Day 22 – Testing Google File Search as a low-cost RAG

🔥

Today I remembered something important.

Google Gemini offers a File Search feature that is supposed to be much cheaper than Pinecone or Supabase. So I made it my goal for today to test it and see how it works inside n8n.

From what I have read, it is a complete simplification of the whole RAG workflow.

Here is what makes it different.

With Pinecone or Supabase you have to build the entire RAG pipeline yourself. That includes:

• Chunking the documents

• Managing embeddings

• Calculating and storing vectors

• Handling metadata

With Google File Search, all of that shall be handled automatically. We will see what will happen.

You shall upload a file and that it is. No chunking, no embedding model selection, no vector Google does all of this internally.

I am curious to see if it really is that simple.

The setup

First I had to get an API key, which meant creating a project in Google Cloud and adding a billing account.Once that was done, I had my key and could start.

In n8n I created a new credential and added everything into the HTTP Request node.

Then I created the store, which is basically the container where the files will be indexed.

Next I added a form trigger to upload the document, followed by an HTTP Request for the file upload and another HTTP Request to import the file into the store.

And here is the interesting part:

My full patent file with all images was accepted by Google without any issues... interesting. Nut i am not sure.

I am curious to see the results later, but not today.This setup already took more time than expected. Still, I am happy that everything worked so smoothly.

Another good step in the journey.

4 comments