Notifig

Reactive typed configuration for Pydantic — a config tree that tells you what changed.

Notifig is a configuration library built on pydantic and pydantic-settings. You declare a config tree of nested sections; the library loads it from layered files (JSON/YAML/TOML), env vars and the CLI; and — the part nothing else offers — every write notifies the observers that care about it, so the runtime reacts to configuration changes instead of re-reading them.

The shape

from notifig import Config, ConfigSection


class ApiConfig(ConfigSection):
    url: str
    model: str | None = None


class Settings(Config):
    api: ApiConfig

• Sections are components. A section is a ConfigSection — a nested model with its own fields, its own validation, and its own position in the tree. A module that owns a feature declares its own section; the config tree just names them.
• Dynamic sections hold entries that type themselves. A DynamicSection[EntryType] instantiates a config block into an entry the moment the block arrives, and re-types it with set_type(name, model) when the model that owns it becomes known — a config file may hold settings for something not yet installed, and the entry waits, visible and editable, until it arrives.
• Writes notify. Every mutation (set, unset, set_type, remove_type, plain section[name] = value) walks the change to every registered observer — above and below the changed section, two-way path matching — and runs it. Writes are sync and atomic; async observers' coroutines are handed back to the caller (set_and_wait is the one async def), and a plain assignment that would reach an async observer refuses loudly before changing anything (AsyncObserversNotCalled).
• Sources layer. Files deep-merge (local over global), env vars and CLI flags override, and legacy keys are a hard break with an explicit map rather than a silent fallback.

settings.api.model = "new-model"  # sync observers run; async ones are returned
settings["api.url"] = "https://..."  # the same, by name


@settings.on_change("api.model")
def model_changed(section, changes):
    print("the model changed:", changes)

Status

Pre-release. The API is real and battle-tested inside its first application, but the standalone packaging, CI and test suite are being built here now.

Install

pip install notifig

Requires Python ≥ 3.13.