Registry

Stores model versions, call state and decision logs. workspace.registry selects it:

URI	Backend
local://./.pecca (default)	files, see workspace/project/call
mlflow://databricks-uc/<catalog>.<schema>	Databricks tracking + Unity Catalog models
mlflow://sqlite/<path>.db, mlflow://http(s)/<host>	local SQLite or an MLflow server
wandb://<entity>/<project>	Weights & Biases artefacts

See MLflow / Unity Catalog. Interface: save_model, load_model, list_versions, get_state, set_state, append_logs, read_logs.