The architecture of Tabby consists of several crates.

• The crates/tabby crate handles core functionalities and the HTTP API.

• The crates/tabby-download crate fetches models from the internet.

• The crates/tabby-inference crate contains traits and utilities for LLM inference.

• The crates/tabby-scheduler crate performs async jobs.

• The ee/tabby-ui crate provides the web UI, while the ee/tabby-webserver crate integrates the UI and provides team management and authentication.