Set up Pinecone for my app's AI memory | Vovy

Create a Pinecone index with the right dimensions, load your documents as embeddings, and query it from your app for fast AI retrieval.

Create a Pinecone index and load my documents into it

How it works

Create a Pinecone project: Vovy opens Pinecone, you sign in, and it creates a project on the free Starter plan and copies the API key to your server secrets.
Create an index that fits: Vovy creates a serverless index with the dimension matching your embedding model and the cosine metric. A mismatch here causes "dimension mismatch" errors later.
Chunk and upload documents: Vovy asks Claude Code for a script that splits documents into chunks, embeds them and upserts them with metadata like title and URL.
Query from your app: Claude Code adds a server function that embeds the user's question and asks Pinecone for the closest chunks.
Show sample results: Vovy runs a few questions and shows the matched chunks with their scores on a card, so you can see retrieval working.

If you already use Supabase, pgvector is simpler. Pinecone shines when you want a managed vector store separate from your database.

The Starter plan is free with limits on storage and usage, which is enough to build and test.

Yes, but you must re-embed everything into a new index, since different models produce different dimensions.