Config containers

OmegaConf.create() with no argument makes an empty DictConfig. A dictionary also becomes a DictConfig, while a list becomes a ListConfig. These containers can be nested. Each key or list position is a node, but reading it gives you its value. For example:

>>> from omegaconf import OmegaConf
>>> cfg = OmegaConf.create({
...     "server": {
...         "host": "localhost",
...         "port": 80,
...     },
...     "users": ["a", "b"],
... })
>>> cfg.server.port
80
>>> cfg["server"]["port"]
80
>>> cfg.users[0]
'a'

Use attribute access for keys that work as Python attributes. Item access also works for other keys and for list positions. You can change a value or add a key to a DictConfig created from a plain dictionary:

>>> cfg.server.port = 8080
>>> cfg.server["public-host"] = "example.org"
>>> cfg.server["public-host"]
'example.org'

Use .get("key", default) when a mapping key may not exist. To pass a config to code expecting plain Python containers, use OmegaConf.to_container().

Dictionary keys may be str, int, bool, float, bytes, or Enum members. An integer path element can refer to either the integer key 1 or the string key "1"; a DictConfig cannot contain both because the path would be ambiguous. This fallback does not infer other non-string key types from path text.

Sequence containers

Lists become mutable ListConfig values. In 2.4, tuples become immutable TupleConfig values. Use a list when you need to change individual elements; replace the whole tuple when you need different tuple contents. The tuple migration guide explains the change from earlier releases.

Next, learn how to mark a value as missing when it must be provided later.