All Products
Search
Document Center

AgentBay:UI automation for mobile use

Last Updated:Jun 02, 2026

The AgentBay SDK provides UI automation for cloud phones, including touch gestures, text input, key events, UI element detection, and screenshots.

Overview

The SDK provides the following UI automation features:

  1. Touch operations: Tap and swipe gestures for cloud phone interaction.

  2. Text input: Enter text and send hardware keypress events.

  3. UI element detection: Find and interact with UI elements.

  4. Screen operations: Capture screenshots for visual verification.

Create a session

from agentbay import AgentBay
from agentbay.session_params import CreateSessionParams

agent_bay = AgentBay()
session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session
# The session is created. You can now automate the cloud phone.

Touch operations

Tap gestures

Tap the screen at specific coordinates:

session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

# Tap at the specified coordinates.
result = session.mobile.tap(x=500, y=300)
if result.success:
    print("Tap successful")  # Output: Tap successful
else:
    print(f"Tap failed: {result.error_message}")

agent_bay.delete(session)

Swipe gestures

Swipe from one point to another:

session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

# Swipe up (from bottom to top).
result = session.mobile.swipe(
    start_x=100,
    start_y=500,
    end_x=100,
    end_y=200,
    duration_ms=300
)
if result.success:
    print("Swipe up successful")  # Output: Swipe up successful

# Swipe left (from right to left).
result = session.mobile.swipe(
    start_x=500,
    start_y=300,
    end_x=100,
    end_y=300,
    duration_ms=300
)
if result.success:
    print("Swipe left successful")  # Output: Swipe left successful

agent_bay.delete(session)

Parameters:

  • start_xstart_y: Starting coordinates.

  • end_xend_y: Ending coordinates.

  • duration_ms: Swipe duration in milliseconds. Default: 300.

Text input

Input text

Enter text into the active input field:

session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

result = session.mobile.input_text("Hello AgentBay!")
if result.success:
    print("Text input successful")  # Output: Text input successful

agent_bay.delete(session)

Send key events

Send Android key events using KeyCode constants:

from agentbay.mobile.mobile import KeyCode

session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

# Press the HOME key.
result = session.mobile.send_key(KeyCode.HOME)
if result.success:
    print("HOME key pressed")  # Output: HOME key pressed

# KeyCode values: HOME=3, BACK=4, VOLUME_UP=24, VOLUME_DOWN=25, POWER=26, MENU=82
print(f"HOME keycode value: {KeyCode.HOME}")  # Output: HOME keycode value: 3

agent_bay.delete(session)

Available KeyCode constants:

KeyCode

Value

Description

KeyCode.HOME

3

Home button

KeyCode.BACK

4

Back button

KeyCode.VOLUME_UP

24

Volume up button

KeyCode.VOLUME_DOWN

25

Volume down button

KeyCode.POWER

26

Power button

KeyCode.MENU

82

Menu button

Note: All key events are sent directly to the Android system.

UI element detection

Retrieve all UI elements

Retrieve all UI elements in the current screen hierarchy:

session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

result = session.mobile.get_all_ui_elements(timeout_ms=2000)
if result.success:
    print(f"Found {len(result.elements)} UI elements")  # Output: Found 2172 UI elements
    for element in result.elements:
        # The element structure varies. Check the element data.
        print(f"Element: {element}")
        # Example output: Element data contains UI hierarchy information
else:
    print(f"Failed: {result.error_message}")

agent_bay.delete(session)

Parameter:

  • timeout_ms: Timeout in milliseconds for UI element retrieval. Default: 2000.

Screen operations

Capture a screenshot

Capture the current cloud phone screen:

session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

result = session.mobile.screenshot()
if result.success:
    screenshot_url = result.data
    print(f"Screenshot URL: {screenshot_url}")
    # Output: Screenshot URL: https://***.***.aliyuncs.com/***/screenshot_1234567890.png?***
else:
    print(f"Screenshot failed: {result.error_message}")

agent_bay.delete(session)

Best practices

Use a cloud phone image

Cloud phone UI automation requires a mobile OS image such as mobile_latest:

# Correct - Use a cloud phone OS image.
session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

# Incorrect - Cannot be used for cloud phone operations.
session_params = CreateSessionParams(image_id="windows_latest")
session = agent_bay.create(session_params).session

Handle screenshot URLs

The screenshot method returns an OSS URL, not image data:

result = session.mobile.screenshot()
if result.success:
    screenshot_url = result.data
    print(f"Screenshot available at: {screenshot_url}")
    # Output: Screenshot available at: https://***.***.aliyuncs.com/***/screenshot_1234567890.png?***
    # Use the URL to download or display the screenshot.
else:
    print(f"Screenshot failed: {result.error_message}")
    

Common use cases

App navigation

from agentbay import AgentBay
from agentbay.session_params import CreateSessionParams

agent_bay = AgentBay()
session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

try:
    # Tap the application icon.
    tap_result = session.mobile.tap(x=200, y=400)
    print(f"App tap result: {tap_result.success}")  # Output: App tap result: True
    
    # Wait for the application to load.
    import time
    time.sleep(2)
    
    # Swipe to navigate.
    swipe_result = session.mobile.swipe(
        start_x=400,
        start_y=600,
        end_x=100,
        end_y=600,
        duration_ms=300
    )
    print(f"Navigation swipe result: {swipe_result.success}")  # Output: Navigation swipe result: True
    
    # Tap the button.
    button_result = session.mobile.tap(x=300, y=800)
    print(f"Button tap result: {button_result.success}")  # Output: Button tap result: True
    
finally:
    agent_bay.delete(session)

Form filling

from agentbay import AgentBay
from agentbay.session_params import CreateSessionParams

agent_bay = AgentBay()
session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

try:
    # Tap the username field.
    username_tap = session.mobile.tap(x=300, y=400)
    print(f"Username field focused: {username_tap.success}")  # Output: Username field focused: True
    
    # Enter the username.
    username_input = session.mobile.input_text("john_doe")
    print(f"Username entered: {username_input.success}")  # Output: Username entered: True
    
    # Tap the password field.
    password_tap = session.mobile.tap(x=300, y=500)
    print(f"Password field focused: {password_tap.success}")  # Output: Password field focused: True
    
    # Enter the password.
    password_input = session.mobile.input_text("secure_password")
    print(f"Password entered: {password_input.success}")  # Output: Password entered: True
    
    # Tap the logon button.
    login_tap = session.mobile.tap(x=300, y=650)
    print(f"Login button pressed: {login_tap.success}")  # Output: Login button pressed: True
    
finally:
    agent_bay.delete(session)

UI element discovery

from agentbay import AgentBay
from agentbay.session_params import CreateSessionParams

agent_bay = AgentBay()
session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

try:
    # Get all clickable elements.
    result = session.mobile.get_clickable_ui_elements(timeout_ms=3000)
    
    if result.success:
        print(f"Found {len(result.elements)} clickable elements")  # Output: Found 3 clickable elements
        
        # Analyze the elements to find the target.
        for i, element in enumerate(result.elements):
            print(f"Element {i+1}: {element}")  
            # Example output:
            # Element 1: UI element with interaction capabilities
            # Element 2: UI element with interaction capabilities  
            # Element 3: UI element with interaction capabilities
    
    # Take a screenshot for verification.
    screenshot = session.mobile.screenshot()
    if screenshot.success:
        screenshot_url = screenshot.data
        print(f"Screenshot URL: {screenshot_url}")
        # Output: Screenshot URL: https://***.***.aliyuncs.com/***/screenshot_1234567890.png?***
    
finally:
    agent_bay.delete(session)

Content scrolling

from agentbay import AgentBay
from agentbay.session_params import CreateSessionParams

agent_bay = AgentBay()
session_params = CreateSessionParams(image_id="mobile_latest")
session = agent_bay.create(session_params).session

try:
    # Scroll down multiple times.
    for i in range(3):
        scroll_result = session.mobile.swipe(
            start_x=300,
            start_y=800,
            end_x=300,
            end_y=200,
            duration_ms=400
        )
        print(f"Scroll down {i+1}: {scroll_result.success}")  # Output: Scroll down 1: True, etc.
        
        # Pause briefly between scrolls.
        import time
        time.sleep(1)
    
    # Scroll up.
    up_result = session.mobile.swipe(
        start_x=300,
        start_y=200,
        end_x=300,
        end_y=800,
        duration_ms=400
    )
    print(f"Scroll up result: {up_result.success}")  # Output: Scroll up result: True
    
finally:
    agent_bay.delete(session)

Troubleshooting

FAQ

  1. "Tool not found" error

    • Ensure you are using a cloud phone OS image, such as image_id="mobile_latest".

    • Verify that the session was created successfully.

    • Verify that the API key and endpoint are configured correctly.

  2. Hardware keypress operations

    • Key events are sent directly to the Android system.

    • Check result.success to verify the key was sent.

    • Error handling example:

      result = session.mobile.send_key(KeyCode.HOME)
      if not result.success:
          print(f"Key press failed: {result.error_message}")
      
  3. UI element detection returns an empty result

    • Increase the timeout_ms value.

    • Capture a screenshot to check the current UI state.

    • Ensure the target screen has finished loading.

  4. Screenshot URLs

    • The screenshot returns an OSS URL, not image data.

    • result.data contains the download URL.

    • Use the URL to download the screenshot if needed.

  5. Swipe gesture does not work as expected

    • Verify the coordinates are within screen boundaries.

    • Adjust duration_ms for different gesture speeds.

    • Ensure the start and end coordinates form a meaningful swipe path.