diff --git a/main.py b/main.py
index c862e84..a6c3229 100644
--- a/main.py
+++ b/main.py
@@ -6,18 +6,33 @@ import time
 import os
 import base64
 
-# Simple in-memory cache
-# Format: {url: {"time": timestamp, "buffer": base64_string}}
-cache = {}
+# Filesystem-based cache for serverless environment
+CACHE_DIR = "/tmp/snapshot_cache"
 CACHE_DURATION = 10  # seconds
 
+def get_cache_path(url):
+    import hashlib
+    url_hash = hashlib.md5(url.encode('utf-8')).hexdigest()
+    return os.path.join(CACHE_DIR, f"{url_hash}.jpg")
+
 def capture_stream_snapshot(youtube_url, output_file="snapshot.jpg", toBuffer=False):
+    if not os.path.exists(CACHE_DIR):
+        os.makedirs(CACHE_DIR, exist_ok=True)
+
+    cache_path = get_cache_path(youtube_url)
+    
     # Check cache first
-    if youtube_url in cache:
-        cached_item = cache[youtube_url]
-        if time.time() - cached_item["time"] < CACHE_DURATION:
-            print(f"Returning cached snapshot for {youtube_url}")
-            return base64.b64decode(cached_item["buffer"]) if toBuffer else True
+    if os.path.exists(cache_path):
+        mtime = os.path.getmtime(cache_path)
+        if time.time() - mtime < CACHE_DURATION:
+            print(f"Returning cached snapshot for {youtube_url} from {cache_path}")
+            if toBuffer:
+                with open(cache_path, "rb") as f:
+                    return f.read()
+            else:
+                import shutil
+                shutil.copy2(cache_path, output_file)
+                return True
 
     # 1. Configure yt-dlp to get the direct stream URL
     ydl_opts = {
@@ -50,6 +65,9 @@ def capture_stream_snapshot(youtube_url, output_file="snapshot.jpg", toBuffer=Fa
 
     result = None
     if success:
+        # Save to cache first
+        cv2.imwrite(cache_path, frame)
+        
         if toBuffer:
             # Encode frame as JPEG to memory buffer
             ret, buf = cv2.imencode(".jpg", frame)
@@ -63,19 +81,6 @@ def capture_stream_snapshot(youtube_url, output_file="snapshot.jpg", toBuffer=Fa
             cv2.imwrite(output_file, frame)
             print(f"Snapshot saved to {output_file}")
             result = True
-            
-        # Update cache if successful
-        if result:
-            if toBuffer:
-                buf_to_cache = result
-            else:
-                with open(output_file, "rb") as f:
-                    buf_to_cache = f.read()
-            
-            cache[youtube_url] = {
-                "time": time.time(),
-                "buffer": base64.b64encode(buf_to_cache).decode('utf-8')
-            }
     else:
         print("Error: Could not read frame from stream.")
 
@@ -113,17 +118,21 @@ def main(args):
                 },
             }
         
-        # Check cache explicitly in main to avoid redundant toBuffer logic if possible
-        if url in cache:
-            cached_item = cache[url]
-            if time.time() - cached_item["time"] < CACHE_DURATION:
-                print(f"Serving {url} from cache.")
+        if not os.path.exists(CACHE_DIR):
+            os.makedirs(CACHE_DIR, exist_ok=True)
+            
+        cache_path = get_cache_path(url)
+        if os.path.exists(cache_path):
+            if time.time() - os.path.getmtime(cache_path) < CACHE_DURATION:
+                print(f"Serving {url} from filesystem cache.")
+                with open(cache_path, "rb") as f:
+                    encoded_content = base64.b64encode(f.read()).decode('utf-8')
                 return {
                     "_shsf": "v2",
                     "_code": 200,
                     "_res": {
                         "message": "done (cached)",
-                        "buffer": cached_item["buffer"]
+                        "buffer": encoded_content
                     },
                     "_headers": {
                         "Content-Type": "application/json",