Plugin architecture

Plugins are ordinary Python distributions. They expose a registration callable through the chimera_ml.plugins entry-point group. On startup, register_all imports built-in modules and then loads each discovered entry point once.

Lifecycle

pip install plugin
  → import chimera_ml.plugins.register_all()
  → discover entry point
  → load object
  → call it when callable
  → registry keys become available to YAML builders

An entry point may target a callable registration function or a module-level object. If the loaded object is callable, Chimera ML calls it; otherwise loading the module is enough for decorator side effects. Plugin loading failures emit a warning and do not replace the core process with a traceback.

Inspect discovery and registration separately:

chimera-ml plugins list
chimera-ml registry list --type models

The plugin how-to and authoring tutorial show a real package layout based on the examples in this repository.