Do I need to store embeddings in the same DB as my app data?

No—you don’t need to store embeddings in the same database as your app data.

Common options

1. Same DB

   • Simpler to manage
   • Easier joins/filters with metadata
   • Good for smaller systems or when using a DB with vector support

2. Separate vector DB / index

   • Better if you have lots of embeddings or high-dimensional similarity search
   • Can scale independently
   • Often faster for retrieval use cases

3. Hybrid

   • App data stays in your main DB
   • Embeddings + vector search in a dedicated vector store
   • Store only the embedding record ID and metadata in the main DB

Typical best practice

• Keep source-of-truth app data in your main relational DB
• Store embeddings where they’re most efficient to query
• Link them with a stable ID

Use the same DB if:

• You want simplicity
• Your dataset is small/medium
• Your database supports vectors well (e.g. Postgres with pgvector)

Use a separate DB if:

• You expect large-scale similarity search
• You need specialized vector indexing/performance
• You want to isolate workloads

If you want, I can recommend an architecture based on your stack (Postgres, MongoDB, Pinecone, etc.).