update app

app.py

@@ -192,6 +192,18 @@ class FaraLocalModel(Model):
 # SELENIUM CHROME SANDBOX
 # -----------------------------------------------------------------------------
 
+def get_system_chrome_path():
+    # Common paths for chromium in Linux/HF Spaces
+    paths = [
+        "/usr/bin/chromium",
+        "/usr/bin/chromium-browser",
+        "/usr/bin/google-chrome",
+    ]
+    for p in paths:
+        if os.path.exists(p):
+            return p
+    return None
+
 class SeleniumSandbox:
     def __init__(self, width=1024, height=768):
         self.width = width
@@ -200,28 +212,37 @@ class SeleniumSandbox:
         
         # Setup Chrome Options
         chrome_opts = ChromeOptions()
-        chrome_opts.add_argument("--headless=new") # Run headless for Space compatibility
+        
+        # Use system binary if available (fixes status 127 in HF Spaces)
+        binary_path = get_system_chrome_path()
+        if binary_path:
+            print(f"Using system Chrome binary at: {binary_path}")
+            chrome_opts.binary_location = binary_path
+        
+        chrome_opts.add_argument("--headless=new") 
         chrome_opts.add_argument(f"--user-data-dir={self.tmp_dir}")
         chrome_opts.add_argument(f"--window-size={width},{height}")
-        chrome_opts.add_argument("--no-sandbox") 
-        chrome_opts.add_argument("--disable-dev-shm-usage")
-        chrome_opts.add_argument("--disable-extensions")
+        chrome_opts.add_argument("--no-sandbox") # Crucial for containers
+        chrome_opts.add_argument("--disable-dev-shm-usage") # Crucial for containers
         chrome_opts.add_argument("--disable-gpu")
-        chrome_opts.add_argument("--force-device-scale-factor=1")
-        chrome_opts.add_argument("--hide-scrollbars")
+        chrome_opts.add_argument("--disable-extensions")
         
         # Initialize Driver
         try:
-            self.driver = webdriver.Chrome(
-                service=ChromeService(ChromeDriverManager().install()), 
-                options=chrome_opts
-            )
-            # Ensure precise viewport size (excluding chrome UI overhead if any)
-            self.driver.set_window_size(width, height)
+            # Check for system driver first
+            system_driver_path = "/usr/bin/chromedriver"
+            if os.path.exists(system_driver_path):
+                print(f"Using system ChromeDriver at: {system_driver_path}")
+                service = ChromeService(executable_path=system_driver_path)
+            else:
+                print("Using webdriver_manager to install ChromeDriver...")
+                service = ChromeService(ChromeDriverManager().install())
             
-            # Start with a blank page
+            self.driver = webdriver.Chrome(service=service, options=chrome_opts)
+            self.driver.set_window_size(width, height)
             self.driver.get("about:blank")
-            print(f"Selenium Chrome Driver started. Data dir: {self.tmp_dir}")
+            print(f"Selenium Chrome Driver started successfully.")
+            
         except Exception as e:
             print(f"Failed to initialize Selenium: {e}")
             self.cleanup()
@@ -233,25 +254,17 @@ class SeleniumSandbox:
         return Image.open(BytesIO(png_data))
 
     def move_mouse_and_click(self, x, y, click_type="left"):
-        """
-        Simulate mouse movement and click using ActionChains.
-        Note: Selenium tracks state, so we move relative to the 'body' tag to ensure absolute positioning simulation.
-        """
         try:
             body = self.driver.find_element(By.TAG_NAME, "body")
             actions = ActionChains(self.driver)
-            
-            # Move to 0,0 of body, then offset
             actions.move_to_element_with_offset(body, 0, 0)
             actions.move_by_offset(x, y)
-            
             if click_type == "left":
                 actions.click()
             elif click_type == "right":
                 actions.context_click()
             elif click_type == "double":
                 actions.double_click()
-                
             actions.perform()
         except Exception as e:
             print(f"Error in move_mouse_and_click: {e}")
@@ -260,12 +273,9 @@ class SeleniumSandbox:
         try:
             body = self.driver.find_element(By.TAG_NAME, "body")
             actions = ActionChains(self.driver)
-            
             actions.move_to_element_with_offset(body, 0, 0)
             actions.move_by_offset(x1, y1)
             actions.click_and_hold()
-            
-            # Move relative from x1,y1 to x2,y2
             actions.move_by_offset(x2 - x1, y2 - y1)
             actions.release()
             actions.perform()
@@ -281,13 +291,11 @@ class SeleniumSandbox:
         try:
             k = getattr(Keys, key_name.upper(), None)
             if not k:
-                # Handle common overrides
                 if key_name.lower() == "enter": k = Keys.ENTER
                 elif key_name.lower() == "space": k = Keys.SPACE
                 elif key_name.lower() == "backspace": k = Keys.BACK_SPACE
                 elif key_name.lower() == "esc": k = Keys.ESCAPE
-                else: k = key_name # Fallback to literal
-
+                else: k = key_name 
             actions = ActionChains(self.driver)
             actions.send_keys(k)
             actions.perform()
@@ -295,9 +303,8 @@ class SeleniumSandbox:
             print(f"Error pressing key: {e}")
 
     def scroll(self, amount, direction="down"):
-        # Selenium doesn't have a great scroll wheel primitive, use JS
         try:
-            scroll_y = amount * 100 # Arbitrary multiplier to match "notches"
+            scroll_y = amount * 100 
             if direction == "up":
                 scroll_y = -scroll_y
             self.driver.execute_script(f"window.scrollBy(0, {scroll_y});")
@@ -604,7 +611,6 @@ def save_final_status(folder, status: str, summary, error_message=None) -> None:
 # UI & APP
 # -----------------------------------------------------------------------------
 
-# CSS
 custom_css = """
 .modal-container { margin: var(--size-16) auto!important; }
 .browser-container { position: relative; width: 100%; height: 600px; border: 1px solid #444; background: #222; display: flex; align-items: center; justify-content: center; overflow: hidden; }
@@ -612,13 +618,6 @@ custom_css = """
 #chatbot { height: 800px!important; }
 """
 
-# HTML Template for the output area (Replaced IFrame with simple image holder logic handled by Gradio Image)
-browser_html = """
-<div class="browser-container">
-    <p style="color: #888;">Browser Screenshot will appear here after steps.</p>
-</div>
-"""
-
 class EnrichedGradioUI(GradioUI):
     def interact_with_agent(
         self,
@@ -632,7 +631,6 @@ class EnrichedGradioUI(GradioUI):
         interaction_id = generate_interaction_id(session_uuid)
         data_dir = os.path.join(TMP_DIR, interaction_id)
         
-        # Initialize Sandbox per run (clean slate)
         sandbox = SeleniumSandbox(width=WIDTH, height=HEIGHT)
         agent = create_agent(data_dir=data_dir, sandbox=sandbox)
         session_state["agent"] = agent
@@ -641,19 +639,15 @@ class EnrichedGradioUI(GradioUI):
             stored_messages.append(gr.ChatMessage(role="user", content=task_input))
             yield stored_messages, None
 
-            # Initial screenshot
             screenshot = sandbox.get_screenshot()
             
-            # Run Agent
             for msg in stream_to_gradio(
                 agent,
                 task=task_input,
                 task_images=[screenshot],
                 reset_agent_memory=False,
             ):
-                # Update Chat
                 if hasattr(agent, "last_marked_screenshot") and msg.content == "-----":
-                     # Add image to chat
                     stored_messages.append(
                         gr.ChatMessage(
                             role="assistant",
@@ -663,11 +657,10 @@ class EnrichedGradioUI(GradioUI):
                             },
                         )
                     )
-                    # Yield updated chat AND the latest screenshot for the side panel
                     yield stored_messages, agent.last_marked_screenshot.to_string()
                 else:
                     stored_messages.append(msg)
-                    yield stored_messages, None # Keep previous side image
+                    yield stored_messages, None
 
             if consent_storage:
                 summary = get_agent_summary_erase_images(agent)
@@ -683,10 +676,8 @@ class EnrichedGradioUI(GradioUI):
             )
             yield stored_messages, None
         finally:
-            # Important: Cleanup Sandbox
             sandbox.cleanup()
 
-# Gradio Block Construction
 theme = gr.themes.Default(
     font=["Oxanium", "sans-serif"], primary_hue="amber", secondary_hue="blue"
 )
@@ -697,7 +688,6 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
     stored_messages = gr.State([])
     
     with gr.Row():
-        # Left Sidebar: Inputs
         with gr.Column(scale=1):
             gr.Markdown("### Fara CUA - Chrome Agent 🌐")
             
@@ -708,22 +698,18 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
             )
             run_btn = gr.Button("Start Task", variant="primary")
             stop_btn = gr.Button("Stop", variant="secondary")
-            
             consent_storage = gr.Checkbox(label="Save logs locally?", value=True)
             
             gr.Examples(
                 examples=[
                     "Go to google.com and search for 'Hugging Face', then click the first link.",
                     "Go to wikipedia.org, type 'Python' in search, and click the search button.",
-                    "Open huggingface.co and find the 'Spaces' link."
                 ],
                 inputs=task_input
             )
 
-        # Right Main: Chat & Live View
         with gr.Column(scale=3):
             with gr.Row():
-                # Side-by-side: Chat and Latest Screenshot
                 with gr.Column(scale=1):
                     chatbot_display = gr.Chatbot(
                         label="Agent Trace",
@@ -740,7 +726,6 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
                         height=600
                     )
 
-    # UI Handler
     agent_ui = EnrichedGradioUI(CodeAgent(tools=[], model=Model(), name="init"))
 
     def interrupt_agent(session_state):
@@ -748,7 +733,6 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
             session_state["agent"].interrupt_switch = True
             return "Interrupted"
     
-    # Run Logic
     run_event = run_btn.click(
         fn=agent_ui.interact_with_agent,
         inputs=[
@@ -764,4 +748,4 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
     stop_btn.click(fn=interrupt_agent, inputs=[session_state], outputs=[])
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(share=True)