Build Your First OptPilot Package

catalog/optpilot_tutorial is the release catalog's teaching package. It is small on purpose: one Environment, one Method, one Run setup, and one Resource. Nothing needs an API key, network access, or third-party dependency.

Building block	Responsibility
Environment	Defines the evaluator, Candidate parameter schema, and metrics.
Method	Proposes Candidates compatible with that schema.
Run setup	Pairs the two and declares the objective, budget, and seed.
Resource	Demonstrates a reusable launchable interface with a visual package map.

Run the complete example:

uv run optpilot package validate catalog/optpilot_tutorial --check-source
uv run optpilot run catalog/optpilot_tutorial/studies/find_best_settings.yaml \
  --package-root catalog/optpilot_tutorial

The six deterministic trials tune a toy factory's worker count, buffer capacity, and operating mode. The Method is deliberately implemented in one short Python file so it can be copied and changed. Each trial retains a small evaluation.json artifact.

To start your own package, copy the folder, replace the 32-character identity in optpilot.package.yaml, and change one component at a time. Keep the same identity when merely moving or renaming your package.