Upstash Vector is a dedicated vector database for storing document embeddings and searching them efficiently. In a retrieval-augmented chat application, uploaded PDF or DOCX content becomes embeddings that support similarity search and grounded responses. section 11.1.1 section 11.1.3
A typical integration connects the database to these application steps:
- Parse and analyze uploaded documents, then convert their content into vector embeddings. section 11.1.1
- Store embeddings externally in Upstash Vector instead of relying on local filesystem storage. section 11.1.1
- Embed a user question, retrieve relevant document chunks, and pass them to the Vercel AI SDK for a context-aware response. section 11.1.1
- Use LangChain.js and the community-supported UpstashVectorStore package to expose the database through retrievers. section 11.1.3
Knowledge-base chat workflow
A user uploads To Kill a Mockingbird to a knowledge base. The application stores its vectorized document content in Upstash Vector; a later question retrieves relevant chunks and produces a conversational answer grounded in that document. section 11.1.1
Figure 11.5 Once the knowledge base contains a few documents, users can start chatting with them conversationally.
Shows the user-facing chat experience after a knowledge base contains documents. section 11.1.1
