Quiet tooling releases matter more than model launch theater. On October 8, Ollama 0.40.2 landed as stable — MLX as the default on Apple Silicon, out of RC — and LM Studio 0.4.26 continued the desktop-local track. If you run agents at home, this is the boring plumbing that keeps the loop alive.
MLX default is the headline
Apple Silicon users should care. Defaults decide what half the community actually benchmarks. MLX as the stable path means fewer “why is this slow on my Mac” tickets and more honest apples-to-apples tests. RC theater is over; the default is the product now.
LM Studio stays the friendly front door for people who do not want to live in a CLI. I use both patterns: Ollama when I am scripting agents and wiring HTTP to a harness, a desktop UI when I am evaluating a model with my eyes and a stopwatch.
House rules
Pin versions. Do not let a background updater surprise a running agent mid-refactor. Keep models and secrets on volumes you understand. Local AI is only “safer” if your ops habits are. Treat the inference box like a small server: updates on purpose, logs you can read, and no mystery ports open to the LAN “just for a minute.”
None of this replaces reading the release notes yourself. Models and agent wrappers change weekly. My bar stays the same: license clarity, local option when secrets are involved, and a workflow I can reproduce without a sales call.
Pin the versions you validate. “Latest” is not a release strategy for anything that can write files.
Have a blessed day!