diff --git a/.env b/.env
index be271e6..0889dd0 100644
--- a/.env
+++ b/.env
@@ -27,8 +27,8 @@ IT_DEPT_EMAIL=da.nguyen8744@gmail.com
 # Application URL (used in email links)
 APP_BASE_URL=https://tickets.ltservicesinc.com
 
-# Anthropic API Key (for AI Chatbot)
-ANTHROPIC_API_KEY=your-anthropic-api-key-here
+# Gemini opic API Key (for AI Chatbot)
+GROQ_API_KEY=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
 
 # File Upload Configuration
 # Must be an absolute path so all gunicorn workers resolve the same directory
diff --git a/app/routes/chatbot.py b/app/routes/chatbot.py
index 064d51c..028fd53 100644
--- a/app/routes/chatbot.py
+++ b/app/routes/chatbot.py
@@ -29,6 +29,53 @@ Your job is to:
 7. Always ask clarifying questions if you need more detail before creating a ticket.
 """
 
+# Groq API — OpenAI-compatible, free tier, no region restrictions.
+# Free tier: 14,400 requests/day. Get a key at: https://console.groq.com
+_GROQ_API_URL = 'https://api.groq.com/openai/v1/chat/completions'
+_GROQ_MODEL   = 'llama-3.3-70b-versatile'
+
+
+def _call_groq(api_key, history, user_msg):
+    """
+    Call the Groq API (OpenAI-compatible) and return the assistant's reply text.
+
+    The system prompt is prepended as a system message. Prior history and the
+    new user message are appended in order.
+
+    Raises requests.HTTPError or requests.exceptions.RequestException on failure.
+    """
+    model = current_app.config.get('GROQ_MODEL', _GROQ_MODEL)
+
+    # Strip any assistant messages containing a create_ticket action block.
+    # These should never reach the model — if they do, the model re-triggers
+    # ticket creation on every subsequent turn.
+    clean_history = [
+        msg for msg in history
+        if not (msg.get('role') == 'assistant' and '"action": "create_ticket"' in msg.get('content', ''))
+    ]
+
+    messages = (
+        [{'role': 'system', 'content': _SYSTEM_PROMPT}]
+        + clean_history
+        + [{'role': 'user', 'content': user_msg}]
+    )
+    resp = requests.post(
+        _GROQ_API_URL,
+        headers={
+            'Authorization': f'Bearer {api_key}',
+            'Content-Type' : 'application/json',
+        },
+        json={
+            'model'      : model,
+            'messages'   : messages,
+            'max_tokens' : 1024,
+            'temperature': 0.4,
+        },
+        timeout=30,
+    )
+    resp.raise_for_status()
+    return resp.json()['choices'][0]['message']['content'].strip()
+
 
 @chatbot_bp.route('/message', methods=['POST'])
 @login_required
@@ -40,40 +87,28 @@ def chat():
     if not user_msg:
         return jsonify({'error': 'Empty message'}), 400
 
-    api_key = current_app.config.get('ANTHROPIC_API_KEY', '')
+    api_key = current_app.config.get('GROQ_API_KEY', '')
     if not api_key:
-        return jsonify({'reply': "The AI assistant is not configured yet. Please contact your IT administrator.", 'ticket': None})
-
-    messages = history + [{'role': 'user', 'content': user_msg}]
+        return jsonify({
+            'reply' : 'The AI assistant is not configured yet. Please contact your IT administrator.',
+            'ticket': None,
+        })
 
     try:
-        resp = requests.post(
-            'https://api.anthropic.com/v1/messages',
-            headers={
-                'x-api-key'        : api_key,
-                'anthropic-version': '2023-06-01',
-                'content-type'     : 'application/json',
-            },
-            json={
-                'model'     : 'claude-sonnet-4-20250514',
-                'max_tokens': 1024,
-                'system'    : _SYSTEM_PROMPT,
-                'messages'  : messages,
-            },
-            timeout=30,
-        )
-        resp.raise_for_status()
-        reply_text = resp.json()['content'][0]['text'].strip()
+        reply_text = _call_groq(api_key, history, user_msg)
     except Exception as exc:
         logger.error(f'[CHATBOT API ERROR] {exc}')
-        return jsonify({'reply': 'Sorry, I encountered an error. Please try again or submit a ticket manually.', 'ticket': None})
+        return jsonify({
+            'reply' : 'Sorry, I encountered an error. Please try again or submit a ticket manually.',
+            'ticket': None,
+        })
 
     # Check if the AI wants to create a ticket
     ticket_data = None
     if '"action": "create_ticket"' in reply_text or "'action': 'create_ticket'" in reply_text:
         try:
-            start = reply_text.find('{')
-            end   = reply_text.rfind('}') + 1
+            start  = reply_text.find('{')
+            end    = reply_text.rfind('}') + 1
             parsed = json.loads(reply_text[start:end])
             if parsed.get('action') == 'create_ticket':
                 # Validate AI-provided enum values against allowed sets to
@@ -89,8 +124,8 @@ def chat():
                     TicketPriority.LOW,    TicketPriority.MEDIUM,
                     TicketPriority.HIGH,   TicketPriority.CRITICAL,
                 }
-                raw_category = parsed.get('category', TicketCategory.OTHER)
-                raw_priority = parsed.get('priority', TicketPriority.MEDIUM)
+                raw_category  = parsed.get('category', TicketCategory.OTHER)
+                raw_priority  = parsed.get('priority', TicketPriority.MEDIUM)
                 safe_category = raw_category if raw_category in _valid_categories else TicketCategory.OTHER
                 safe_priority = raw_priority if raw_priority in _valid_priorities else TicketPriority.MEDIUM
                 if raw_category != safe_category:
@@ -118,10 +153,10 @@ def chat():
                 notify_new_ticket(ticket)
 
                 ticket_data = {
-                    'id'            : ticket.id,
-                    'ticket_number' : ticket.ticket_number,
-                    'title'         : ticket.title,
-                    'url'           : f'/tickets/{ticket.id}',
+                    'id'           : ticket.id,
+                    'ticket_number': ticket.ticket_number,
+                    'title'        : ticket.title,
+                    'url'          : f'/tickets/{ticket.id}',
                 }
                 reply_text = (
                     f"✅ **Ticket Created!**\n\n"
@@ -132,4 +167,4 @@ def chat():
         except (json.JSONDecodeError, KeyError) as exc:
             logger.warning(f'[CHATBOT PARSE ERROR] Could not parse ticket JSON: {exc}')
 
-    return jsonify({'reply': reply_text, 'ticket': ticket_data})
+    return jsonify({'reply': reply_text, 'ticket': ticket_data})
\ No newline at end of file
diff --git a/app/templates/base.html b/app/templates/base.html
index bde97a5..c0ddfca 100644
--- a/app/templates/base.html
+++ b/app/templates/base.html
@@ -628,13 +628,18 @@ async function sendChat(){
     typing.remove();
     const reply = d.reply || 'Sorry, I encountered an error.';
     appendBubble(reply,'bot');
-    chatHistory.push({role:'assistant',content:reply});
     if(d.ticket){
+      // Clear history after ticket creation — keeping the prior conversation
+      // (which contains the AI's JSON action block) causes the model to
+      // re-trigger ticket creation on every subsequent message.
+      chatHistory = [];
       const notif=document.createElement('div');
       notif.className='chat-bubble bot';
       notif.style.cssText='background:rgba(45,212,191,.1);border:1px solid rgba(45,212,191,.3);';
       notif.innerHTML=`🎫 <strong>${d.ticket.ticket_number}</strong> — <a href="${d.ticket.url}">View Ticket</a>`;
       document.getElementById('chat-messages').appendChild(notif);
+    } else {
+      chatHistory.push({role:'assistant',content:reply});
     }
   }catch(e){typing.remove();appendBubble('Sorry, something went wrong.','bot');}
   scrollChat();
diff --git a/app/templates/tickets/notifications.html b/app/templates/tickets/notifications.html
index 499b748..c32ede5 100644
--- a/app/templates/tickets/notifications.html
+++ b/app/templates/tickets/notifications.html
@@ -19,11 +19,14 @@
         {% if notifs.items %}
           {% for n in notifs.items %}
           <a href="{{ n.link or '#' }}"
+             data-notif-id="{{ n.id }}"
+             data-is-read="{{ 'true' if n.is_read else 'false' }}"
+             class="notif-list-item"
              style="display:flex;gap:14px;padding:16px 20px;border-bottom:1px solid var(--border);color:var(--text);
                     {% if not n.is_read %}border-left:3px solid var(--accent3);background:rgba(0,180,216,.03);{% endif %}">
             <div style="margin-top:2px;flex-shrink:0;">
               {% if not n.is_read %}
-                <div style="width:8px;height:8px;border-radius:50%;background:var(--accent3);margin-top:3px;"></div>
+                <div class="notif-unread-dot" style="width:8px;height:8px;border-radius:50%;background:var(--accent3);margin-top:3px;"></div>
               {% else %}
                 <i class="bi bi-check2" style="color:var(--muted);font-size:14px;"></i>
               {% endif %}
@@ -78,4 +81,45 @@
     </div>
   </div>
 </div>
-{% endblock %}
+
+<script>
+(function(){
+  // On page load, fetch the live unread count and sync the bell badge.
+  // The server-rendered {{ unread_notifications }} is only accurate at render
+  // time — if the user has been clicking around, it can be stale.
+  fetch('/api/notifications/unread-count')
+    .then(r => r.json())
+    .then(d => { if(typeof updateBadge === 'function') updateBadge(d.count); })
+    .catch(() => {});
+
+  // Intercept every notification link click:
+  // 1. If the notification is unread, call the mark-read API.
+  // 2. Decrement the bell badge immediately (optimistic update).
+  // 3. Navigate to the destination.
+  document.querySelectorAll('a.notif-list-item').forEach(function(link){
+    link.addEventListener('click', async function(e){
+      const id     = this.dataset.notifId;
+      const isRead = this.dataset.isRead === 'true';
+      const dest   = this.getAttribute('href');
+
+      if(!isRead && id){
+        e.preventDefault();
+        try {
+          await csrfPost('/api/notifications/' + id + '/read');
+        } catch(_){}
+        // Update badge — decrement by 1
+        if(typeof updateBadge === 'function') updateBadge(-1, true);
+        // Mark the row as visually read
+        this.dataset.isRead = 'true';
+        this.style.borderLeft = '';
+        this.style.background = '';
+        const dot = this.querySelector('.notif-unread-dot');
+        if(dot) dot.remove();
+      }
+
+      if(dest && dest !== '#') window.location.href = dest;
+    });
+  });
+})();
+</script>
+{% endblock %}
\ No newline at end of file
diff --git a/config/config.py b/config/config.py
index 44c3380..4098296 100644
--- a/config/config.py
+++ b/config/config.py
@@ -1,7 +1,10 @@
 import os
 from dotenv import load_dotenv
 
-load_dotenv()
+# Use an absolute path so .env is found regardless of the process working
+# directory — under systemd the CWD may not be the project root.
+_here = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+load_dotenv(os.path.join(_here, '.env'))
 
 class Config:
     SECRET_KEY = os.environ.get('SECRET_KEY') or 'dev-secret-key-change-in-production'
@@ -44,8 +47,10 @@ class Config:
     MAX_CONTENT_LENGTH = int(os.environ.get('MAX_CONTENT_LENGTH', 16 * 1024 * 1024))
     ALLOWED_EXTENSIONS = {'png', 'jpg', 'jpeg', 'gif', 'pdf', 'doc', 'docx', 'txt', 'zip', 'log'}
 
-    # Anthropic AI
-    ANTHROPIC_API_KEY = os.environ.get('ANTHROPIC_API_KEY', '')
+    # AI Chatbot — Groq (free tier: 14,400 req/day, no credit card, no region restrictions)
+    # Get a key at: https://console.groq.com → API Keys
+    GROQ_API_KEY = os.environ.get('GROQ_API_KEY', '')
+    GROQ_MODEL   = os.environ.get('GROQ_MODEL', 'llama-3.3-70b-versatile')
 
     # Rate limiting storage.
     # The `limits` library (used by Flask-Limiter) does not support MySQL as a