Implementing efficient machine learning (ML) is extremely difficult for Internet of Things (IoT) devices with limited computational power, memory, and energy resources. We introduce TinyQL, a quantum machine learning framework for Internet of Things (IoT) devices with limited resources. The framework tackles three main issues: dimensionality reduction using variational quantum circuits while maintaining anomaly detection capabilities; model compression using entropy-based structured pruning and parameter quantization; and effective encoding of classical network traffic data into quantum states using amplitude embedding. TinyQL can be deployed on devices with limited quantum simulation capabilities by combining a quantum autoencoder with traditional optimization approaches. The CICIDS 2017 intrusion detection dataset is used to assess the framework, and it is contrasted with traditional autoencoder methods. According to experimental data, TinyQL requires substantially fewer parameters and has a smaller memory footprint than classical approaches while achieving detection accuracy that is equivalent. Deployment on resource-constrained IoT devices is made possible by the optimized models' 98.08% detection accuracy and up to 93.75% memory savings when compared to the original quantum model. An entropy-based pruning strategy tailored for quantum circuits, a novel quantum autoencoder architecture optimized for IoT constraints, a thorough evaluation framework contrasting quantum and classical approaches for IoT anomaly detection, and useful recommendations for implementing quantum machine learning models on edge devices are among the contributions.