This artifact contains the implementation and datasets for PromptEcho, a passive eavesdropping attack that infers semantically meaningful portions of natural-language responses from encrypted network traffic via a merged token-length side channel. The artifact includes two-stage fine-tuning scripts, an inference pipeline, an evaluation framework, training/test datasets of token-length sequences, and anonymized exemplar network traces from ChatGPT and DeepSeek API sessions. Full details are provided in the README.