WEBVTT 00:00.360 --> 00:03.720 Today I'm going to show you how to connect Supabase. 00:06.080 --> 00:07.200 And PostgreSQL. 00:08.440 --> 00:09.280 2010. 00:09.760 --> 00:16.680 If you are building a agents in 2010, especially RAC agents, and want to give them a long term memory, 00:17.080 --> 00:24.000 you can't use the simple memory node because it only stores short term memory for the current session. 00:24.320 --> 00:27.840 So once the conversation ends, everything is lost. 00:28.240 --> 00:36.040 Instead, you are going to need two essential things a memory so your agent can remember conversations 00:36.600 --> 00:40.880 and a vector database for RAC functionality. 00:41.280 --> 00:46.840 So today I'm going to show you how to connect both PostgreSQL and Supabase 2010. 00:47.520 --> 00:51.600 Before we start building, let me explain what we are working with. 00:51.960 --> 01:00.320 PostgreSQL is this powerful open source database that we are going to use to store our agents conversation 01:00.320 --> 01:00.920 memory. 01:01.600 --> 01:09.600 Think of it like giving a agent a brain that actually remembers what you talk about. 01:09.600 --> 01:20.120 Yesterday by Supabase is built on top of PostgreSQL, but gives us this clean interface and vector capabilities 01:20.120 --> 01:24.440 that make it perfect for storing document embeddings. 01:24.680 --> 01:31.720 So by the end of this tutorial, you will have fully functional AI agent that can remember your conversations 01:31.960 --> 01:38.120 thanks to PostgreSQL and pull relevant information from your documents. 01:38.560 --> 01:40.360 Thanks to Supabase vector store. 01:40.760 --> 01:43.040 So here is what we are going to accomplish today. 01:43.320 --> 01:50.480 First, we sign up for Sparkbase account, which is completely free by the way, and you can also self-host 01:50.520 --> 01:50.840 it. 01:51.480 --> 01:55.600 Then we will create a new project and grab all the connection details we need. 01:56.360 --> 02:02.760 The tricky part is getting all those connection strings because there are quite a few different URLs 02:02.760 --> 02:04.080 and credentials involved. 02:04.520 --> 02:06.960 I will walk you through each one step by step. 02:08.360 --> 02:13.680 Also, we will need to add credentials for our Postgres chat memory. 02:17.120 --> 02:23.200 And this is where most people get stuck because the connection details can be confusing. 02:24.320 --> 02:30.960 So we need to connect and configure things like host, username, password, port, all that connection 02:30.960 --> 02:31.440 stuff. 02:32.840 --> 02:39.680 And the reason we need both is simple because the PostgreSQL memory node stores our conversation history. 02:39.920 --> 02:40.640 So the end. 02:40.680 --> 02:43.600 So the agent remembers what we talk about. 02:43.880 --> 02:52.000 While the vector store lets us upload documents that we have the agents search through them intelligently. 02:52.840 --> 02:53.960 Let's get started. 02:54.200 --> 02:59.040 First I'm going to add a chat message trigger node. 03:03.870 --> 03:06.350 Now let's add another agents. 03:10.150 --> 03:11.590 To search for an agent. 03:14.990 --> 03:19.190 Let's rename this guy to Rack agent. 03:25.310 --> 03:30.230 All right, now let's connect it to some large language model. 03:30.430 --> 03:36.230 I'm going to use OpenAI's GPT four mini. 03:50.470 --> 03:50.830 All right. 03:50.830 --> 03:51.270 Great. 03:53.710 --> 03:57.430 Now I'm going to create a fresh super base account so you can follow along. 03:57.470 --> 03:58.150 Exactly. 03:59.350 --> 04:02.590 So head over to Supabase Superpages.com. 04:04.950 --> 04:06.510 Click on starting project. 04:09.830 --> 04:10.990 And click sign up. 04:15.750 --> 04:17.110 Now create your account. 04:17.670 --> 04:19.230 So provide email and password. 04:20.470 --> 04:25.710 Now once you sign up you should get this email to confirm your email address. 04:26.550 --> 04:27.830 So just simply confirm. 04:28.950 --> 04:32.470 Now once you confirm your email you will need to create an organization. 04:33.310 --> 04:38.110 Just leave it as personal for now since we are using the free plan. 04:38.710 --> 04:41.230 So type personal free plan. 04:41.990 --> 04:43.470 Click on Create organization. 04:46.150 --> 04:51.150 Now when you create your project you will need to set a database password. 04:51.510 --> 04:52.710 So this is super important. 04:52.710 --> 04:59.030 Make sure you save this password somewhere safe because we are going to need it later when we connect 04:59.030 --> 05:00.150 to PostgreSQL. 05:03.910 --> 05:05.150 So type your password. 05:10.150 --> 05:13.870 I'm going to copy mine and paste it into my notes. 05:14.590 --> 05:21.230 So this password is specifically for accessing the PostgreSQL database that Supabase creates for you. 05:22.230 --> 05:23.590 Now let's create the project. 05:23.590 --> 05:24.870 So creating project. 05:25.710 --> 05:28.230 So this is going to take a few minutes to set up. 05:28.430 --> 05:29.950 So we just need to wait. 05:32.150 --> 05:33.950 Here you can see the project status. 05:36.150 --> 05:37.790 So it is doing behind the scenes. 05:37.790 --> 05:41.470 It's setting up the PostgreSQL instance. 05:42.150 --> 05:49.470 Now once your project is set up you need to go to Project Settings and click on it from the left sidebar. 05:52.190 --> 05:55.230 Then under configuration click on database. 05:58.870 --> 06:01.190 And now you can see connection string has moved. 06:01.350 --> 06:04.830 And you can find Project Connect details by clicking connect in the top bar. 06:05.190 --> 06:08.630 So you need to click on that connect button in that in the top. 06:13.710 --> 06:17.830 And you will see this dashboard with a bunch of different API keys and URLs. 06:17.990 --> 06:22.550 And this is where people usually get confused because there is so much information here. 06:23.070 --> 06:26.910 But we only need specific pieces for our PostgreSQL connection. 06:27.550 --> 06:29.910 So now let's go back to our workflow. 06:30.470 --> 06:36.030 Now as memory for our rack agent let's add Postgres memory. 06:39.190 --> 06:40.670 Let's create new credential. 06:43.550 --> 06:48.230 And as you can see we need to provide host database user and password. 06:49.070 --> 06:52.030 So now I'm going to show you where you can find these credentials. 06:52.550 --> 06:57.470 Now we need to change the connection type to PostgreSQL. 07:00.670 --> 07:04.500 And then scroll down To the transaction puller. 07:05.140 --> 07:09.900 So this is where we'll find our host username password and port. 07:10.180 --> 07:13.340 So the host tells N810 where to find your database. 07:13.540 --> 07:16.900 The username and password authenticate us. 07:16.940 --> 07:21.500 And the port specifies which connection method to use for the host. 07:21.500 --> 07:25.260 We are looking for everything after dash h. 07:27.300 --> 07:33.060 Which obviously stands for host up to dash P. 07:33.940 --> 07:35.060 So let's copy it. 07:37.620 --> 07:43.260 Let's go back to our editing workflow and paste it into host field. 07:44.100 --> 07:47.300 Next we need the database and username. 07:47.540 --> 07:49.580 So let's go to our connection string. 07:51.700 --> 08:00.220 Here you can see that we have a dash for D which is our database Postgres. 08:00.220 --> 08:01.140 So let's copy it. 08:05.620 --> 08:08.380 Is the same as default and for username. 08:14.740 --> 08:16.980 We have everything after you. 08:17.020 --> 08:20.620 So Postgres dot and some random characters. 08:21.620 --> 08:22.700 So let's copy it. 08:25.100 --> 08:27.620 And paste it into the user field. 08:27.900 --> 08:30.100 So the database stays Postgres. 08:30.500 --> 08:32.820 That's just the default database name. 08:33.500 --> 08:34.660 Now for the password. 08:35.140 --> 08:39.580 This is where you use that password you created when setting up the project. 08:39.740 --> 08:42.580 So this authenticates you as the database owner. 08:43.340 --> 08:44.420 Now let's scroll down. 08:49.140 --> 08:50.660 And let's find the port number. 08:52.460 --> 08:54.460 So go back to our connection string. 08:58.220 --> 09:01.500 And here after dash P you can see the port number. 09:01.820 --> 09:02.700 Let's copy it. 09:06.420 --> 09:07.900 And paste into this field. 09:11.020 --> 09:15.060 And you should see that green check mark is telling us everything is working. 09:15.420 --> 09:20.620 So now I'm going to rename this credentials Something tutorial. 09:24.500 --> 09:24.900 All right. 09:24.900 --> 09:25.380 Great. 09:26.820 --> 09:27.580 Perfect. 09:27.820 --> 09:29.700 Now our chat memory is connected. 09:30.500 --> 09:31.380 You can see. 09:33.820 --> 09:39.500 That is going to use our connected chat trigger node as the session key. 09:39.980 --> 09:43.460 And it will store everything in a table in Supabase. 09:43.860 --> 09:49.420 So the session key is important because it keeps different conversations separate. 09:49.940 --> 09:52.980 Now let me quickly show you by testing this connection. 09:53.540 --> 09:54.780 So let's open the chat. 10:00.140 --> 10:03.740 And first actually let's go to our database. 10:03.740 --> 10:03.820 base. 10:04.420 --> 10:08.340 And when you go to table editor. 10:11.140 --> 10:13.300 You can see that we don't have any tables yet. 10:14.300 --> 10:16.940 But when I start the conversation with our agent. 10:22.820 --> 10:24.380 I want to send a simple message. 10:24.380 --> 10:24.860 Hello. 10:26.260 --> 10:32.900 And now let me check the table editor and refresh it. 10:37.740 --> 10:39.900 We should see this conversation start. 10:40.420 --> 10:40.700 Yes. 10:40.700 --> 10:41.220 Here we go. 10:41.220 --> 10:47.100 So we have a new table called N810 Chat Histories with two messages. 10:47.100 --> 10:50.180 When I click on it we have my message. 10:50.380 --> 10:51.100 So hello. 10:52.900 --> 10:54.340 And the agent's reply. 10:56.900 --> 10:58.620 So hello how can I assist you today? 10:59.060 --> 10:59.700 Perfect. 10:59.940 --> 11:05.340 Yes, this is how all your conversations get stored based on session Session ID? 11:06.300 --> 11:09.180 So the session ID comes from our chat trigger node. 11:10.260 --> 11:13.780 So each conversation gets its own unique ID. 11:13.820 --> 11:18.540 This means if you have multiple users the conversations stay separate. 11:18.820 --> 11:20.660 Now lets add vector style. 11:20.980 --> 11:23.700 So for this part we need to create another credential. 11:23.900 --> 11:26.860 But this time it's for Spa based API. 11:27.540 --> 11:28.660 So lets add tool. 11:31.100 --> 11:34.780 For Supabase select Supabase vector store. 11:37.540 --> 11:38.740 Create credential. 11:40.420 --> 11:44.980 And this is where people get confused because we need two different sets of credentials. 11:44.980 --> 11:47.580 Because we are accessing Supabase in two different ways. 11:47.980 --> 11:50.020 So go back to your Supabase dashboard. 11:53.900 --> 11:55.780 Go to Project settings. 12:00.420 --> 12:01.340 Data API. 12:04.850 --> 12:06.970 And from here you can grab your URL. 12:07.450 --> 12:08.330 So copy it. 12:11.930 --> 12:13.930 And paste it into the host field. 12:17.970 --> 12:20.010 Now you need to provide service raw secret. 12:25.050 --> 12:26.370 Go to API keys. 12:30.970 --> 12:32.570 And reveal its service role. 12:34.250 --> 12:35.050 Copy it. 12:37.290 --> 12:38.770 And paste it into this field. 12:39.490 --> 12:42.730 So the project goes into the host field. 12:43.210 --> 12:45.450 And the service roll secret is that long token. 12:45.930 --> 12:52.050 And the reason we need different credentials is because the vector store communicates through Supabase 12:52.090 --> 12:59.210 rest API, while the PostgreSQL memory connects directly to the database. 12:59.330 --> 13:01.010 So there are two different protocols. 13:01.570 --> 13:02.530 Now let's save it. 13:04.330 --> 13:05.130 Excellent. 13:06.090 --> 13:08.250 Now I'll rename this credential as well. 13:18.450 --> 13:18.810 All right. 13:18.810 --> 13:19.170 Cool. 13:19.970 --> 13:25.610 Now before we can actually use the vector store we need to set up the documents table in Supabase. 13:25.930 --> 13:31.210 So this is a crucial step that allows for vector similarity search. 13:31.770 --> 13:32.810 So in Supabase. 13:38.610 --> 13:39.610 Go to docs. 13:40.330 --> 13:47.090 Here in the documentation you can find the exact command we need to copy and paste to the SQL editor 13:47.090 --> 13:47.890 in Starbase. 13:48.410 --> 13:53.410 So scroll down, click on Quickstart for setting up your vector store. 13:59.570 --> 14:01.330 Copy this SQL code. 14:01.330 --> 14:02.650 Don't change anything here. 14:05.410 --> 14:06.490 Simply copy it. 14:08.250 --> 14:10.050 Go to the dashboard. 14:13.970 --> 14:17.530 In the left sidebar, find SQL editor. 14:18.450 --> 14:20.330 Click on new SQL snippet. 14:24.010 --> 14:29.650 And here you are going to run this command to set up our documents table with Vector Store. 14:30.410 --> 14:31.650 So paste the command here. 14:34.410 --> 14:35.250 Click on run. 14:38.850 --> 14:40.410 And you should see success. 14:40.610 --> 14:41.690 No rows returned. 14:42.490 --> 14:49.850 And then when I go to table editor we have a new table called documents. 14:50.330 --> 14:54.410 So now let me show you how this works by adding a document to our vector store. 14:55.210 --> 14:58.450 I'm going to use a Google Drive node to grab the documents. 14:59.050 --> 15:02.370 So let's add Google Drive node. 15:09.770 --> 15:10.850 Let's do not file. 15:12.250 --> 15:14.010 And from list I'm going to choose. 15:17.490 --> 15:21.730 A voice agent prompt framework migrate. 15:22.770 --> 15:24.170 Now when I test this node. 15:29.130 --> 15:30.330 To get the binary data. 15:30.330 --> 15:32.610 So this document will become our knowledge base. 15:33.530 --> 15:36.890 So now I will connect this to our supervised vector node. 15:37.730 --> 15:39.290 So let's search for vector store. 15:44.370 --> 15:45.010 Vector store. 15:45.970 --> 15:47.930 We are going to add documents to vector store. 15:51.090 --> 15:52.250 For the credentials. 15:52.610 --> 15:56.850 I'm going to choose the credential I've just made the operation mode. 15:56.850 --> 16:03.570 We have insert documents and you have to choose the table we want to push the data to. 16:04.610 --> 16:09.250 So the table you just created in our super base vector store called documents. 16:09.890 --> 16:11.170 So let's select documents. 16:12.810 --> 16:13.050 Right. 16:13.050 --> 16:13.610 Perfect. 16:14.290 --> 16:16.730 Now let's add the document loader. 16:17.850 --> 16:22.250 Now I need to configure the document loader and embeddings to. 16:22.930 --> 16:24.250 So the document loader. 16:26.290 --> 16:30.170 Will be default data loader to load data from previous step in the workflow. 16:32.490 --> 16:35.970 So basically the document loader breaks the file into chunks. 16:38.570 --> 16:41.530 Let's change the type of data to binary. 16:43.250 --> 16:44.050 All right great. 16:44.930 --> 16:46.050 And embeddings model. 16:49.730 --> 16:51.890 We're going to use OpenAI models. 16:54.730 --> 16:56.130 Text embedding XML. 16:56.130 --> 16:57.050 So default model. 16:59.250 --> 17:06.090 And basically this model converts each chunk into vectors that the large language model can search through. 17:07.880 --> 17:08.280 All right. 17:08.280 --> 17:08.680 Cool. 17:11.600 --> 17:18.840 And now, when I test this step, it processes the document and splits it into chunks. 17:20.840 --> 17:26.000 Now, as you can see, it created 13 chunks from our document. 17:27.000 --> 17:28.720 Now back in Supabase. 17:31.640 --> 17:37.760 We now have 13 vectorized chunks of our document with embeddings and metadata. 17:38.200 --> 17:42.680 So the vectors are those long arrays of numbers. 17:43.000 --> 17:48.800 That's how the AI so large language models understands and searches the content. 17:50.400 --> 17:51.680 We also have metadata. 17:53.600 --> 17:57.760 And here we've got access to all the content splitted into chunks. 18:00.600 --> 18:03.320 So as you can see I can read through each of them. 18:06.840 --> 18:07.440 Perfect. 18:08.080 --> 18:17.320 Now, regarding metadata, we've got stuff like source and blob type, all this kind of information. 18:17.320 --> 18:22.280 So this is where we could also add more metadata if we wanted to. 18:22.520 --> 18:27.880 Things like document categories, dates or any other information that might help with searching. 18:28.560 --> 18:34.680 So the metadata is really useful because it helps the AI understand context about each document chunk. 18:35.280 --> 18:41.000 So when someone asks a question, the AI doesn't just look at the content. 18:41.720 --> 18:46.840 It can also consider whether that content came from and what type of document it is. 18:48.280 --> 18:56.920 So all this content comes from this file I store in my Google Drive and it has been vectorized. 18:57.360 --> 19:01.440 Split it into chunks and added to the documents table in Supabase. 19:03.040 --> 19:03.760 All right. 19:04.080 --> 19:10.960 Now that we successfully vectorized the file and store the data in our super base table. 19:11.280 --> 19:14.160 It's time to connect our agent to this vector database. 19:14.640 --> 19:16.200 So in Superbase vector store. 19:19.440 --> 19:24.280 Make sure you are connected to the same super base account for the operation mode. 19:24.600 --> 19:28.000 We've got retrieve documents as tool for agent. 19:28.800 --> 19:34.080 So basically you want to retrieve documents from vector store to be used as tool with eye nodes. 19:35.600 --> 19:38.920 Then for the description you can say something like. 19:41.280 --> 19:43.280 Use this to get. 19:45.560 --> 19:46.440 Information. 19:49.080 --> 19:49.920 About. 19:53.600 --> 19:56.800 The AI voice agent prompt engineering framework. 20:16.080 --> 20:16.720 All right. 20:17.840 --> 20:21.600 So this description helps the I know when to use this tool. 20:22.360 --> 20:24.760 And make sure you select the correct table. 20:25.200 --> 20:33.320 So from list select documents not editing chart histories because we want to because we want to retrieve 20:33.320 --> 20:37.760 information based on the files we vectorized. 20:39.480 --> 20:39.840 All right. 20:39.840 --> 20:40.280 Great. 20:40.800 --> 20:43.680 And now let's add our embedding model. 20:45.480 --> 20:49.800 We are going to use OpenAI's embedding model the default one. 20:50.160 --> 20:53.040 So text embedding XML perfect. 20:54.280 --> 20:57.720 And now let's test it first let's hit save. 20:59.040 --> 21:00.040 Let's open chat. 21:02.560 --> 21:05.320 And let's ask it for example. 21:11.270 --> 21:18.470 What is the part two of the framework about? 21:20.670 --> 21:21.550 And hit send. 21:26.150 --> 21:27.030 As you can see. 21:29.390 --> 21:32.430 It retrieves the relevant information from our vector database. 21:35.670 --> 21:37.030 And is preparing the response. 21:37.630 --> 21:42.550 So part two of the Prompt Engineering Framework focuses on defining the personality of the agent. 21:42.830 --> 21:44.110 Here are the key components. 21:44.870 --> 21:51.430 And in our document we have personality emotional and conversational style of the agent. 21:51.750 --> 21:52.510 Okay great. 21:52.910 --> 21:59.430 So the AI searches through our document chunks, finds the most relevant ones, and uses that information 21:59.430 --> 22:00.630 to answer the question. 22:01.470 --> 22:06.710 So if we check our chat histories table, we will see both the question. 22:08.830 --> 22:10.150 And the IRS response. 22:10.390 --> 22:11.110 Start here. 22:12.390 --> 22:17.110 So our agent maintenance memory while also accessing the vector database. 22:17.630 --> 22:19.110 So our documents table. 22:19.710 --> 22:26.670 And this is awesome because by combining both systems we have persistent memory and intelligent document 22:26.670 --> 22:29.230 search using similarity search. 22:29.870 --> 22:30.790 So that's it. 22:30.910 --> 22:36.590 The beauty of this setup is that you can add more documents, um, handle multiple conversations. 22:36.990 --> 22:40.070 And it's all stored securely in your smart base project. 22:40.870 --> 22:46.070 Now we know how to connect Supabase and PostgreSQL to your agents in N810. 22:46.750 --> 22:51.750 Thanks to that, your agents can remember conversations through PostgreSQL. 22:54.190 --> 22:59.230 And intelligently search through documents using Supabase vector database. 23:00.390 --> 23:05.310 So we've got everything you need to build powerful agents that actually remember everything. 23:05.750 --> 23:07.270 I hope this was helpful. 23:07.470 --> 23:09.230 Thanks for watching and see you in the next one.