This paper describes a distributed agent-based paradigm for hybrid (soft and hard) computation in information (data) fusion. The computations are the outcome of deterministic, nondeterministic or stochastic interaction among the agents, in an environment where information arises from database and multiple sensors. These interactions are like chemical reactions and the evolution of the multiset of agents can mimic the data fusion in a complex sensory system to achieve a required outcome. Since the reaction rules are inherently parallel, any number of actions can be performed cooperatively or competitively among the subsets of the agents. We also describe how scale-free and small world networks arise in the connectivity structure of agents during the information fusion and can result in swarming.