Learning to use local AI is exciting, overwhelming, and frustrating

by | Oct 11, 2026 | Technology

Learning to use local AI is exciting, overwhelming, and frustrating

As local artificial intelligence capabilities expand, a journalist has begun experimenting with running large language models directly on personal computers rather than using cloud-based services. The motivation stems partly from privacy concerns about sending sensitive data to third-party companies like OpenAI, Google, Microsoft, and Anthropic.

The author installed the open-source Hermes Agent, a self-hosted desktop application compatible with macOS, Windows, and Linux, on an M5 Ultra Mac Studio equipped with 256GB of unified memory. Initial setup proved straightforward, with the Hermes interface allowing selection from numerous available models. The author chose the Qwen 3.8 Flash model, a 125-billion parameter system requiring approximately 105GB of storage, and also plans to test smaller variants on devices with more limited resources, including an M6 Mac Mini, MacBook Air, and upcoming Windows machines.

The practical exploration revealed both promise and challenges in finding meaningful applications for local AI. An initial attempt to generate daily morning briefings that scan email, calendar, and weather required troubleshooting, including discovering that the Mac needed to remain powered on for scheduled tasks to execute properly. More successful was using Hermes to reorganize a Steam game library containing over 400 titles into genre-based categories while preserving custom collections. The process required granting limited API access that could be revoked afterward.

Additional experiments included financial data analysis and creation of laptop comparison spreadsheets. The author notes that discovering practical uses for the technology requires deliberate thinking rather than obvious applications, though the ability to process computationally intensive tasks locally without ongoing subscription costs presents potential advantages. Testing continues across multiple hardware configurations to evaluate viability across different device types.

Article Attribution | Read More at Article Source

Article summary produced by Claude AI