Index Google Drive files into a Supabase vector store with OpenAI embeddings

📺 Full walkthrough video: https://youtu.be/r5kNla0O7I

Author: [Cole Medin](https://www.youtube.com/@ColeMedin)

Who it's for

This workflow is for developers, data engineers, and knowledge management teams who need to automatically ingest documents stored in Google Drive into a searchable vector database — supporting RAG (retrieval-augmented generation) pipelines or semantic search applications.

How it works

1. One-time setup: A chat trigger runs SQL queries to create the required Postgres tables (documents, documentmetadata, documentrows) and the vector similarity match function in Supabase/Postgres. 2. Trigger: Two Google Drive triggers detect newly created or updated files in a watched folder and pass them into a batch loop. 3. Clean old data: For each file, stale document rows and vector embeddings are deleted from Supabase before re-processing. 4. Metadata upsert & download: Document metadata (ID, title, URL) is upserted into Postgres, then the file binary is downloaded from Google Drive. 5. Route by file type: A Switch node directs each file to the correct extractor — PDF, Word/Office document, Excel spreadsheet, or CSV. 6. Tabular data storage: Excel and CSV rows are inserted as raw JSONB records into Postgres and aggregated into a summary. 7. Embedding & storage: All extracted text (documents, PDFs, tabular summaries) is chunked with a character text splitter, embedded via OpenAI, and inserted into the Supabase vector store.

How to set up

- [ ] Connect Google Drive OAuth2 credentials to the two trigger nodes and the download node - [ ] Add Supabase credentials to the delete and vector store insert nodes - [ ] Add Postgres credentials to all Postgres nodes (table creation, metadata upsert, schema update, row insert) - [ ] Set your OpenAI API key in the OpenAI Embeddings node - [ ] Run the setup flow once via the chat trigger to create all database tables and the vector match function - [ ] Set the Google Drive folder ID to watch in both trigger nodes - [ ] Tune the Character Text Splitter chunk size and overlap to fit your document sizes