NEWS · MODELS · #1827
Hands-on with local AI: Hermes Agent running Qwen 3.8 on an M5 Ultra Mac Studio
The author installed the open-source Hermes Agent on an M5 Ultra Mac Studio and ran the 125B-parameter Qwen 3.8 Flash Next model locally, using it for tasks such as a daily morning briefing and reorganizing a Steam library; they highlight the privacy appeal, high hardware requirements, and UX friction around model selection and permissions.
KEY POINTS
- The author installed the open-source Hermes Agent on an M5 Ultra Mac Studio and ran the 125B-parameter Qwen 3.8 Flash Next model locally, using it for tasks such as a daily morning briefing and reorganizing a Steam library; they highlight the privacy appeal, high hardware requirements, and UX friction around model selection and permissions.
- Demonstrates that powerful LLMs can be run locally for privacy-preserving automation but also exposes significant hardware, configuration, and usability frictions for mainstream users.
- Learning to use local AI is exciting, overwhelming, and frustrating
WHY IT MATTERS
Demonstrates that powerful LLMs can be run locally for privacy-preserving automation but also exposes significant hardware, configuration, and usability frictions for mainstream users.