fix(proxy): read LITELLM_ENABLE_MODEL_CONFIG_CREDENTIAL_OVERRIDES env var

The credential routing docs offer the env var as an alternative to
litellm_settings.enable_model_config_credential_overrides, but nothing
ever read it, so enabling the feature that way left it inert.

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/__init__.py

@@ -373,7 +373,9 @@ prompt_name_config_map: Dict[str, PromptSpec] = {}
 enable_preview_features: bool = False
 return_response_headers: bool = False  # get response headers from LLM Api providers - example x-remaining-requests,
 enable_json_schema_validation: bool = False
-enable_model_config_credential_overrides: bool = False
+enable_model_config_credential_overrides: bool = (
+    os.getenv("LITELLM_ENABLE_MODEL_CONFIG_CREDENTIAL_OVERRIDES", "false").lower() == "true"
+)
 enable_key_alias_format_validation: bool = (
     False  # opt-in validation of key_alias format on /key/generate and /key/update
 )

litellm/proxy/litellm_pre_call_utils.py

@@ -2380,7 +2380,6 @@ def _apply_credential_overrides_from_model_config(
     5. Team default override (defaultconfig)
     6. Deployment default (no action needed)
     """
-    # Feature flag gate — disabled by default, opt in with litellm.enable_model_config_credential_overrides = True
     if not litellm.enable_model_config_credential_overrides:
         return
 

tests/test_litellm/proxy/test_litellm_pre_call_utils.py

@@ -2,6 +2,9 @@ import asyncio
 import copy
 import json
 import os
+import subprocess
+import sys
+import textwrap
 import time
 from datetime import datetime, timezone
 from types import SimpleNamespace
@@ -4951,6 +4954,35 @@ def test_apply_overrides_feature_flag_disabled_by_default():
     assert "api_key" not in data
 
 
+@pytest.mark.parametrize(
+    "env_value, expected",
+    [("true", True), ("True", True), ("false", False), (None, False)],
+)
+def test_credential_overrides_flag_reads_env_var_at_import(env_value, expected):
+    """LITELLM_ENABLE_MODEL_CONFIG_CREDENTIAL_OVERRIDES is read at import, so check it in a fresh interpreter."""
+    env = {k: v for k, v in os.environ.items() if k != "LITELLM_ENABLE_MODEL_CONFIG_CREDENTIAL_OVERRIDES"}
+    if env_value is not None:
+        env["LITELLM_ENABLE_MODEL_CONFIG_CREDENTIAL_OVERRIDES"] = env_value
+    result = subprocess.run(
+        [
+            sys.executable,
+            "-c",
+            textwrap.dedent(
+                """
+                import litellm
+                print(litellm.enable_model_config_credential_overrides)
+                """
+            ),
+        ],
+        capture_output=True,
+        text=True,
+        env=env,
+        timeout=180,
+    )
+    assert result.returncode == 0, result.stderr
+    assert result.stdout.strip() == str(expected)
+
+
 def test_extract_credential_provider_hint_prefers_exact_match():
     """Provider hint selects the correct provider in a multi-provider entry."""
     entry = {