Moss
Moss is a semantic search runtime that embeds directly in an application, built for conversational AI agents that need retrieval latency under 10 milliseconds without running a separate vector database.
Moss runs hybrid semantic-and-keyword search and embedding generation inside an application's own process, rather than calling out to a remote vector database, so there is no network hop on the retrieval hot path. It ships as an SDK for Python, TypeScript, Elixir and C, plus a separate WebAssembly build that runs client-side in the browser.
It includes built-in embedding models, so no external embedding API key is required, though one can be supplied, along with metadata filtering operators. On a benchmark of 100,000 documents and 750 queries, Moss reports single-digit-millisecond p50 through p99 latency, compared to hundreds of milliseconds for Pinecone, Qdrant and ChromaDB in the same test.
It requires signing up at moss.dev for a project ID and key, with a free tier available, and needs Python 3.10+ or Node.js 20+.