The Event-Driven AI Stack: Supabase + n8n (Video Companion)

Support the work ⚡ Supabase Sponsor Link

👇 Workflow links below

Most “AI Automation” tutorials teach you to poll APIs. You set up a cron job to check a database every minute: ”Is there new data? No. Is there new data? No.”

It works, but it’s fragile, wasteful, and slow.

In this video, I break down a better way: Event-Driven Architecture using Supabase Webhooks and n8n. This allows your AI agents to react instantly—the millisecond a user uploads a file or saves a draft.

Here is the technical breakdown of the two patterns we built.

Pattern 1: Storage Events for RAG 📂

The goal: User uploads a PDF → System automatically ingests it into a Vector Store.

Instead of watching a folder, we use Supabase Storage Webhooks.

1. Trigger: A file is uploaded to the documents bucket.

2. Event: Supabase fires a POST request to an n8n webhook.

3. Action: n8n grabs the file, chunks it, embeds it, and stores it in your Vector Database (which is also Supabase/Postgres).

This transforms a complex “file watching” script into a simple, reactive workflow.

Pattern 2: Database Events for AI Agents 🤖

The goal: User creates a “Recipe Title” → AI writes the full recipe, tags it, and generates an image.

We use Database Webhooks on the blog_posts table.

1. Trigger: A new row is inserted with status: draft.

2. Event: Supabase sends the payload (including the Row ID) to n8n.

3. Action:

• n8n generates the recipe content using OpenAI.

• n8n generates an image using DALL-E 3.

• n8n updates the original row with the new content and sets status: published.

The frontend (if using Supabase Realtime) updates instantly before the user’s eyes.

The Stack / Links

⚡ Supabase Sponsor Link

🚀 N8N Workflows)

Rag Workflow 👉 Cole Medin’s Channel

If you are building internal tools or AI backends, stop polling and start listening to events. It’s cleaner, faster, and scales with you.