Seed Supabase Vector Store with Ollama Embeddings (n8n)

Seed your Supabase Vector Store with Ollama embeddings (n8n)

This n8n workflow loads your own reference text into a Supabase pgvector table—using Ollama embeddings—so you can build an AI assistant that retrieves relevant answers from your documents.

What this workflow does

• Manual start: Run the workflow when you’re ready to seed or update your vector database.
• Load reference entries: Define a list of reference text entries inside the workflow.
• Chunk for reliable retrieval: Split each entry into smaller overlapping chunks for consistent similarity search.
• Generate embeddings with Ollama: Create vector embeddings for each chunk using Ollama’s nomic-embed-text model.
• Store in Supabase: Insert chunk text, metadata, and embeddings into a Supabase pgvector table using the configured match_reference_documents SQL function.

Use cases

• Customer support FAQ bots: Seed product FAQs, troubleshooting guides, and policy text for fast retrieval.
• HR onboarding: Turn onboarding documents and SOPs into searchable knowledge for new hires.
• Operations knowledge base: Make internal procedures searchable by chunk-based vector lookup.
• Research & reference organization: Store notes and reference data so an AI can pull the right excerpts.

Technical details (built with n8n)

• Integrations: Supabase (pgvector via reference_documents + match_reference_documents), Ollama (local embeddings)
• Embeddings model: nomic-embed-text
• Nodes / stack: Code, Sticky Note, Manual Trigger, n8nn8n-nodes-langchainembeddings ollama, n8nn8n-nodes-langchainvector store supabase, n8nn8n-nodes-langchaindocument default data loader
• Setup requirements: Supabase project with pgvector enabled; Ollama running locally with the nomic-embed-text model pulled
• Customization: Replace placeholder text in the reference chunk list; adjust chunk size/overlap as needed (and optionally swap embeddings providers as supported).