diff --git a/README.md b/README.md
index e927c80b8b4..d2cceafddf0 100644
--- a/README.md
+++ b/README.md
@@ -104,6 +104,10 @@ response = completion(model="openai/gpt-4o", messages=[{"role": "user", "content
 response = completion(model="anthropic/claude-sonnet-4-20250514", messages=[{"role": "user", "content": "Hello!"}])
 ```
 
+By default, LiteLLM loads `.env` when imported in DEV mode. To disable this for the SDK and proxy, set
+`LITELLM_DISABLE_DOTENV=1` in the process environment before importing LiteLLM. Setting it inside `.env` is too late
+because LiteLLM reads the flag before loading that file
+
 ### AI Gateway (Proxy Server)
 
 [**Getting Started - E2E Tutorial**](https://docs.litellm.ai/docs/proxy/docker_quick_start) - Setup virtual keys, make your first request
diff --git a/litellm/__init__.py b/litellm/__init__.py
index c8df4394a06..f2b42054241 100644
--- a/litellm/__init__.py
+++ b/litellm/__init__.py
@@ -11,12 +11,14 @@ warnings.filterwarnings("ignore", message=".*Accessing the.*attribute on the ins
 # cannot enforce it at runtime, which floods proxy boot once such a type is schema-walked
 warnings.filterwarnings("ignore", message=".*`ReadOnly` qualifier.*")
 ### INIT VARIABLES #########################
-import threading
 import os
+import threading
 
 # Load .env before any other litellm imports so env vars (e.g. LITELLM_UI_SESSION_DURATION) are available
 import dotenv as _dotenv
 
+from ._dotenv_loader import should_load_dotenv as _should_load_dotenv
+
 
 def _dev_env_hot_reload_enabled() -> bool:
     """The proxy exports this flag when started with ``--reload``. A reloaded
@@ -26,7 +28,7 @@ def _dev_env_hot_reload_enabled() -> bool:
     return os.getenv("LITELLM_DEV_ENV_HOT_RELOAD") == "True"
 
 
-if os.getenv("LITELLM_MODE", "DEV") == "DEV":
+if _should_load_dotenv():
     _dotenv.load_dotenv(override=_dev_env_hot_reload_enabled())
 
 from collections.abc import Mapping, Sequence
diff --git a/litellm/_dotenv_loader.py b/litellm/_dotenv_loader.py
new file mode 100644
index 00000000000..399505bf892
--- /dev/null
+++ b/litellm/_dotenv_loader.py
@@ -0,0 +1,9 @@
+import os
+from typing import Final
+
+_TRUTHY_VALUES: Final = frozenset({"1", "true", "t", "yes", "y"})
+
+
+def should_load_dotenv() -> bool:
+    disable_dotenv: Final = os.getenv("LITELLM_DISABLE_DOTENV", "").strip().casefold()
+    return os.getenv("LITELLM_MODE", "DEV") == "DEV" and disable_dotenv not in _TRUTHY_VALUES
diff --git a/litellm/proxy/proxy_cli.py b/litellm/proxy/proxy_cli.py
index ac69f2e3894..2ff008cb0c7 100644
--- a/litellm/proxy/proxy_cli.py
+++ b/litellm/proxy/proxy_cli.py
@@ -18,6 +18,7 @@ from dotenv import load_dotenv
 from pydantic import BaseModel, ConfigDict
 
 import litellm
+from litellm._dotenv_loader import should_load_dotenv
 from litellm.constants import DEFAULT_NUM_WORKERS_LITELLM_PROXY
 from litellm.proxy.db.pgbouncer import (
     PgBouncerError,
@@ -52,7 +53,7 @@ sys.path.append(os.getcwd())
 config_filename: Final = "litellm.secrets"
 
 litellm_mode: Final = os.getenv("LITELLM_MODE", "DEV")  # "PRODUCTION", "DEV"
-if litellm_mode == "DEV":
+if should_load_dotenv():
     load_dotenv()
 from enum import Enum
 
diff --git a/tests/test_litellm/test__dotenv_loader.py b/tests/test_litellm/test__dotenv_loader.py
new file mode 100644
index 00000000000..c242bb26c05
--- /dev/null
+++ b/tests/test_litellm/test__dotenv_loader.py
@@ -0,0 +1,84 @@
+import os
+import subprocess
+import sys
+import textwrap
+from collections.abc import Mapping
+from pathlib import Path
+from typing import Final
+
+import pytest
+
+
+_CONTROL_ENV_VARIABLES: Final = frozenset(
+    {
+        "LITELLM_DISABLE_DOTENV",
+        "LITELLM_DEV_ENV_HOT_RELOAD",
+        "LITELLM_DOTENV_LOADED",
+        "LITELLM_MODE",
+    }
+)
+_REPOSITORY_ROOT: Final = Path(__file__).resolve().parents[2]
+
+
+def _dotenv_was_loaded(module: str, environment: Mapping[str, str]) -> bool:
+    process_environment: Final = {
+        key: value for key, value in os.environ.items() if key not in _CONTROL_ENV_VARIABLES
+    } | dict(environment)
+    script: Final = textwrap.dedent(
+        f"""
+        import os
+        import dotenv
+
+        def load_dotenv(*args, **kwargs):
+            os.environ["LITELLM_DOTENV_LOADED"] = "1"
+            return True
+
+        dotenv.load_dotenv = load_dotenv
+        import {module}
+        print(f"LITELLM_DOTENV_LOADED={{os.environ.get('LITELLM_DOTENV_LOADED', '0')}}")
+        """
+    )
+    result: Final = subprocess.run(
+        [sys.executable, "-c", script],
+        cwd=_REPOSITORY_ROOT,
+        env=process_environment,
+        capture_output=True,
+        text=True,
+        check=True,
+    )
+    marker: Final = "LITELLM_DOTENV_LOADED="
+    output_line: Final = next(line for line in result.stdout.splitlines() if line.startswith(marker))
+    return output_line.removeprefix(marker) == "1"
+
+
+def test_default_dev_mode_loads_dotenv() -> None:
+    assert _dotenv_was_loaded("litellm", {})
+
+
+@pytest.mark.parametrize("value", ("1", "true", "t", "yes", "y", "TRUE", "YeS"))
+def test_disable_dotenv_truthy_values_prevent_loading(value: str) -> None:
+    assert not _dotenv_was_loaded("litellm", {"LITELLM_DISABLE_DOTENV": value})
+
+
+def test_disable_dotenv_false_value_preserves_loading() -> None:
+    assert _dotenv_was_loaded("litellm", {"LITELLM_DISABLE_DOTENV": "false"})
+
+
+def test_production_mode_still_skips_dotenv() -> None:
+    assert not _dotenv_was_loaded("litellm", {"LITELLM_MODE": "PRODUCTION"})
+
+
+def test_disable_dotenv_wins_over_dev_hot_reload() -> None:
+    environment: Final = {
+        "LITELLM_DISABLE_DOTENV": "1",
+        "LITELLM_DEV_ENV_HOT_RELOAD": "True",
+    }
+    assert not _dotenv_was_loaded("litellm", environment)
+
+
+def test_proxy_cli_respects_disable_dotenv() -> None:
+    assert not _dotenv_was_loaded("litellm.proxy.proxy_cli", {"LITELLM_DISABLE_DOTENV": "1"})
+
+
+def test_proxy_cli_default_dev_mode_loads_dotenv() -> None:
+    assert _dotenv_was_loaded("litellm.proxy.proxy_cli", {})