Local AI
Choosing, running and evaluating open models on your own hardware: what fits, what finishes, and what it costs in time.
6 essays
Splash! A Test of Real Work
Two Qwen 3.8 27B builds audited a wine e-commerce site. Splash finished sooner, but their paths diverged. The session logs show what we can compare.
The Best Local Models for a 64 GB MacBook Pro
The page I keep updated with the local model I would actually run on a 64 GB MacBook Pro today, why I picked it, and what else I tried.
Battle of the Qwens: 27B Versus Custom Flash
I built a custom Qwen Flash model to beat the 27B I rely on. It is quicker on the work both of them finish, and it writes well — but sometimes it never stops.
Which Qwen 3.8 27B Build Should You Run?
Half a dozen downloads all call themselves Qwen 3.8 27B. I compared them on a 64 GB MacBook Pro to find the one worth keeping, and when the others are worth it.
Running Local AI Models on a 64 GB MacBook Pro
What a 64 GB MacBook Pro can actually do offline, and how I choose a local AI model package, a runtime and the memory settings, tested mainly on coding and structured tasks.
How I Made Qwen Flash-Next Fit on a 64 GB MacBook Pro
Qwen's Flash-Next ships bigger than my Mac's memory. I cut the expert bank, mixed the precision and left the biggest file on the SSD. The build runs, with limits on memory headroom and task completion.