Ollama

Running open-weight models locally, made boring in the best way. One command, sane defaults, and it stays out of the way. The obvious first stop for local inference.

How I tested it

I ran Ollama on macOS · M-series, version 0.5, across two weeks of ordinary project work — the refactors, the migrations, the test coverage. Not a benchmark harness; the real, unglamorous middle of building something. Last retested 6 days ago, and the notes below come from that run.

Where it lands

The honest read after that stretch is in the verdict and the rating, both of which I revisit on every retest. When either moves, the changelog says why. Nothing here is a one-time score.

Verdict

It earns a place in the daily rotation. Use it — and keep verifying the output, like everything else.