The introduction of new variants and the difficulty of forecasting future market demand and developments aggravate the synchronisation of assembly lines. This ultimately leads to cycle time spreads and thus to efficiency losses, e.g. due to lower employee utilisation. In response, matrix-structured assembly systems have been developed as a concept of cycle time independent flow production. Essential characteristics of this type of assembly systems are the dissolution of both one-dimensionally arranged assembly stations as well as cycle times across assembly stations. In recent years, the focus has been on assembly control for the routing of orders through a matrix-structured assembly system. However, order release strategies have largely been neglected, which means that the actually promised performance of this new organisational form of assembly cannot be fulfilled. An agent-based release decision enables the optimal scheduling of new orders taking into account current information from the assembly system such as station states or the processing progress of orders that have already been released. This work extends and builds on existing agent-based approaches to control matrix-structured assembly systems in regard to order release. This results in a theoretical improvement in key performance indicators such as throughput time and station utilisation. For this purpose, the release process, as well as the associated calculation logics and constraints, are described and the implementation in an environmental model is outlined. An essential part of calculation logics is the prediction of all possible paths and capacity requirements resulting from routing and sequence flexibility. This work contributes to the practical realisation and economic operation of matrix-structured assembly systems.