chore: enforce dotenv override and warn on key conflict (google#282)

Enforces dotenv override in entry points and warns on conflicting API keys.

Author: aksg87

benchmarks/benchmark.py

@@ -45,6 +45,8 @@
 from typing import Any
 import urllib.error
 
+import dotenv
+
 from benchmarks import config
 from benchmarks import plotting
 from benchmarks import utils
@@ -55,6 +57,7 @@
 import langextract.io as lio
 
 # Load API key from environment
+dotenv.load_dotenv(override=True)
 GEMINI_API_KEY = os.environ.get(
     "GEMINI_API_KEY", os.environ.get("LANGEXTRACT_API_KEY")
 )

examples/custom_provider_plugin/test_example_provider.py

@@ -17,6 +17,7 @@
 
 import os
 
+import dotenv
 # Import the provider to trigger registration with LangExtract
 # Note: This manual import is only needed when running without installation.
 # After `pip install -e .`, the entry point system handles this automatically.
@@ -27,6 +28,7 @@
 
 def main():
   """Test the custom provider."""
+  dotenv.load_dotenv(override=True)
   api_key = os.getenv("GEMINI_API_KEY") or os.getenv("LANGEXTRACT_API_KEY")
 
   if not api_key:

examples/ollama/demo_ollama.py

@@ -52,9 +52,13 @@
 import urllib.error
 import urllib.request
 
+import dotenv
+
 import langextract as lx
 from langextract.providers import ollama
 
+dotenv.load_dotenv(override=True)
+
 DEFAULT_MODEL = "gemma2:2b"
 DEFAULT_OLLAMA_URL = os.environ.get("OLLAMA_HOST", "http://localhost:11434")
 OUTPUT_DIR = "test_output"

langextract/factory.py

@@ -24,6 +24,7 @@
 import dataclasses
 import os
 import typing
+import warnings
 
 from langextract import providers
 from langextract.core import base_model
@@ -72,11 +73,23 @@ def _kwargs_with_environment_defaults(
 
     for provider_prefix, env_vars in env_vars_by_provider.items():
       if provider_prefix in model_lower:
+        found_keys = []
         for env_var in env_vars:
-          api_key = os.getenv(env_var)
-          if api_key:
-            resolved["api_key"] = api_key
-            break
+          key_val = os.getenv(env_var)
+          if key_val:
+            found_keys.append((env_var, key_val))
+
+        if found_keys:
+          resolved["api_key"] = found_keys[0][1]
+
+          if len(found_keys) > 1:
+            keys_list = ", ".join(k[0] for k in found_keys)
+            warnings.warn(
+                f"Multiple API keys detected in environment: {keys_list}. "
+                f"Using {found_keys[0][0]} and ignoring others.",
+                UserWarning,
+                stacklevel=3,
+            )
         break
 
   if "ollama" in model_id.lower() and "base_url" not in resolved:
@@ -126,7 +139,6 @@ def create_model(
   if not config.model_id and not config.provider:
     raise ValueError("Either model_id or provider must be specified")
 
-  # Load providers before any resolution
   providers.load_builtins_once()
   providers.load_plugins_once()
 
@@ -144,6 +156,8 @@ def create_model(
 
   model_id = config.model_id
 
+  model_id = config.model_id
+
   kwargs = _kwargs_with_environment_defaults(
       model_id or config.provider or "", config.provider_kwargs
   )

tests/test_live_api.py

@@ -38,7 +38,7 @@
 import langextract as lx
 from langextract.providers import gemini_batch as gb
 
-dotenv.load_dotenv()
+dotenv.load_dotenv(override=True)
 
 DEFAULT_GEMINI_MODEL = "gemini-2.5-flash"
 DEFAULT_OPENAI_MODEL = "gpt-4o"