Libmnemosyne consists of a number of components, which can be swapped in and out. This is handled by the ComponentManager. Examples of components are the scheduler, the storage layer, card types, plugins, ...
The basic unit of information from which Cards are derived is called a Fact, containing a set of fields and their associated values. E.g., for a three-sided CardType, these fields are foreign word, pronunciation and translation.
A FactView collects a number of the fields of a Fact into question and answers. E.g., the three-sided CardType has a recognition FactView, where the question contains the foreign word, and the answer contains the pronunciation and the translation.
As mentioned before, a Fact is linked to a CardType, and each CardType lists a set of FactViews.
The actual Cards are generated from the data in Fact using the recipe of a certain FactView. A Card also contains all the repetition data needed for the Scheduler to do its work. Since the question and answers are generated from the Fact data each time a Card is shown, related Cards (i.e. Cards with different FactViews of the same Fact) are always consistent.
The actual displaying of the data in a Card is handled by a Renderer. The default Renderer takes the fields from the Fact, adds them into a html template and applies a CSS for formatting.
At several points in the program, plugin writers can hook in their code using the Hook mechanism.
Before the data is passed to the Renderer, Filters can be applied to it. This is an extra level of flexibility, and can be used to generate LaTeX, convert relative paths to absolute paths, etc ...
To determine which cards are active (i.e.) participate in the review process, a mechanism of ActivityCriterion and CriterionApplier is used.
In order to make it easier for other GUI frontends to be written, all the logic typically needed for GUIs, but that is independent of the actual GUI toolkit used, is abstracted in controllers. In order to get more flexibility, there are two of them: one related to the review process (which is different for different schedulers), and one related to the rest of the program (which normally won’t change).
There is also mechanism for plugins to add new statistical data to the standard statistics in an integrated way.