Serve text embeddings and reranking privately, behind one API with a per VM credential and no default login.
Infinity is an open source, high throughput REST server for text embeddings, reranking and CLIP models. It serves models such as BGE, GTE, E5 and Sentence Transformers with dynamic batching behind an OpenAI compatible embeddings API, so existing OpenAI SDK code works unchanged. You call it from your own applications, from the OpenAI SDK or from the CLI.
The Infinity server is bound to loopback and fronted by an nginx TLS proxy that enforces HTTP Basic auth, so nothing is reachable without a credential. A fresh admin password and TLS certificate are generated on first boot into a root only file, so there is no default login. A small open weights model is pre downloaded so the API answers immediately, it generates embeddings on CPU and offloads to the GPU when one is present, and 24/7 cloudimg support plus a paired deploy guide are included.