Creating Databases and Vector Databases in Supabase

This guide walks you through creating your first database table in Supabase, and then a special kind called a vector database — used to power AI features like chatbots that search your own documents. No coding experience required; everything below can be done by clicking through the Supabase dashboard, with a small amount of copy-paste where noted.

---

What's the difference?

A normal database table stores information the way a spreadsheet does — rows and columns, like a list of customers, orders, or products.

A vector database stores that same kind of information, but adds one special column that captures the meaning of a piece of text (a paragraph, a support article, a PDF chunk) as a list of numbers. This is what allows an AI assistant to find "the most relevant answer" to a question, rather than just matching exact keywords.

---

Prerequisites

• Access to your project in the Supabase dashboard (provided as part of your GLBNXT platform).

---

Steps

Step 1: Create a normal database table

1. Open your Supabase dashboard and select your project.
2. In the left sidebar, click Table Editor.
3. Click New Table.
4. Give your table a name, for example customers.
5. Add the columns you need — think of each column as a category of information, like name, email, or signup_date. Click Add Column for each one, and choose the matching type (text, number, date, etc.).
6. Click Save.

Your table now exists, and Supabase automatically makes it available to any connected tool or app, with no extra setup needed.

💡 Tip: You don't need to plan every column perfectly upfront. You can always come back and add or remove columns later.

---

Step 2: Turn on vector support

Before you can create a vector database, your project needs one setting turned on.

1. In the left sidebar, click Database.
2. Click Extensions.
3. Search for vector.
4. Toggle it on.

This only needs to be done once per project.

---

Step 3: Create your vector table

Vector tables need one small piece of setup that isn't yet available as a simple button in the dashboard, so this step uses the SQL Editor.

1. In the left sidebar, click SQL Editor.
2. Click New Query.
3. Paste the following:

create table documents (
  id bigserial primary key,
  content text,
  metadata jsonb,
  embedding vector(1536)
);

1. Click Run.

In plain terms, this creates a table called documents with:

• content — the actual text (e.g. a paragraph from a manual).
• metadata — extra details about where it came from (e.g. the file name).
• embedding — the "meaning fingerprint" of that text, generated automatically by an AI model elsewhere in your workflow (for example in n8n or Langflow).

💡 Tip: The number 1536 matches the AI model most commonly used to generate these "fingerprints". If your GLBNXT setup uses a different model, we can help you confirm the right number to use here.

---

Step 4: Let your table connect to sources (optional, but recommended)

If your documents come from different places — different files, folders, or knowledge bases — it's useful to also keep a simple normal table listing those sources, and link your vector table to it. This way, every AI search result can also tell you exactly which file or document it came from.

create table sources (
  id bigserial primary key,
  name text
);

alter table documents add column source_id bigint references sources(id);

Now every row in your vector table can point back to a row in your normal table — so results always come with clear context, not just raw text.

---

What happens next?

Once these tables exist, the rest of the work — actually generating the "meaning fingerprints" and searching through them — happens in the tools connected to your Supabase project, such as n8n or Langflow.

---

Need More Information?

If anything in this guide is unclear, or you'd like help deciding how to structure your specific tables, reach out or visit the Supabase Documentation Page.

---

Need help? Contact the GLBNXT support team or ask a GLBNXT agent to walk you through the setup.