ChatGPT Limits and Local LLMs: An Experiment with Hermes Agent

The author conducted an experiment to replace cloud-based Hermes Agent capabilities with local LLMs using a home PC equipped with an RTX 3080 and 32GB of RAM. The study compared Qwen and GLM models across architectural tasks, text editing, and data extraction. Various quantization levels were tested, and a connection was established between a Hermes client on Mac and a local server on Windows. Despite successful technical integration, the author notes that achieving response quality comparable to cloud solutions has not yet been possible. The article details testing conditions, performance metrics, and the rationale behind choosing the Qwen 3.5 9B model. The experiment highlights the current limitations of local LLMs when attempting to fully replace paid API services for daily professional workflows.