In this paper we present characteristics, features and computational strategies used in MAGI (Multi-Agent Geosimulation Infrastructure) which are relevant for strongly geo-spatially oriented agent-based simulations. These characteristics make MAGI an effective modelling and simulation environment, in terms of generality, user friendliness, modelling flexibility, interoperability with GIS datasets and computational efficiency. The infrastructure is composed of a development environment for building and executing simulation models, and a class library based on open source components. Differently from most of the existing tools for geosimulation, both raster and vector representation of simulated entities are allowed and managed with efficiency. This is obtained through the integration of a geometry engine implementing a core set of operations on spatial data through robust geometric algorithms, and an efficient spatial indexing strategy for moving agents.