diff --git a/.env.example b/.env.example
index ec98c23..1f30890 100644
--- a/.env.example
+++ b/.env.example
@@ -1,11 +1,24 @@
 # Smriti - Environment Variables
-# Copy this file to .env and fill in the values
+# Copy this file to .env and fill in the values.
+#
+# The backend loads this file via python-dotenv at startup so API keys
+# are available even when the server is launched by a process that does
+# not inherit your shell profile (e.g. uvicorn from an IDE, a CI
+# runner, or a subprocess-based tool host like Claude Code).
+#
+# Existing env vars are NOT overridden — if you also export a key in
+# your shell profile, the shell value takes precedence.
 
 # Database
 DATABASE_URL=postgresql://smriti:smriti@localhost:5432/smriti
 
-# OpenAI (leave empty to use mock provider)
+# Provider API keys — set the ones you want to use.
+# Leave empty to fall back to config/providers.yaml or mock provider.
 OPENAI_API_KEY=
+ANTHROPIC_API_KEY=
+OPENROUTER_API_KEY=
+
+# Default model for background intelligence (extraction, draft, review)
 OPENAI_MODEL=gpt-4o-mini
 
 # App
diff --git a/backend/app/config_loader.py b/backend/app/config_loader.py
index 6dc2933..59e1db1 100644
--- a/backend/app/config_loader.py
+++ b/backend/app/config_loader.py
@@ -1,6 +1,12 @@
 """
 Provider configuration loader.
 Priority: env vars > config/providers.yaml > built-in defaults.
+
+The .env file (at the project root, one level above backend/) is loaded
+via python-dotenv at import time so that API keys set there are visible
+to os.environ.get() even when the backend is started by a process that
+does not inherit the user's shell profile (e.g. uvicorn launched from
+an IDE, a CI runner, or a subprocess-based tool host).
 """
 from __future__ import annotations
 
@@ -8,6 +14,19 @@ import os
 from dataclasses import dataclass, field
 from pathlib import Path
 
+from dotenv import load_dotenv
+
+# Load .env from the project root (smriti/) — one level above
+# backend/. Falls back silently if the file does not exist (e.g. in
+# Docker where env vars are injected by the runtime).
+_PROJECT_ROOT_ENV = Path(__file__).parent.parent.parent / ".env"
+_BACKEND_DIR_ENV = Path(__file__).parent.parent / ".env"
+# Try project root first (where .env.example lives), then backend dir.
+if _PROJECT_ROOT_ENV.is_file():
+    load_dotenv(_PROJECT_ROOT_ENV, override=False)
+elif _BACKEND_DIR_ENV.is_file():
+    load_dotenv(_BACKEND_DIR_ENV, override=False)
+
 try:
     import yaml
     _YAML_AVAILABLE = True
diff --git a/backend/app/providers/registry.py b/backend/app/providers/registry.py
index a276b03..189c1b5 100644
--- a/backend/app/providers/registry.py
+++ b/backend/app/providers/registry.py
@@ -2,10 +2,13 @@
 from __future__ import annotations
 
 import json as _json
+import logging
 
 from app.config_loader import get_provider_config, ProviderNotConfiguredError
 from app.providers.base import ProviderAdapter
 
+logger = logging.getLogger(__name__)
+
 
 # Canned JSON response returned by the mock adapter when the caller asks
 # for JSON mode (response_format={"type": "json_object"}). Covers every
@@ -74,6 +77,13 @@ def get_adapter(provider: str, allow_mock: bool = False) -> ProviderAdapter:
         cfg = get_provider_config(p)
     except ProviderNotConfiguredError:
         if allow_mock:
+            logger.warning(
+                "Provider '%s' has no API key configured — falling back to "
+                "MockAdapter. Extraction, drafting, and review endpoints will "
+                "return deterministic mock content instead of real LLM output. "
+                "Set the API key in .env or config/providers.yaml to fix this.",
+                provider,
+            )
             return MockAdapter()
         raise
 
diff --git a/backend/pyproject.toml b/backend/pyproject.toml
index a80a7b2..74ed8a5 100644
--- a/backend/pyproject.toml
+++ b/backend/pyproject.toml
@@ -16,6 +16,7 @@ dependencies = [
     "anthropic>=0.40.0",
     "pyyaml>=6.0",
     "pgvector>=0.2.0",
+    "python-dotenv>=1.0.0",
 ]
 
 [project.optional-dependencies]