Chat with Supabase Storage files using n8n AI Agent

Auto-vectorize PDFs and text files from a Supabase bucket, then answer contextual questions via a chat interface

Who

Researchers, analysts, business owners, or anyone managing a large document collection who needs quick contextual retrieval

You get

After uploading files, ask the AI questions about document contents and receive context-based answers

You need

n8n · supabase

Steps

1. Fetch file list from Supabase

   Use an HTTP Request node (or Supabase node) in n8n to send a POST request to Supabase Storage and list files in the target bucket. Replace the hard-coded Supabase URL in the template with your own project endpoint. Add logic to filter out empty-folder placeholders (.emptyFolderPlaceholder) to avoid incorrect processing.

2. Compare and filter files

   Use an Aggregate node to summarize the file list from Storage, then compare it against existing records in the Supabase files table. Use an If node to exclude duplicates and placeholders, keeping only new unprocessed files.

3. Download new files

   For files that pass the filter, use an HTTP Request node to download them. Configure authentication according to your Storage public/private access settings (the template uses supabaseApi credentials). Adjust the GET request URL and parameters to match your Supabase setup.

4. Process by file type

   Use a Switch node to route by extension: PDFs go to an Extract Document PDF node; text files are processed directly. The template Switch checks whether $binary.data.fileExtension equals pdf or is undefined (text).

5. Chunk content

   Split large extracted text using the Recursive Character Text Splitter node. The source default: chunkSize=500 tokens, chunkOverlap=200 tokens. Adjust based on your document structure.

6. Create vector embeddings

   Use the Embeddings OpenAI node (model text-embedding-3-small) to generate embeddings for each chunk. Configure metadata in the Default Data Loader to carry file_id for later retrieval. Data source expression: {{ $('Merge').item.json.data ?? $('Merge').item.json.text }}.

7. Store in Supabase vector store

   Use the Insert into Supabase Vectorstore node to write vectors into Supabase. Table name: documents; query function: match_documents. Ensure your Supabase project has the vector extension enabled and the corresponding table structure created. [First-test safety] Before the first test, use a test Supabase project/bucket and test documents table, or temporarily disconnect Insert into Supabase Vectorstore; write to production only after chunking and embeddings look correct.

8. Connect AI chatbot

   Build the second scenario: Chat Trigger node receives user messages → AI Agent node → linked Vector Store Tool (Supabase Vector Store) retrieves relevant chunks → OpenAI Chat Model generates answers. The Vector Store Tool metadata can specify file_id for targeted queries across multiple documents.

You are setting up n8n template 2621, "AI agent to chat with files in Supabase Storage". Proceed in this order:
1. Confirm you have an n8n instance, a Supabase project (with Storage and vector extension), and an OpenAI API Key.
2. Import the template workflow JSON into n8n and replace all hard-coded Supabase URLs, bucket names, and credentials with your own resources.
3. In Supabase, create the vector table "documents" and the "match_documents" query function (the source does not provide SQL; refer to Supabase official vector docs).
4. Configure supabaseApi and openAiApi credentials.
5. Upload test files to Storage and run Scenario 1 (file processing flow) to confirm vectors are written successfully.
6. Test Scenario 2 (chat flow) by sending a question to the Chat Trigger and verify it answers based on document content.
7. Check the If node filter expression and Switch node extension checks against your actual file types.
Note: chunkSize, chunkOverlap, and topK can be tuned based on actual results.