With an increase in flexibility in industries adopting the tenets of shared, skill-based, and modular production, a higher variability of products a single manufacturer offers can be anticipated. An immediate consequence is a potential inability to instruct assembly personnel in an adequate and detailed manner. This problem would primarily affect small and medium enterprises engaged in individual and small series productions. To face the issue, this paper examines the suitability of facilitating an AI assistant to support workers handling an ever-expanding range of assembly tasks. This assistant would be realized through a retrieval-augmented generation system, founded on an Large Language Model (LLM) and a knowledge base. We especially propose a locally trained and hosted LLM, aiming to enhance effective flexibility and applicability while minimizing individual installation and setup times. By strictly formalizing descriptions of assembly or disassembly steps in a knowledge base, manufacturing difficulties can be presented as informational problems, at which LLMs excel. Through this, we are trying to extend the increased efficiency of knowledge workers empowered by utilization of LLMs such as GPT-4 into manufacturing. Using verbal inputs as well as reading the generated feedback back to a worker, we aim to keep a worker engaged with their primary tasks and, furthermore, reduce idle times caused by knowledge gaps. By taking this verbal/auditive-only approach, we secondarily aim towards increasing worker autonomy by answering miscellaneous workplace-related questions alongside knowledge-based problems of the day-to-day business.