We describe a framework for modeling and performing inference on complex systems, based on an Attributed Programmed Graph Grammar (APGG) meta-model. We outline this APGG framework, and how it can be used. In particular, show how we can use the underlying system structure as an organising principle for model specification and analysis. In addition, we examine the attribution we assign to the higher-level systems models, namely a constraintbased language, and how a distributed agent-based approach can regulate inference performed asynchronously on several multi-level models.