fastembed-rs

Why we included this project

Rust teams often face a choice between calling a hosted embedding API or standing up their own inference service. fastembed-rs sidesteps that by running ONNX models like BGE, MiniLM, and multilingual E5 directly inside your binary, using the ort runtime and Hugging Face tokenizers. That makes it handy for RAG and semantic search work when you want synchronous, dependency-light inference without pulling in Tokio. It also means documents never leave your machine, which matters for privacy-sensitive or offline setups. Just be aware it is an inference library, not a search system, so you will still pair it with a vector store and your own retrieval logic.