Learning to use local AI is exciting, overwhelming, and frustrating — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Hermes Agent, an open-source, free self-hosted AI desktop app for macOS/Windows/Linux, was installed on an M5 Ultra Mac Studio with 256GB unified memory and paired with the Qwen 3.8 Flash Next model — 125 billion parameters weighing roughly 105GB
- A daily morning briefing cron job scanning email, calendar, and weather was the first real-world test; it initially failed because macOS was asleep at the 7:30am trigger time but now runs reliably
- Hermes sorted the author's 400+ game Steam library by genre in minutes using a temporary Steam web API key, which was revoked immediately after the one-time sort
- The author ran financial data analysis and built a laptop spec comparison spreadsheet on embargoed hardware locally — explicitly stating they 'cannot or will not trust' that data to any cloud service
- Currently in progress: scripting laptop benchmark test automation via Hermes-generated Python, which the author calls 'a work in progress' that has already been 'a bit of an endeavor'
- The author frames local AI as just a tool, not a companion — refusing to say 'please' or 'thank you' and remaining 'stone-cold and indifferent' when issuing commands
Why it matters: This diary shows local AI on a $12,000 M5 Ultra Mac Studio can now handle real busy work — sorting 400+ Steam games, crunching financial records, building spreadsheets from embargoed laptop specs. The hardware barrier is steep, but the author demonstrates the privacy-respecting alternative is no longer purely theoretical for routine automation.
Ask SkimNews
