AKIRA: a Framework for MABS Giovanni Pezzulo, Gianguglielmo Calvi ISTC-CNR viale Marx, 15 – 00137 Roma Italy g.pezzulo@istc.cnr.it , calvi@noze.it Abstract. Here we present AKIRA, a framework for Agent-based cognitive and social simulations. AKIRA is an open-source project, currently developed mainly at ISTC-CNR, that exploits state-of-the-art techniques and tools. It gives to the programmer a number of facilities for building Agents at different level of complexity (e.g. reactive, deliberative, layered). Here we describe the main architectural features (such as agents Hybridism and the Energy Pool) and the theoretical assumptions that motivated it; we also present some simulations. 1 Introduction AKIRA is an open source project [16] that is conceived as a general framework allowing both social and cognitive Agent Based simulations. Many existing platforms for social simulation [17, 18] allow to build complex simulations using only very simple Agents. At the contrary, the aim of the Artificial Intelligence Group at ISTC- CNR is to perform social simulations that take advantage of more complex Agents: a theoretical mayor claim of the Group is that many socio-cognitive phenomena (e.g. involving trust [1] and reputation [2]) can be modeled only with Agents having a certain set of cognitive features (e.g. BDI-like Agents [12]). Another current issue of the Group is to implement and test a range of single-Agent cognitive models (e.g. about beliefs and goals dynamics, anticipation and expectations, epistemic actions) that were developed in the Institute throughout the years [1, 2, 11]. So, our first requirement for a platform is representing and implementing in the same framework both the cognitive components of single cognitive Agents and their social dynamics, as well as a dynamical environment. The general AKIRA paradigm is MAS, but the complexity of an Agent can be set at different levels. Each single cognitive Agent is conceived as a complex-Agent, composed by many cooperating and concurrent simple-Agents, (e.g. each representing a goal, or a belief, etc). Simple and complex-Agents can be the actors of a social simulation within the same framework. More, the environment (with its rules and dynamics) can be modeled as an Agent, too, interacting in a transparent way with the other Agents within the framework. Some related requirements are: allow both turn-based and real-time (parallel) simulations; furnish utilities for rapid prototyping of Agents at different complexities (e.g. reactive, BDI-like). At the same time, AKIRA has some built-in properties: Agents can form Coalitions; there is integration of top-down control and emerging behavior, intrinsic pressure of each active Agent over the whole computation, intrinsic concurrency, access to a limited pool of resources.