diff --git a/server/.env.example b/server/.env.example
index 83058852..853ee50a 100644
--- a/server/.env.example
+++ b/server/.env.example
@@ -10,6 +10,13 @@ MAX_MESSAGES_PER_SESSION=5
 # Google Authentication
 GOOGLE_CLIENT_ID=your-client-id
 
+# Ollama base URL — server-side default when no ollama_url is sent in the request.
+# For local dev this is fine. For deployed instances (e.g. Vercel), users must pass
+# their own ollama_url in each API request (e.g. an ngrok tunnel to their local machine),
+# because the server cannot reach a user's localhost.
+# Example for a self-hosted or Docker instance: OLLAMA_BASE_URL="http://192.168.1.100:11434"
+OLLAMA_BASE_URL="http://localhost:11434"
+
 # Plugin Architecture Config
 # PLUGINS_DIRS="path/to/plugins1,path/to/plugins2" # Optional: Override the default plugins locations
 # ENABLED_PLUGINS="example_plugin,another_plugin" # Optional: Comma-separated list of plugins to enable
\ No newline at end of file
diff --git a/server/api/config.py b/server/api/config.py
index 351c7aec..8771efbd 100644
--- a/server/api/config.py
+++ b/server/api/config.py
@@ -27,3 +27,4 @@ class Config:
     ENABLED_PLUGINS = os.getenv("ENABLED_PLUGINS", None) # Comma-separated list of plugin folder names
     MAX_MESSAGES_PER_SESSION = int(os.getenv("MAX_MESSAGES_PER_SESSION", 10))
     CONTACT_EMAIL = os.getenv("CONTACT_EMAIL", "support@privgpt-studio.com")
+    OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://localhost:11434")
diff --git a/server/api/routes/chat_routes.py b/server/api/routes/chat_routes.py
index 43fe2f12..5e632cc7 100644
--- a/server/api/routes/chat_routes.py
+++ b/server/api/routes/chat_routes.py
@@ -164,11 +164,14 @@ def chat():
         model_name = request.form.get("model_name", "")
         session_id = request.form.get("session_id", "1")
         session_name = request.form.get("session_name", "")
-        
+        # User-supplied Ollama URL (e.g. an ngrok tunnel) lets deployed instances
+        # proxy to the user's local Ollama without changing the server default.
+        ollama_url = request.form.get("ollama_url") or Config.OLLAMA_BASE_URL
+
         # Plugin: before_prompt
         if plugin_manager:
             user_msg = plugin_manager.before_prompt(user_msg)
-            
+
         user_timestamp = datetime.now() - timedelta(seconds=10)
         session_id = request.form.get("session_id", "1")
 
@@ -286,7 +289,7 @@ def chat():
                 payload["system"] = system_prompt
             try:
                 latency_ms = datetime.now()
-                response = requests.post("http://localhost:11434/api/generate", json=payload, timeout=60)
+                response = requests.post(f"{ollama_url}/api/generate", json=payload, timeout=60)
                 latency_ms = int((datetime.now() - latency_ms).total_seconds() * 1000)
                 bot_reply = response.json().get("response", "No reply.")
                 
@@ -429,11 +432,14 @@ def chat_stream():
         model_name = request.form.get("model_name", "")
         session_id = request.form.get("session_id", "1")
         session_name = request.form.get("session_name", "")
+        # User-supplied Ollama URL (e.g. an ngrok tunnel) lets deployed instances
+        # proxy to the user's local Ollama without changing the server default.
+        ollama_url = request.form.get("ollama_url") or Config.OLLAMA_BASE_URL
 
         # Plugin: before_prompt
         if plugin_manager:
             user_msg = plugin_manager.before_prompt(user_msg)
-            
+
         if is_session_locked(session_id):
             def error_generator_locked():
                 err_msg = "This chat is locked and cannot receive new messages."
@@ -549,7 +555,7 @@ def generate_stream():
                             payload["options"]["seed"] = seed
                         if system_prompt:
                             payload["system"] = system_prompt
-                        response = requests.post("http://localhost:11434/api/generate", json=payload, stream=True, timeout=60)
+                        response = requests.post(f"{ollama_url}/api/generate", json=payload, stream=True, timeout=60)
                         response.raise_for_status()
                         
                         for line in response.iter_lines():
diff --git a/server/api/routes/model_routes.py b/server/api/routes/model_routes.py
index 30bdb852..8f990b64 100644
--- a/server/api/routes/model_routes.py
+++ b/server/api/routes/model_routes.py
@@ -8,11 +8,14 @@ def models():
     """
     Returns available local and cloud models.
 
+    Accepts optional query param ?ollama_url= so deployed frontends can
+    proxy through a user-supplied Ollama address (e.g. an ngrok tunnel).
+
     Returns:
     JSON: Dictionary with local_models and cloud_models keys.
     """
-
-    local_models = get_available_models()
+    ollama_url = request.args.get("ollama_url") or None
+    local_models = get_available_models(ollama_url=ollama_url)
     cloud_models = ["gemini"]
     return jsonify({
         "local_models": local_models,
@@ -23,7 +26,7 @@ def models():
 def model_info():
     """
     Returns detailed information about a specific model.
-    Expects JSON: { "model_name": "name", "model_type": "local"|"cloud" }
+    Expects JSON: { "model_name": "name", "model_type": "local"|"cloud", "ollama_url": "..." (optional) }
     """
     data = request.json
     if not data:
@@ -31,6 +34,7 @@ def model_info():
 
     model_name = data.get("model_name")
     model_type = data.get("model_type", "local")
+    ollama_url = data.get("ollama_url") or None
 
     if not model_name:
         return jsonify({"error": "Model name is required"}), 400
@@ -51,10 +55,10 @@ def model_info():
             }
         })
 
-    details = get_model_details(model_name)
+    details = get_model_details(model_name, ollama_url=ollama_url)
     if details:
         return jsonify(details)
-    
+
     return jsonify({"error": "Failed to fetch model info"}), 500
 
 select_model_bp = Blueprint('select_model_bp', __name__)
diff --git a/server/api/services/ollama_services.py b/server/api/services/ollama_services.py
index 26230344..568eaa95 100644
--- a/server/api/services/ollama_services.py
+++ b/server/api/services/ollama_services.py
@@ -1,31 +1,38 @@
 import requests
+from api.config import Config
 
-def get_available_models():
+def get_available_models(ollama_url=None):
     """
     Fetches list of available local models from Ollama.
 
+    Args:
+    ollama_url (str, optional): Custom Ollama base URL. Defaults to Config.OLLAMA_BASE_URL.
+
     Returns:
     list: Names of available local models (with full tags).
     """
+    base_url = ollama_url or Config.OLLAMA_BASE_URL
     try:
-        res = requests.get("http://localhost:11434/api/tags", timeout=5)
+        res = requests.get(f"{base_url}/api/tags", timeout=5)
         # Return full model names including tags (e.g., "gemma3:1b" instead of just "gemma3")
         return sorted(m['name'] for m in res.json().get("models", []))
     except:
         return []
 
-def get_model_details(model_name):
+def get_model_details(model_name, ollama_url=None):
     """
     Fetches detailed information for a specific local model from Ollama.
 
     Args:
     model_name (str): The name of the model to inspect.
+    ollama_url (str, optional): Custom Ollama base URL. Defaults to Config.OLLAMA_BASE_URL.
 
     Returns:
     dict: The JSON response from Ollama's /api/show endpoint, or None if failed.
     """
+    base_url = ollama_url or Config.OLLAMA_BASE_URL
     try:
-        res = requests.post("http://localhost:11434/api/show", json={"name": model_name}, timeout=5)
+        res = requests.post(f"{base_url}/api/show", json={"name": model_name}, timeout=5)
         if res.status_code == 200:
             return res.json()
         return None