Upstash Vector

Upstash Vector is a serverless vector database from Upstash, designed for vector search at scale without infrastructure management. Developers store high-dimensional vectors, attach metadata, and query by similarity through a REST API or SDKs. The documentation lists use cases such as natural language processing, retrieval augmented generation, recommendation engines, image recognition, and clustering or classification.

What It Is

Upstash Vector is a managed vector database. Vector databases store and search high-dimensional vectors that can represent text, images, sounds, or other media, enabling similarity search across content that traditional databases handle poorly. The docs say Upstash Vector uses DiskANN, an approximate nearest neighbor algorithm, to deliver high recall and low latency.

How It Works

Developers upsert vectors, or raw data that is embedded with a supported embedding model, into an index. They then query by vector or by data, optionally combining the search with metadata filters that limit results to records matching a filter expression. Similarity functions include Euclidean distance, cosine similarity, and dot product. Other documented features include namespaces, sparse and hybrid indexes, and resumable queries.

SDKs and Integrations

The docs describe a REST API and SDKs with first-class Python and TypeScript support, plus PHP (with Laravel), Go, and semantic cache libraries for JavaScript and Python. Documented integrations include Vercel AI SDK, LangChain, LlamaIndex, LlamaParse, Flowise, and Langflow. Tutorials cover semantic search, RAG with LlamaIndex, Hugging Face embeddings, and Gradio applications.