diff --git a/chutes/chute/template/embedding.py b/chutes/chute/template/embedding.py
index c9fa9e0..c131c8a 100644
--- a/chutes/chute/template/embedding.py
+++ b/chutes/chute/template/embedding.py
@@ -259,6 +259,43 @@ async def monitor_subprocess():
                 pass
             await asyncio.sleep(1)
 
+    @chute.on_event("shutdown")
+    async def cleanup_vllm(self):
+        """
+        Cleanup vLLM embedding subprocess and monitor task on shutdown.
+        """
+        logger.info("Cleaning up vLLM embedding subprocess...")
+        try:
+            # Cancel the monitor task
+            if hasattr(self, "_monitor_task") and self._monitor_task:
+                self._monitor_task.cancel()
+                try:
+                    await self._monitor_task
+                except asyncio.CancelledError:
+                    pass
+
+            # Terminate the subprocess gracefully
+            if hasattr(self, "_vllm_process") and self._vllm_process:
+                try:
+                    self._vllm_process.terminate()
+                    # Wait up to 5 seconds for graceful shutdown
+                    try:
+                        self._vllm_process.wait(timeout=5)
+                        logger.info("vLLM embedding subprocess terminated gracefully")
+                    except subprocess.TimeoutExpired:
+                        # Force kill if it doesn't terminate
+                        logger.warning("vLLM embedding subprocess did not terminate, forcing kill")
+                        self._vllm_process.kill()
+                        self._vllm_process.wait()
+                        logger.info("vLLM embedding subprocess force killed")
+                except ProcessLookupError:
+                    # Process already terminated
+                    logger.debug("vLLM embedding subprocess already terminated")
+                except Exception as e:
+                    logger.error(f"Error cleaning up vLLM embedding subprocess: {e}")
+        except Exception as e:
+            logger.error(f"Unexpected error during vLLM embedding cleanup: {e}")
+
         self.passthrough_headers["Authorization"] = f"Bearer {api_key}"
         logger.info("✅ Embedding server initialized successfully!")
 
diff --git a/chutes/chute/template/sglang.py b/chutes/chute/template/sglang.py
index d04f775..060cab1 100644
--- a/chutes/chute/template/sglang.py
+++ b/chutes/chute/template/sglang.py
@@ -3,6 +3,7 @@
 import os
 import re
 import sys
+import subprocess
 import uuid
 from enum import Enum
 from loguru import logger
@@ -371,6 +372,43 @@ async def monitor_subprocess():
         await warmup_model(self, api_key=api_key)
         await validate_auth(self, api_key=api_key)
 
+    @chute.on_event("shutdown")
+    async def cleanup_sglang(self):
+        """
+        Cleanup SGLang subprocess and monitor task on shutdown.
+        """
+        logger.info("Cleaning up SGLang subprocess...")
+        try:
+            # Cancel the monitor task
+            if hasattr(self, "_monitor_task") and self._monitor_task:
+                self._monitor_task.cancel()
+                try:
+                    await self._monitor_task
+                except asyncio.CancelledError:
+                    pass
+
+            # Terminate the subprocess gracefully
+            if hasattr(self, "_sglang_process") and self._sglang_process:
+                try:
+                    self._sglang_process.terminate()
+                    # Wait up to 5 seconds for graceful shutdown
+                    try:
+                        self._sglang_process.wait(timeout=5)
+                        logger.info("SGLang subprocess terminated gracefully")
+                    except subprocess.TimeoutExpired:
+                        # Force kill if it doesn't terminate
+                        logger.warning("SGLang subprocess did not terminate, forcing kill")
+                        self._sglang_process.kill()
+                        self._sglang_process.wait()
+                        logger.info("SGLang subprocess force killed")
+                except ProcessLookupError:
+                    # Process already terminated
+                    logger.debug("SGLang subprocess already terminated")
+                except Exception as e:
+                    logger.error(f"Error cleaning up SGLang subprocess: {e}")
+        except Exception as e:
+            logger.error(f"Unexpected error during SGLang cleanup: {e}")
+
     def _parse_stream_chunk(encoded_chunk):
         chunk = encoded_chunk if isinstance(encoded_chunk, str) else encoded_chunk.decode()
         if "data: {" in chunk:
diff --git a/chutes/chute/template/vllm.py b/chutes/chute/template/vllm.py
index 8c22155..fd26dce 100644
--- a/chutes/chute/template/vllm.py
+++ b/chutes/chute/template/vllm.py
@@ -413,6 +413,43 @@ async def monitor_subprocess():
         await validate_auth(self, base_url="http://127.0.0.1:10101", api_key=api_key)
         logger.info("✅ vLLM server warmed up and ready to roll!")
 
+    @chute.on_event("shutdown")
+    async def cleanup_vllm(self):
+        """
+        Cleanup vLLM subprocess and monitor task on shutdown.
+        """
+        logger.info("Cleaning up vLLM subprocess...")
+        try:
+            # Cancel the monitor task
+            if hasattr(self, "_monitor_task") and self._monitor_task:
+                self._monitor_task.cancel()
+                try:
+                    await self._monitor_task
+                except asyncio.CancelledError:
+                    pass
+
+            # Terminate the subprocess gracefully
+            if hasattr(self, "_vllm_process") and self._vllm_process:
+                try:
+                    self._vllm_process.terminate()
+                    # Wait up to 5 seconds for graceful shutdown
+                    try:
+                        self._vllm_process.wait(timeout=5)
+                        logger.info("vLLM subprocess terminated gracefully")
+                    except subprocess.TimeoutExpired:
+                        # Force kill if it doesn't terminate
+                        logger.warning("vLLM subprocess did not terminate, forcing kill")
+                        self._vllm_process.kill()
+                        self._vllm_process.wait()
+                        logger.info("vLLM subprocess force killed")
+                except ProcessLookupError:
+                    # Process already terminated
+                    logger.debug("vLLM subprocess already terminated")
+                except Exception as e:
+                    logger.error(f"Error cleaning up vLLM subprocess: {e}")
+        except Exception as e:
+            logger.error(f"Unexpected error during vLLM cleanup: {e}")
+
     def _parse_stream_chunk(encoded_chunk):
         chunk = encoded_chunk if isinstance(encoded_chunk, str) else encoded_chunk.decode()
         if "data: {" in chunk:
diff --git a/chutes/entrypoint/ssh.py b/chutes/entrypoint/ssh.py
index d2d72b9..56ef5ee 100644
--- a/chutes/entrypoint/ssh.py
+++ b/chutes/entrypoint/ssh.py
@@ -61,8 +61,9 @@ async def setup_ssh_access(ssh_public_key):
                 ],
                 check=True,
             )
-        subprocess.Popen(["/usr/sbin/sshd", "-D", "-f", sshd_config_path])
+        sshd_process = subprocess.Popen(["/usr/sbin/sshd", "-D", "-f", sshd_config_path])
         logger.info(f"SSH server started successfully on port 2202 for user {username}")
+        return sshd_process
 
     except Exception as e:
         logger.error(f"Failed to setup SSH access: {e}")