An installation of Annif may contain multiple independent projects, each of which specifies a set of settings (e.g. backend, language and indexing vocabulary). Projects are limited to a single language; multilingual indexing can be performed by defining multiple projects, one per language. See Project configuration for more details.

Each project defines a (typically large) number of subjects, which reflect concepts from an indexing vocabulary. Subjects are typically created from a corpus extracted from existing metadata records and/or indexed documents. See Subject vocabulary formats for more information about the supported subject formats.

Further, there can be several independent backends that provide subject suggestions. Backends can be integrated into Annif itself, or external services queried via APIs. A project can make use of several backends and combine their analysis results.

Annif has a core application (using Flask and Connexion), which provides both a REST API (when run as a web application) and a command line interface. The command line interface supports administrative functionalities such as corpus management, quality evaluation and other auxiliary functions.