Vector databases are the infrastructure behind AI memory, search and RAG and most AI users have never heard of them

When you use an AI tool that can search your documents by meaning rather than by keyword, or that remembers relevant context from previous interactions, there is almost certainly a vector database involved. IBM's explainer https://www.ibm.com/think/topics/vector-database covers how they work without requiring a database engineering background.

The key concept: instead of storing documents as text and searching by exact words, vector databases store mathematical representations of the meaning of content. Searching by meaning rather than by exact keywords is what allows semantic search to find relevant documents even when the query uses different words from the document.

For people building AI products this is essential infrastructure knowledge. For people using AI products it explains why "AI-powered search" produces different results from traditional keyword search and why sometimes that is dramatically better and sometimes frustratingly different from what you expected.

The RAG connection is direct: the retrieval step in RAG uses a vector database to find the most semantically relevant documents for a query before passing them to the LLM as context.

For developers in the community: which vector database are you using in your current AI projects and what has been the most significant limitation or surprise you encountered when working with them at scale?