail node setup way faster. 57:33.760 --> 57:34.880 So let's do it now. 57:34.920 --> 57:38.720 So as you can see we can let the model define the parameter. 57:41.080 --> 57:42.200 So let's click on that. 57:43.440 --> 57:47.440 The same for the subject and for the message. 57:49.280 --> 57:53.400 And the second option is to set up it manually. 57:53.920 --> 58:01.040 So I will show you the manual way how to configure this node using the from OpenAI expression. 58:01.520 --> 58:02.600 So that's useful. 58:02.800 --> 58:07.120 Um, if for some reason the default AI option doesn't work for your case. 58:07.400 --> 58:11.960 Or if you want more control and want to understand how it works behind the scenes. 58:12.560 --> 58:21.360 But both of these methods works not just for Gmail, but for every node you use as a tool for your agents 58:21.360 --> 58:26.760 like sending emails, retrieving inbox content, managing calendars, and so on. 58:27.640 --> 58:36.400 So you can stick with the quick helper or go deeper with full manual setup, and both options are available. 58:37.720 --> 58:39.480 So now let's move on to the second method. 58:39.840 --> 58:43.280 So we are going to use this cool expression. 58:44.760 --> 58:47.560 So we will use the from I function. 58:48.040 --> 58:56.560 This helps the AI extract details from a user's request and map them into fields automatically. 58:57.040 --> 58:58.040 So it's really cool. 58:59.120 --> 59:05.950 So basically this expression helps LLM Alarm interpret an incoming query. 59:06.710 --> 59:12.190 So I'm planning to make a separate tutorial on explaining this in detail, because this is the best 59:12.190 --> 59:16.310 way to give you give you agents tools in any ten. 59:18.190 --> 59:19.870 For now, let's keep it simple. 59:20.790 --> 59:26.190 So let's copy this expression including curly braces. 59:31.390 --> 59:32.870 And paste it to this field. 59:36.510 --> 59:39.430 Then in the field switch over to expression. 59:43.390 --> 59:45.430 Now we need to change the placeholder. 59:47.070 --> 59:50.430 So name it email address. 59:57.710 --> 01:00:01.070 You can also add a description but it's optional. 01:00:01.470 --> 01:00:08.310 So it helps glide the LLM in figuring out how to retrieve the information and in what format. 01:00:09.790 --> 01:00:11.790 So you can add comma. 01:00:19.230 --> 01:00:20.870 And say something like. 01:00:23.190 --> 01:00:24.390 The recipient's. 01:00:27.430 --> 01:00:28.310 Email address. 01:00:33.990 --> 01:00:36.070 We do the same thing with other fields. 01:00:42.350 --> 01:00:43.630 Now the next field. 01:00:45.150 --> 01:00:46.310 So our subject. 01:00:47.070 --> 01:00:48.270 Let's do the same thing. 01:00:56.470 --> 01:00:59.110 Don't forget to switch over to expression. 01:01:14.150 --> 01:01:18.830 Also we need to change the image type to text. 01:01:23.310 --> 01:01:25.510 And for message expression. 01:01:31.310 --> 01:01:34.550 Let's say email body. 01:01:37.190 --> 01:01:42.670 Also we need to add option append and end distribution. 01:01:43.390 --> 01:01:44.270 Turn it off. 01:01:46.750 --> 01:01:48.510 And add sender name. 01:01:50.310 --> 01:01:53.110 This just tells Gmail who the email is from. 01:01:53.510 --> 01:01:54.430 But it won't. 01:01:54.470 --> 01:01:56.790 It won't actually sign off the message. 01:01:57.230 --> 01:01:59.110 We need to do it in our prompt. 01:01:59.750 --> 01:02:03.030 So I will put my name here since it's my personal assistant. 01:02:07.710 --> 01:02:08.870 I will show you in a bit. 01:02:10.950 --> 01:02:13.750 Now we can test our workflow. 01:02:16.270 --> 01:02:21.430 So let's see if the agent can send an email like it's supposed to. 01:02:22.110 --> 01:02:31.430 So since our supervisor agent will be the one calling, this agent will create a sample request to simulate 01:02:31.430 --> 01:02:31.830 that. 01:02:32.550 --> 01:02:35.390 So in the execute workflow trigger. 01:02:37.990 --> 01:02:38.390 Um. 01:02:40.470 --> 01:02:44.910 We need to we need to give a query to our agent. 01:02:45.830 --> 01:02:48.510 So when task two send an email. 01:02:50.190 --> 01:02:58.420 So for example I might say can you send email to Damon at n810 demos Gmail.com to check how he is doing. 01:02:59.140 --> 01:03:03.060 Now let's save it in your data. 01:03:05.900 --> 01:03:06.660 Save it. 01:03:11.980 --> 01:03:13.500 And let's test the workflow. 01:03:19.580 --> 01:03:20.860 There is no problem specified. 01:03:20.860 --> 01:03:23.420 So now let's fix it. 01:03:25.420 --> 01:03:25.780 Oh yeah. 01:03:25.780 --> 01:03:28.860 You have to define below. 01:03:30.180 --> 01:03:34.180 We don't want to take data from the previous node automatically. 01:03:34.220 --> 01:03:36.780 We want to define it below. 01:03:37.700 --> 01:03:42.940 We just simply grab this query and drop it here. 01:03:45.860 --> 01:03:46.500 Save. 01:03:46.660 --> 01:03:48.020 And let's test it again. 01:04:08.820 --> 01:04:10.220 Now let's check my inbox. 01:04:16.060 --> 01:04:16.420 Great. 01:04:16.420 --> 01:04:17.940 So I hope this message finds you well. 01:04:18.420 --> 01:04:18.940 Perfect. 01:04:18.940 --> 01:04:22.740 So now our agent is able to send messages. 01:04:24.020 --> 01:04:27.860 Just a quick tip to get an authorization error. 01:04:28.260 --> 01:04:30.900 Our Gmail credentials might have expired. 01:04:31.340 --> 01:04:34.020 So just refresh the connection and try again. 01:04:34.300 --> 01:04:39.460 So the credentials seem to expire every few days, meaning you just have to sign in again. 01:04:39.980 --> 01:04:46.140 But if you publish your app, you should be good to go without having to refresh it all the time. 01:04:48.660 --> 01:04:56.620 Now, after successful execution, when you check this tool, you can see the actual message. 01:04:58.060 --> 01:05:02.980 So the email I got and the email subject and email address. 01:05:12.460 --> 01:05:22.260 Going back to the email I got, the LM grabbed the recipient, which was Damien, but since we didn't 01:05:22.260 --> 01:05:26.420 specify a sender name, it signed off as your name by default. 01:05:26.820 --> 01:05:33.100 If you actually if you actually want it to sign off properly, you need to prompt the supervisor agent. 01:05:33.860 --> 01:05:42.980 So a quick way to do this is just by is just by going to the supervisor agent in the system message. 01:05:45.580 --> 01:05:51.140 You can say something like when sending emails through Gmail, always sign off with Damian and never 01:05:51.140 --> 01:05:55.660 use placeholders, so this makes sure every email has the correct sign off. 01:05:57.540 --> 01:05:57.900 Great. 01:05:57.900 --> 01:06:00.220 So our agent knows how to send emails. 01:06:00.220 --> 01:06:05.300 Now we need to add another tool to fetch emails from our inbox. 01:06:05.740 --> 01:06:07.740 So let's go back to our email agent. 01:06:11.860 --> 01:06:15.660 Now we want our agent to retrieve emails from our inbox. 01:06:16.260 --> 01:06:21.700 So you can simply use the same tool as before. 01:06:25.820 --> 01:06:28.020 But now you have to configure it differently. 01:06:29.180 --> 01:06:35.260 So you can really expand on this agent and give it tons of different actions like. 01:06:38.620 --> 01:06:39.620 Adding labels. 01:06:39.660 --> 01:06:40.780 Deleting messages. 01:06:41.220 --> 01:06:42.820 Marking emails as unread. 01:06:43.020 --> 01:06:46.660 You could essentially build a tool for all these functions. 01:06:47.220 --> 01:06:49.700 So for now, let's just focus on receiving emails. 01:06:50.020 --> 01:06:52.780 So we need to select get money. 01:06:57.620 --> 01:06:58.860 So first we need to. 01:07:01.180 --> 01:07:04.020 Set a limit on how many emails we are pulling in. 01:07:05.580 --> 01:07:08.300 So for now let's say five. 01:07:09.580 --> 01:07:12.060 We also need to define some filters. 01:07:14.380 --> 01:07:16.780 Like who the emails are coming from. 01:07:18.340 --> 01:07:20.020 So we'll set the sender field. 01:07:21.820 --> 01:07:29.980 As you can see there are multiple ways to filter by date by status, etc. for example, if you only 01:07:29.980 --> 01:07:38.060 wanted to get emails from the past five days, you could configure it by setting received after and 01:07:38.060 --> 01:07:39.500 received before filters. 01:07:40.340 --> 01:07:44.220 For now, let's keep it simple and use just send the filter. 01:07:45.300 --> 01:07:52.850 You can also use from I expression to set a limit and specify the sender. 01:07:53.490 --> 01:07:56.890 So let's use from I and as our sender. 01:07:57.530 --> 01:08:01.490 So I'm just gonna copy and paste this expression again. 01:08:06.090 --> 01:08:07.330 And paste it right here. 01:08:13.050 --> 01:08:15.090 As a placeholder you can enter. 01:08:17.930 --> 01:08:18.890 Send the email. 01:08:25.930 --> 01:08:28.570 And for the limit you have to switch over to expression. 01:08:30.850 --> 01:08:32.210 And paste it right here. 01:08:34.410 --> 01:08:36.170 We can say something like. 01:08:39.050 --> 01:08:39.690 Limit. 01:08:45.010 --> 01:08:45.650 Great. 01:08:47.250 --> 01:08:48.610 And now, one important thing. 01:08:48.610 --> 01:08:51.690 We don't want to simplify the results. 01:08:52.130 --> 01:08:53.810 If this setting is on. 01:08:54.090 --> 01:08:59.730 It returns a simplified version of the email data instead of the full content. 01:09:00.010 --> 01:09:03.730 So in some cases that might that might be fine. 01:09:03.930 --> 01:09:10.930 But if you are dealing with long emails and want all the details, you would want to turn this off. 01:09:12.570 --> 01:09:12.970 All right. 01:09:12.970 --> 01:09:20.330 So now at this point this email agent can send and retrieve emails. 01:09:21.450 --> 01:09:33.210 So what we want to do here is we want this email agent to respond back to our supervisor agent. 01:09:33.450 --> 01:09:35.730 So we need to add another node. 01:09:38.050 --> 01:09:44.250 Set node and simply just name it as a response. 01:09:48.170 --> 01:09:49.850 And grab this output. 01:09:52.290 --> 01:09:56.850 And put here in this field as the value. 01:09:58.570 --> 01:10:00.570 Our email agent is ready to go. 01:10:01.410 --> 01:10:07.450 Just need to unpin data and we can connect it to our supervisor agent. 01:10:08.050 --> 01:10:10.250 So let's go back to our main workflow. 01:10:10.810 --> 01:10:14.690 Now back in our main workflow we are going to add a new tool. 01:10:15.210 --> 01:10:18.890 And for this we'll choose call another workflow node. 01:10:20.570 --> 01:10:23.610 So simply click on Add Tool button. 01:10:26.090 --> 01:10:30.970 We need to choose this option because we want to trigger a different workflow. 01:10:32.010 --> 01:10:33.450 So our email agent. 01:10:37.770 --> 01:10:41.690 We will name this one Email agent. 01:10:53.170 --> 01:10:56.010 Now, as a description, we can say something like. 01:10:58.290 --> 01:11:06.050 Call this tool to take action in email. 01:11:13.090 --> 01:11:15.250 Now we need to configure this node. 01:11:17.370 --> 01:11:21.330 And set up which workflow this tool should connect to. 01:11:24.730 --> 01:11:34.170 So you can select the workflow either from a list or by ID, but picking from the list is the easiest. 01:11:38.570 --> 01:11:50.690 If you can see the workflow which consists your email agent, just simply save and refresh. 01:12:39.760 --> 01:12:44.040 So we called our email agent workflow email agent. 01:12:48.120 --> 01:12:49.480 So let's search for it. 01:12:55.800 --> 01:12:56.200 Now. 01:12:56.240 --> 01:12:57.240 Fit to return. 01:12:58.360 --> 01:13:01.440 This is where we previously set up the response field. 01:13:01.680 --> 01:13:10.280 So when the supervisor agent sends a request to the email, agent is going to look for a field named 01:13:10.280 --> 01:13:11.080 response. 01:13:11.800 --> 01:13:17.840 That's why in the email agent workflow I'm going to show you. 01:13:22.400 --> 01:13:34.880 That's why in this agent we added a set node named the field response and assign it to the actual email 01:13:34.880 --> 01:13:35.480 image data. 01:13:36.960 --> 01:13:41.960 So the supervisor agent knows exactly where to look for the response. 01:13:43.560 --> 01:13:50.560 Now we are going to create our calendar sub agent and I will move a bit quicker with this, since the 01:13:50.560 --> 01:13:54.560 process is pretty much the same as what we did with the email agent. 01:13:55.360 --> 01:14:00.680 So the first thing we need to do, we have to create a completely new workflow. 01:14:02.320 --> 01:14:04.520 So let's go back to the home page. 01:14:07.240 --> 01:14:11.240 So let's trigger this workflow when called by another workflow. 01:14:14.400 --> 01:14:16.240 So when called by another workflow. 01:14:20.880 --> 01:14:22.360 We'll add the agent. 01:14:23.040 --> 01:14:25.640 So let's search for a agent. 01:14:33.520 --> 01:14:34.920 Now as the chat model. 01:14:34.920 --> 01:14:42.760 So our agents brain since GPT four mini worked well for the email agent, we will stick with it again. 01:14:45.600 --> 01:14:47.200 We can remove this one. 01:14:47.200 --> 01:14:49.560 Chat message received trigger note. 01:14:52.880 --> 01:14:53.960 Add chat model. 01:14:57.320 --> 01:14:58.920 Select chat model. 01:15:11.960 --> 01:15:15.320 We don't need memory for this agent so we will leave that off. 01:15:15.680 --> 01:15:17.600 Now we can start adding the tools. 01:15:18.400 --> 01:15:20.080 This is our calendar agent. 01:15:20.080 --> 01:15:23.720 So we want to create and get events from our calendar right. 01:15:24.160 --> 01:15:27.160 So search for Google Calendar. 01:15:32.320 --> 01:15:34.160 So other counties already connected. 01:15:34.160 --> 01:15:35.200 So we are good to go. 01:15:35.480 --> 01:15:41.680 If you are using the cloud service, you might not need to enter the client ID and secret to set up 01:15:41.680 --> 01:15:42.480 credentials. 01:15:45.520 --> 01:15:48.240 But for the rest of us, we already done the hard part. 01:15:48.640 --> 01:15:52.080 So the first tool we are going to add is to create an event. 01:15:52.520 --> 01:15:54.560 So let's rename it. 01:16:03.200 --> 01:16:05.320 Will connect it to our calendar. 01:16:05.320 --> 01:16:08.200 And now we just need to configure a few details. 01:16:10.800 --> 01:16:15.640 The leaf resource as an event we want to create an email. 01:16:16.880 --> 01:16:19.080 Select your email address from the list. 01:16:22.120 --> 01:16:29.480 Now we have to specify the start, start time and end time and any additional properties like attendees. 01:16:30.360 --> 01:16:35.720 We will also add a summary, which is basically the event's title. 01:16:36.000 --> 01:16:38.240 So the text that shows up on the calendar. 01:16:43.640 --> 01:16:45.600 So let's do it one by one. 01:16:47.840 --> 01:16:57.920 We will pull the start time using from AI expression once again to let the large language model to do 01:16:57.920 --> 01:16:59.960 it automatically. 01:17:01.960 --> 01:17:11.120 So copy this from AI expression from curly braces all the way to the closing curly braces. 01:17:14.520 --> 01:17:16.680 Remember to switch to the expression. 01:17:19.080 --> 01:17:20.320 And paste it right here. 01:17:23.320 --> 01:17:26.880 Change the placeholder name to start time. 01:17:31.510 --> 01:17:33.990 You can add the description if you want. 01:17:36.350 --> 01:17:37.350 So it's the time. 01:17:40.950 --> 01:17:46.270 The user asked for the event. 01:17:49.350 --> 01:17:50.270 To start. 01:17:54.230 --> 01:17:54.910 Great. 01:17:56.870 --> 01:18:01.390 Now we have to do exactly the same thing for the end time. 01:18:02.070 --> 01:18:03.910 So let's just copy and paste it. 01:18:04.550 --> 01:18:08.190 Now let's change the placeholder name to End time. 01:18:12.150 --> 01:18:13.510 And the description. 01:18:18.390 --> 01:18:19.110 That's great. 01:18:24.350 --> 01:18:26.270 Now we have to add summary. 01:18:35.110 --> 01:18:36.190 For the summary. 01:18:36.230 --> 01:18:41.710 We will also pull that from I expression. 01:18:43.630 --> 01:18:45.190 So let's copy it. 01:18:49.590 --> 01:18:54.310 And just change the placeholder and the description. 01:18:55.510 --> 01:18:58.150 So let's say simply name. 01:19:04.830 --> 01:19:07.190 And it's the name of the calendar event. 01:19:17.430 --> 01:19:18.150 Great. 01:19:19.750 --> 01:19:28.110 Next we will add a sample query in the Execute workflow trigger node so we can play around with the 01:19:28.110 --> 01:19:28.710 data. 01:19:33.350 --> 01:19:39.270 Let's say we ask the calendar agent to set up an event at 3:00 pm for a meeting. 01:19:53.830 --> 01:19:56.270 So here we have to provide the prompt. 01:19:58.550 --> 01:20:03.550 Set up and event. 01:20:07.190 --> 01:20:08.310 At 3 p.m.. 01:20:10.350 --> 01:20:11.070 Meeting. 01:20:15.190 --> 01:20:15.790 Great. 01:20:19.750 --> 01:20:20.870 Now let's save it. 01:20:22.630 --> 01:20:23.710 The data is pinned. 01:20:27.590 --> 01:20:31.750 In this case, we are just blocking out time in our supervisor agent. 01:20:32.030 --> 01:20:35.710 But since we didn't specify an end time in the query. 01:20:35.990 --> 01:20:38.470 So how long is the meeting for? 01:20:39.750 --> 01:20:49.030 Later, we modified the agents prompt to assume the event lasts for 30 minutes or an hour. 01:20:49.030 --> 01:20:54.630 After the start time of the meeting in our supervisor agents prompt. 01:20:56.750 --> 01:20:58.190 Now let's save the workflow. 01:21:02.510 --> 01:21:03.390 And test it. 01:21:10.790 --> 01:21:16.870 Yes, we got an error because we didn't specify the prompt, so let's do it now. 01:21:21.510 --> 01:21:22.750 So define below. 01:21:24.950 --> 01:21:27.670 And simply drag and drop the query property. 01:21:31.230 --> 01:21:37.830 Now we will update the agent's prompt to include the current date and time so it knows when to create 01:21:37.830 --> 01:21:38.470 events. 01:21:38.790 --> 01:21:46.190 Because while these agents are pretty good at figuring out which tools to call, adding system prompts 01:21:46.190 --> 01:21:51.030 like this can make a big difference, especially depending on the model you are using. 01:21:51.830 --> 01:21:52.830 So let's do it now. 01:21:55.430 --> 01:22:04.670 Click on add option, Add system message and we'll keep the prompt short and to the point since it's 01:22:04.670 --> 01:22:06.350 just a calendar agent. 01:22:12.430 --> 01:22:18.220 You are a calendar assistant with access to various tools to manage Calendar tasks. 01:22:18.260 --> 01:22:20.380 The current date and time is. 01:22:20.980 --> 01:22:32.500 I forgot to switch over to the expression mode, so thanks to this, agent is able to figure out what 01:22:32.500 --> 01:22:34.740 the current date and time is. 01:22:36.300 --> 01:22:37.980 So create event. 01:22:40.820 --> 01:22:45.180 Create event with attendee and also get events. 01:22:48.700 --> 01:22:49.860 Okay now let's test it. 01:22:52.380 --> 01:22:53.100 Hit save. 01:22:58.740 --> 01:22:59.820 And this workflow. 01:23:06.380 --> 01:23:06.940 Great. 01:23:08.420 --> 01:23:10.620 Now let's look at the execution. 01:23:31.980 --> 01:23:34.300 As you can see, it's working perfectly fine. 01:23:41.740 --> 01:23:45.140 Now let's add another tool to create events with attendees. 01:23:45.460 --> 01:23:52.940 We are doing this in a separate tool because adding attendees requires extra configuration. 01:23:53.540 --> 01:23:56.220 We need to add additional fields. 01:23:56.700 --> 01:24:00.420 If we left it all in one tool, it could cause issues. 01:24:00.820 --> 01:24:02.220 If there is no attendee. 01:24:02.380 --> 01:24:06.380 It might return an error when trying to create the event. 01:24:07.140 --> 01:24:08.820 So let's make it easy. 01:24:08.820 --> 01:24:13.380 So we need just to copy our create Eight. 01:24:14.020 --> 01:24:14.460 Events. 01:24:14.500 --> 01:24:14.980 Note. 01:24:20.980 --> 01:24:22.060 And paste it here. 01:24:24.940 --> 01:24:26.220 Let's rename it to. 01:24:32.980 --> 01:24:34.580 Event with attendee. 01:24:46.500 --> 01:24:48.460 So everything is exactly the same. 01:24:48.460 --> 01:24:50.500 We just need to specify attendee. 01:24:52.140 --> 01:24:53.180 So scroll down. 01:24:56.100 --> 01:24:58.820 Add filled attendees. 01:25:03.100 --> 01:25:05.220 Let's use our awesome expression. 01:25:09.300 --> 01:25:18.980 So the placeholder is attendee email and the description is the email of the user asks the event to 01:25:18.980 --> 01:25:20.060 be scheduled with. 01:25:21.020 --> 01:25:23.980 Now that we've got event creation working. 01:25:26.620 --> 01:25:29.740 Let's add one more tool to get events from the calendar. 01:25:30.620 --> 01:25:35.860 So this will allow the agent to retrieve events based on specific dates. 01:25:37.300 --> 01:25:42.340 Let's just actually let's rename this agent to Calendar agent. 01:25:53.780 --> 01:25:55.860 And also our workflow to. 01:25:58.620 --> 01:25:59.740 Calendar agent. 01:26:03.860 --> 01:26:04.660 And hit save. 01:26:16.770 --> 01:26:21.010 And this node to create event. 01:26:29.770 --> 01:26:33.410 Let's search for Google Calendar again. 01:26:39.010 --> 01:26:44.490 This time we want to retrieve emails from our inbox. 01:26:45.170 --> 01:26:47.690 So change operation to get money. 01:26:50.170 --> 01:26:51.530 Select your email address. 01:26:56.770 --> 01:26:58.930 We want all the emails to be returned. 01:26:59.130 --> 01:27:00.450 So enable this. 01:27:09.050 --> 01:27:13.490 We want to get events from one day before the requested date. 01:27:14.810 --> 01:27:18.210 So at option select after. 01:27:20.410 --> 01:27:25.170 Which the expression we will use from I expression again. 01:27:27.690 --> 01:27:33.370 As a placeholder we have start date and as the description the day before the requested date. 01:27:35.930 --> 01:27:38.930 We have to do the same for before field. 01:27:45.290 --> 01:27:50.530 But this time we want to get events up to one day after the requested date. 01:27:52.130 --> 01:27:54.850 So switch over to expression mode. 01:27:59.010 --> 01:28:04.090 And date as a placeholder and as a description the day after the requested date. 01:28:05.170 --> 01:28:12.050 This helps us retrieve events dynamically based on the user's input, instead of just pulling events 01:28:12.050 --> 01:28:14.730 from a fixed date like today. 01:28:16.370 --> 01:28:22.850 Finally, we need to ensure that the calendar agent sends responses back to the supervisor agent. 01:28:37.330 --> 01:28:40.930 Let's rename it to get events. 01:28:50.650 --> 01:29:01.210 We will add a set node as in the previous example in the email agent to search for set node. 01:29:05.730 --> 01:29:07.650 We need to add a response field. 01:29:08.930 --> 01:29:12.810 So the name is response again. 01:29:15.090 --> 01:29:16.330 Grab the output. 01:29:18.410 --> 01:29:19.410 Drop over here. 01:29:21.730 --> 01:29:22.370 Great job. 01:29:22.410 --> 01:29:27.050 Now let's connect this calendar agent to our supervisor agent. 01:29:27.610 --> 01:29:28.490 So hit save. 01:29:32.250 --> 01:29:33.050 Adding a tool. 01:29:35.490 --> 01:29:37.570 Select called Workflow tool. 01:29:41.370 --> 01:29:43.330 Rename it to Calendar agent. 01:29:58.530 --> 01:30:00.010 We need to provide a description. 01:30:00.010 --> 01:30:03.330 So They still. 01:30:07.530 --> 01:30:08.690 Manage the calendar. 01:30:13.370 --> 01:30:14.250 Great stuff. 01:30:15.890 --> 01:30:21.010 Now we are going to connect the calendar agent workflow we just created. 01:30:23.890 --> 01:30:26.490 We have to refresh this workflow. 01:30:29.450 --> 01:30:34.210 Because we didn't refresh it since we have created our calendar agent. 01:30:34.650 --> 01:30:35.530 So hit save. 01:30:37.290 --> 01:30:40.170 Now our calendar agent appears in the list. 01:30:41.650 --> 01:30:42.610 So select it. 01:30:45.170 --> 01:30:50.090 And the response field will handle sending data back to the supervisor agent. 01:30:53.410 --> 01:30:54.010 Great. 01:30:56.050 --> 01:30:56.930 Let's save it. 01:31:00.800 --> 01:31:01.440 Great. 01:31:01.480 --> 01:31:08.840 Now we are setting up a research agent that can answer questions by searching the internet, checking 01:31:08.840 --> 01:31:14.400 articles, and using sources like Wikipedia and set API. 01:31:15.000 --> 01:31:22.440 It's a lot easier to set up this agent than the email and calendar agents, but still very useful. 01:31:23.520 --> 01:31:29.880 So the goal here is to show you how you can assign different tools to your agents, making them more 01:31:29.880 --> 01:31:34.520 flexible and reusable across multiple workflows. 01:31:35.520 --> 01:31:39.400 So it's exactly the same process we need to create completely new workflow. 01:31:41.560 --> 01:31:46.200 So let's go back to our home page at First Step. 01:31:49.800 --> 01:31:55.320 So we want to trigger this workflow when it's called by another workflow. 01:32:01.240 --> 01:32:02.280 Other agents. 01:32:12.520 --> 01:32:13.600 Remove the trigger. 01:32:20.880 --> 01:32:22.640 Large language model as a chat model. 01:32:24.880 --> 01:32:25.840 To do the same. 01:32:27.920 --> 01:32:29.200 GPT four mini. 01:32:32.400 --> 01:32:33.560 We don't need a memory. 01:32:35.320 --> 01:32:38.720 Then we will add tools starting with Wikipedia. 01:32:39.840 --> 01:32:40.720 So add tool. 01:32:42.160 --> 01:32:43.400 Search for Wikipedia. 01:32:50.560 --> 01:32:53.000 Let's add the prompt to our agent. 01:32:56.520 --> 01:32:59.200 So adoption system message. 01:33:02.080 --> 01:33:08.600 The system message I have prepared is you are a research assistant with access to Wikipedia and serve 01:33:08.640 --> 01:33:09.080 API. 01:33:09.520 --> 01:33:15.560 First search for search Wikipedia and if you don't find the answer, use search API. 01:33:18.760 --> 01:33:26.760 So this setup is efficient because it uses resources first and only calls setup API when needed. 01:33:26.960 --> 01:33:28.720 Saving API credits. 01:33:29.600 --> 01:33:34.760 Setup API offers a free plan that includes 100 searches per month. 01:33:36.000 --> 01:33:48.480 So basically you can use this research agent 100 times per month in order to connect API ad to search 01:33:48.480 --> 01:33:49.680 for Serp API. 01:33:53.160 --> 01:33:56.760 Now we have to connect credentials. 01:33:57.640 --> 01:33:59.480 So go to servicom. 01:34:01.000 --> 01:34:02.120 Create an account. 01:34:03.080 --> 01:34:04.480 Go to API key. 01:34:09.080 --> 01:34:10.600 Create your API key. 01:34:10.920 --> 01:34:13.960 Copy it and paste it into an A-10. 01:34:15.720 --> 01:34:17.480 So we need to create new credential. 01:34:20.480 --> 01:34:21.760 And paste it right here. 01:34:23.320 --> 01:34:24.360 Then click save. 01:34:25.000 --> 01:34:27.360 And it should show as successful. 01:34:30.400 --> 01:34:31.680 Then save workflow. 01:34:36.120 --> 01:34:36.600 All right. 01:34:36.640 --> 01:34:39.240 Now let's connect this agent to our supervisor. 01:34:39.240 --> 01:34:39.760 Agent. 01:34:43.960 --> 01:34:45.520 First let's rename it to. 01:34:48.080 --> 01:34:49.160 Research agent. 01:34:52.800 --> 01:34:56.760 And also our subagent to make it clear. 01:35:10.960 --> 01:35:11.560 Great. 01:35:13.600 --> 01:35:14.240 Save it. 01:35:16.920 --> 01:35:19.240 Let's go back to our main workflow. 01:35:26.480 --> 01:35:30.240 Let's move this tool down. 01:35:35.920 --> 01:35:36.760 Adding a tool. 01:35:39.560 --> 01:35:41.120 Call a workflow tool. 01:35:45.840 --> 01:35:49.520 Rename it to please search agent. 01:36:08.470 --> 01:36:14.510 I will call this tool to search the internet and answer the user's questions. 01:36:17.070 --> 01:36:20.270 So simply call tool. 01:36:23.070 --> 01:36:25.590 To search the internet. 01:36:29.070 --> 01:36:30.070 And answer. 01:36:34.670 --> 01:36:35.870 Their users questions. 01:36:52.270 --> 01:36:55.470 Just select the research agent from the list. 01:37:02.910 --> 01:37:03.670 And that's it. 01:37:05.830 --> 01:37:06.550 Hit save. 01:37:14.990 --> 01:37:16.150 Congratulations. 01:37:16.310 --> 01:37:23.150 You just built your team of AI agents and you have learned how to create subagents like the email agent, 01:37:23.150 --> 01:37:31.470 calendar agent, and the research agent all working together under the supervision of the main supervisor 01:37:31.470 --> 01:37:32.070 agent. 01:37:32.550 --> 01:37:40.270 You can easily add more sub agents to handle new tasks, and if you want your agents to remember things 01:37:40.270 --> 01:37:46.190 over time, you can connect a vector database to give them persistent memory. 01:37:46.790 --> 01:37:47.790 So great job!