Qualitative coding, the process of assigning labels to text as part of qualitative analysis, is time-consuming and repetitive, especially for large datasets. While available QDAS sometimes allows the semi-automated extension of annotations to unseen data, recent user studies revealed critical issues. In particular, the integration of automated code suggestions into the coding process is not transparent and interactive. In this work, we present Cody, a system for semi-automated qualitative coding that suggests codes based on human-defined coding rules and supervised machine learning (ML). Suggestions and rules can be revised iteratively by users in a lean interface that provides explanations for code suggestions. In a preliminary evaluation, 42% of all documents could be coded automatically based on code rules. Cody is the first coding system to allow users to define query-style code rules in combination with supervised ML. Thereby, users can extend manual annotations to unseen data to improve coding speed and quality.