Load a consumer package's config

Define a Pydantic model for non-sensitive settings. Credentials belong in axm-vault and should be resolved separately.

Declare and load a model

The following Python program selects a named profile for this process and persists a native integer. Use an unused profile name when trying it.

Python

import os
from pydantic import BaseModel
from axm_config import delete, get_file, load, set_

os.environ["AXM_PROFILE"] = "docs-consumer"

class ResearchConfig(BaseModel):
    dataset: str
    timeout: int = 30

set_("research.demo", "dataset", "sample")
set_("research.demo", "timeout", 45)
cfg = load("research.demo", ResearchConfig)
assert cfg.dataset == "sample"
assert cfg.timeout == 45
assert get_file("research.demo", "timeout") == 45

os.environ["AXM_RESEARCH__DEMO_TIMEOUT"] = "10"
assert load("research.demo", ResearchConfig).timeout == 10
assert get_file("research.demo", "timeout") == 45

del os.environ["AXM_RESEARCH__DEMO_TIMEOUT"]
delete("research.demo", "timeout")
assert load("research.demo", ResearchConfig).timeout == 30
delete("research.demo", "dataset")

load(namespace, model) resolves each field name, omits unresolved fields and calls model.model_validate. Model defaults and default factories then apply. Required missing fields and validation errors become ConfigError. Pydantic performs conversions; get itself keeps environment values as strings. Aliases are not used as configuration lookup keys.

Diagnose an override

Run both commands with the same profile and environment as the consumer:

Bash

AXM_PROFILE=docs-consumer AXM_RESEARCH__DEMO_TIMEOUT=10 axm-config doctor research.demo
AXM_PROFILE=docs-consumer AXM_RESEARCH__DEMO_TIMEOUT=10 axm config_doctor --namespace research.demo

The report names the winning layer without returning the value. It still parses the file internally and may create or chmod the AXM home. Specify the namespace to discover environment-only settings.