The author is testing local AI deployment on high-end hardware, specifically using the Hermes Agent desktop application. This tool runs on macOS, Windows, and Linux and supports local large language models without cloud dependency.
Testing was conducted on a Mac Studio equipped with an M5 Ultra chip and 256GB of unified memory. The setup utilized the Qwen 3.8 Flash Next model, which features 125 billion parameters and requires approximately 105GB of storage.
Practical applications included automating a daily morning briefing via a Telegram bot, which aggregates email and calendar data. The author also used the agent to categorize a Steam library containing over 400 games by genre.
Additional experiments involved local data analysis of financial records and the creation of a laptop specification comparison spreadsheet. These tasks were performed locally to ensure data privacy and handle embargoed information.
Future development plans include automating laptop benchmark testing scripts. The author is also preparing to test smaller Qwen models on various systems, including an M6 Mac Mini, an M5 MacBook Air, and an Asus TUF Gaming A14.
