Multi-agent systems (MAS) is an accepted paradigm in safety-critical systems, like the production automation. Agents control the underlying machinery they are representing and interact with each other to achieve an optimal production rate. However, the key requirement is to minimize downtime of the production system in order to allow just-in-time delivery and minimal production costs. Nevertheless, as any other distributed system MAS are prone to failures as well, and consequently an agent may crash. In this case it is important that after recovery the agent achieves its optimal production state as fast as possible. Given the recovery problem, in this paper we address the assimilation sub-problem, i.e. the problem of catching up with the external environment. We propose an abstraction framework called MozartSpaces that keeps the state information, necessary for an agent to behave correctly, up-to-date transparently to the agents. This allows the failed agent to continue its work after recovery with zero-delay in the state where it would have been without a crash.