diff --git a/grafana-configs/loki_logger.py b/grafana-configs/loki_logger.py
index 59a901b..a86a969 100644
--- a/grafana-configs/loki_logger.py
+++ b/grafana-configs/loki_logger.py
@@ -5,6 +5,7 @@
 """
 
 import json
+import sys
 import time
 from datetime import datetime
 from threading import Thread
@@ -111,9 +112,12 @@ def _send_to_loki_sync(self, level: str, message: str) -> None:
 
     def _log(self, level: str, message: str) -> None:
         """Queue log entry for async processing (non-blocking)"""
-        # Print to console immediately for real-time feedback
+        # Print to console immediately for real-time feedback. Written to
+        # stderr (not stdout) so callers that capture a subprocess's stdout
+        # for its return value (e.g. decrypt_vault_secrets.py) never pick up
+        # log lines mixed in with the actual output.
         timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-        print(f"[{timestamp}] {level: <8} | {message}")  # noqa: T201
+        print(f"[{timestamp}] {level: <8} | {message}", file=sys.stderr)  # noqa: T201
 
         # Queue for async Loki sending (non-blocking)
         try:
diff --git a/src/loki_logger.py b/src/loki_logger.py
index b70be5b..d780fce 100644
--- a/src/loki_logger.py
+++ b/src/loki_logger.py
@@ -13,6 +13,7 @@
 """
 
 import json
+import sys
 import time
 from datetime import datetime
 from threading import Thread
@@ -119,9 +120,12 @@ def _send_to_loki_sync(self, level: str, message: str) -> None:
 
     def _log(self, level: str, message: str) -> None:
         """Queue log entry for async processing (non-blocking)"""
-        # Print to console immediately for real-time feedback
+        # Print to console immediately for real-time feedback. Written to
+        # stderr (not stdout) so callers that capture a subprocess's stdout
+        # for its return value (e.g. decrypt_vault_secrets.py) never pick up
+        # log lines mixed in with the actual output.
         timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-        print(f"[{timestamp}] {level: <8} | {message}")  # noqa: T201
+        print(f"[{timestamp}] {level: <8} | {message}", file=sys.stderr)  # noqa: T201
 
         # Queue for async Loki sending (non-blocking, drops log if queue is full)
         try:
diff --git a/src/vector_indexer/loki_logger.py b/src/vector_indexer/loki_logger.py
index 59a901b..a86a969 100644
--- a/src/vector_indexer/loki_logger.py
+++ b/src/vector_indexer/loki_logger.py
@@ -5,6 +5,7 @@
 """
 
 import json
+import sys
 import time
 from datetime import datetime
 from threading import Thread
@@ -111,9 +112,12 @@ def _send_to_loki_sync(self, level: str, message: str) -> None:
 
     def _log(self, level: str, message: str) -> None:
         """Queue log entry for async processing (non-blocking)"""
-        # Print to console immediately for real-time feedback
+        # Print to console immediately for real-time feedback. Written to
+        # stderr (not stdout) so callers that capture a subprocess's stdout
+        # for its return value (e.g. decrypt_vault_secrets.py) never pick up
+        # log lines mixed in with the actual output.
         timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-        print(f"[{timestamp}] {level: <8} | {message}")  # noqa: T201
+        print(f"[{timestamp}] {level: <8} | {message}", file=sys.stderr)  # noqa: T201
 
         # Queue for async Loki sending (non-blocking)
         try: