4KAgent is a NeurIPS 2025 research project that builds an intelligent agent for image super-resolution. The agent leverages large multimodal language models (MLLMs) and vision-language models to reason about and orchestrate a series of computer vision steps — such as face enhancement, text removal, and detail restoration — to transform any input image into high-quality 4K resolution. It represents an agentic approach to low-level vision tasks, using LLMs to plan and coordinate specialized vision modules.