Key Info

An OpenCode developer describes training a small local model to classify shell commands for a UI issue. While newer capabilities can already handle the same task, local execution may still be preferable — and the setup could be a useful way to generate a lot of synthetic data.

Highlights

  • A small local model was trained to classify shell commands, solving a UI issue.
  • The task is now doable by newer tools or models, but a local approach still has advantages.
  • The developer suggests this pattern could be a strong source of synthetic training data.