At the core is an event and entity system. The entities are component based, and access named components via get() and set() methods, which is not great but allows for nice behavior by triggering events on property set, allowing other code to subscribe to changes on entities. For example, the render binds to the "setposition" event and will mark any entity for re-rendering when it has moved.
Entities don't get any logic or behavior. Separately, Behavior types are defined and serve two purposes. Firstly, they get a tick() method called, given a time delta since the last call, and are allowed to do any updating to the entity states. This is where I'm integrating Box2Djs, for example. Secondly, behaviors can subscribe to any events in entities they are applied on, so a Draggable behavior watching an object has an "entity.mouse.click" event fired when any of its entities "mouse.click" event triggers, allowing the behavior to response in whatever way is appropriate.
I am also happy to have, so far, kept the rendering component completely isolated. When I get around to writing a WebGL renderer, it should be able to do so without any changes to entities, behaviors, or anything else.
I'll be posting up a version of both the core JS library and the engine some time this week.