LLM-dCache: Improving Tool-Augmented LLMs with GPT-Driven Localized Data Caching