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 is a free, open-source, self-hosted AI desktop app that runs on macOS, Windows, and Linux, and was installed on the M5 Ultra Mac Studio with 256GB of unified memory
- Qwen 3.8 Flash Next, a 125-billion-parameter model weighing roughly 105GB, was selected as the first model, with the author also planning to test smaller Qwen models on an M6 Mac Mini, M5 MacBook Air, and Asus TUF Gaming A14
- A daily morning briefing cron job scans email, calendar, and weather; it initially failed because macOS was asleep at the 7:30am trigger, and the author later acknowledged the briefing 'already broke, multiple times'
- Steam library reorganization sorted more than 400 games by genre in minutes using a web API key the author revoked afterward, preserving custom categories like co-op titles and party games
- Financial records and an embargoed laptop spec comparison were crunched locally — tasks the author explicitly said they 'cannot or will not trust with any cloud service' because of the sensitive or pre-release data
- Apple Mac desktops and new RTX Spark Windows machines with up to 128GB of RAM are being marketed for local and agentic AI workloads
- The author frames local AI strictly as a tool — 'not a digital friend' — and deliberately avoids saying please or thank you when issuing commands
Why it matters: The author's own use cases show local AI is the difference between using the technology and not: a 400-game Steam library, a morning briefing, and especially financial records and embargoed laptop specs are all tasks they said they would never upload to a cloud LLM. That makes the privacy case concrete and replicable for anyone handling sensitive or pre-release data.
Ask SkimNews
