AgentSpring is a new agent-based modeling framework especially suited to model and simulate complex socio-technical systems, such as energy markets or transport infrastructures. Common problems encountered when modeling and analyzing such systems are how to represent the variety of facts that describe the system and how to allow agents to make decisions based on those loosely related pieces of information. AgentSpring proposes a solution to these problems by modeling the systems as graphs consisting of agents, artifacts and relations between them. A graph is one of the most flexible data structures, and is made of vertices and edges that connect them. Agents use sophisticated queries to reason over and traverse the graph data structure and make decisions based on the results they discover. They continuously update the graph to include the effects of the decisions they have made. AgentSpring represents the simulated system as a constantly evolving graph of interconnected facts that can transparently be observed and queried in real time by both agents in the model, and the modeler. In addition, AgentSpring allows for the composition of agent behavior modules, where sophisticated behavior can be produced by combining simpler decision making rules. The modeler can easily create a number of heterogeneous agents by combining these behavioral "lego" bricks. Finally, AgentSpring is open-source and is continuously developed based on the feature requests and contributions of the modeler community.