While the technological capabilities of Generative Artificial Intelligence (GenAI) are extensively documented, understanding how teachers exercise professional agency in AI-mediated instruction remains underexplored, particularly in non-Western contexts where institutional structures, technological infrastructures, and policy frameworks differ markedly from settings where most existing research has been conducted. This study addresses this gap by examining how Chinese university English teachers enact agency when integrating localized AI platforms into their pedagogical practices. Drawing on the Ecological Approach to Agency (Priestley et al., 2015), this qualitative inquiry examines the lived experiences of university English teachers who employ domestically developed AI tools in their instruction. Data were collected through semi-structured interviews and teacher reflection journals, and analyzed using thematic analysis. The ecological framework enables examination of agency across multiple temporal and contextual dimensions: past technological experiences, future professional aspirations, and immediate institutional constraints. Findings reveal that teacher agency in AI-mediated contexts operates as a dynamic negotiation rather than a binary acceptance-rejection decision. Teachers demonstrate varied approaches to AI integration, shaped by factors including research performance pressures, student assessment requirements, peer knowledge networks, and ethical concerns regarding academic integrity. The study contributes empirical evidence of teacher agency beyond Western settings and highlights the role of localized technological ecologies in shaping pedagogical decision-making. These findings offer implications for professional development programs that honor teachers' situated expertise in navigating technological change, and for policies aimed at supporting context-sensitive AI integration in higher education.