In the olden days, most people rolled their own servers to expose their ML models to client applications. In fact, Roboflow Inference's HTTP interface and REST API are built on FastAPI.
In this day and age, it's certainly still possible to start from scratch, but you'll be reinventing the wheel and will run into a lot of footguns others have already solved along the way. It's usually better and faster to use one of the existing ML-focused servers.
Choose FastAPI or Flask if: your main goal is learning the intricacies ofmaking an inference server.
LitServe is a lightweight and customizable inference server focused on serving models with minimal overhead. It is fairly minimalistic but flexible and self-contained.
Like Triton, LitServe is task-agnostic, meaning it is designed to balance the needs of vision models with NLP, audio, and tabular models. This means it's not as feature-rich for computer vision applications (for example, it doesn't have any built-in features for streaming video). It is also highly focused on model serving without an abstraction layer like Workflows (offered by Roboflow Inference) for model chaining and integrations with other tools.
Choose LitServe if: you are working on general-purpose machine learningtasks and were previously considering rolling your own server but wanta more featureful starting point.
Inference turns any computer or edge device into a command center for your computer vision projects.