{ "@context": "https://schema.org", "@id": "https://aggr.aymericbeaumet.com/items/philschmid-de/2026-06-25-control-an-android-phone-with-gemini-3-5-flash-computer-use/#webpage", "@type": [ "WebPage", "ArchiveComponent" ], "articleBody": "June 25, 2026 9 minute read [View Code](https://github.com/google-gemini/gemini-android-computer-use-quickstart)\n\nGemini 3.5 Flash has [built-in Computer Use](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-computer-use-gemini-3-5-flash/). The model looks at a screenshot, decides what to do, and returns a function call like `click(y=300, x=500)`. You execute that action on the device via ADB, take a new screenshot, and send it back. Repeat until the task is done.\n\nThis guide walks you through controlling an Android emulator using the `mobile` environment and the Python SDK.\n\nGithub Repository: [https://github.com/google-gemini/gemini-android-computer-use-quickstart](https://github.com/google-gemini/gemini-android-computer-use-quickstart)\n\n## What is Computer Use?\n\n[Computer Use](https://ai.google.dev/gemini-api/docs/computer-use) is a native tool in Gemini 3.5 Flash. You give the model a screenshot and a goal (\"open Settings and turn on dark mode\"). The model returns structured actions: taps, text input, swipes, app launches. Your code executes those actions on the target device.\n\nIt works like function calling. The model proposes actions, you execute them, you send back the result. The model stays in the loop through screenshots.\n\nGemini supports three Computer Use environments: `browser` for desktop web automation, `mobile` for mobile devices/emulators, and `desktop` for operating system-level control. This guide uses `mobile`.\n\n```plaintext\n Screenshot ──> Gemini 3.5 Flash ──> function_call\n ▲ │\n │ ▼\n └──────── Execute via ADB ◄──────────┘\n```\n\n### Pseudocode\n\nPython\n\n```python\nfrom google import genai\n \nclient = genai.Client()\nbridge = ADBBridge()\n \n# Take initial screenshot and send first request\nscreenshot = bridge.screenshot()\n \ninteraction = client.interactions.create(\n model=\"gemini-3.5-flash\",\n input=[\n {\"type\": \"text\", \"text\": \"Open Settings and enable dark mode\"},\n {\"type\": \"image\", \"data\": b64(screenshot), \"mime_type\": \"image/png\"},\n ],\n tools=[{\"type\": \"computer_use\", \"environment\": \"mobile\"}],\n)\n \n# Agent loop: execute actions, send results back\nwhile interaction has function_calls:\n for call in interaction.function_calls:\n bridge.execute(call.name, call.args) # click, type, open_app...\n \n screenshot = bridge.screenshot()\n interaction = client.interactions.create(\n model=\"gemini-3.5-flash\",\n previous_interaction_id=interaction.id,\n input=[function_results + screenshot],\n tools=[{\"type\": \"computer_use\", \"environment\": \"mobile\"}],\n )\n \nprint(interaction.output_text)\n```\n\n### Supported actions (mobile environment)\n\nThese are the actions the model can request when using `environment: \"mobile\"`. Your code must handle each one.\n\n| Action | Description | Arguments |\n| ----------------- | ---------------------------------- | -------------------------------------------------------------- |\n| `open_app` | Opens an app by name | `app_name: str`, `intent: str` (optional) |\n| `click` | Tap a screen coordinate | `y: int (0-999)`, `x: int (0-999)` |\n| `type` | Type text | `text: str`, `press_enter: bool` (default false) |\n| `long_press` | Long-press a coordinate | `y: int (0-999)`, `x: int (0-999)`, `seconds: int` (default 2) |\n| `drag_and_drop` | Drag between two points | `start_y`, `start_x`, `end_y`, `end_x` |\n| `press_key` | Press a key (home, back, enter...) | `key: str` |\n| `go_back` | Navigate back | (none) |\n| `wait` | Pause execution | `seconds: int` (default 1) |\n| `list_apps` | List installed apps | (none) |\n| `take_screenshot` | Capture the screen | (none) |\n\nCoordinates are normalized to a 0-999 grid. `(0, 0)` is top-left, `(999, 999)` is bottom-right. Your bridge converts these to actual pixel coordinates based on the device resolution.\n\n## Setup\n\nNo Android Studio GUI required. Run the [setup script](https://github.com/google-gemini/gemini-android-computer-use-quickstart/blob/main/setup_emulator.sh) on your Mac to install the Android SDK, emulator, and create a virtual device from the terminal.\n\n### 1\\. Run the setup script\n\nInstall script link: [https://github.com/google-gemini/gemini-android-computer-use-quickstart/blob/main/setup_emulator.sh](https://github.com/google-gemini/gemini-android-computer-use-quickstart/blob/main/setup_emulator.sh)\n\nBash\n\n```bash\nchmod +x setup_emulator.sh\n./setup_emulator.sh\n```\n\n### 2\\. Install Python dependencies\n\nBash\n\n```bash\npip install google-genai\n```\n\nThe Python agent script handles locating SDK paths and starting the emulator in the background automatically, so no additional manual environment variable exporting or terminal setup is required.\n\n## The agent loop\n\nFull working script [agent.py](https://github.com/google-gemini/gemini-android-computer-use-quickstart/blob/main/agent.py). It automatically sets up the environment variables, launches the emulator in the background (if it isn't already running), waits for it to boot, and begins the Computer Use loop.\n\nPython\n\n```python\nimport base64\nimport json\nimport os\nimport re\nimport subprocess\nimport sys\nimport time\n \nfrom google import genai\n \nBASE_DIR = os.path.dirname(os.path.abspath(__file__))\n \n \ndef setup_android_env():\n paths_to_check = [\n os.environ.get(\"ANDROID_HOME\"),\n \"/opt/homebrew/share/android-commandlinetools\",\n \"/usr/local/share/android-commandlinetools\",\n ]\n \n android_home = None\n for p in paths_to_check:\n if p and os.path.exists(p):\n android_home = p\n break\n \n if not android_home:\n print(\"Error: ANDROID_HOME not found. Run setup_emulator.sh first.\", file=sys.stderr)\n sys.exit(1)\n \n os.environ[\"ANDROID_HOME\"] = android_home\n sdk_paths = [\n os.path.join(android_home, \"cmdline-tools\", \"latest\", \"bin\"),\n os.path.join(android_home, \"emulator\"),\n os.path.join(android_home, \"platform-tools\"),\n ]\n \n current_path = os.environ.get(\"PATH\", \"\")\n for p in sdk_paths:\n if p not in current_path:\n current_path = p + os.pathsep + current_path\n \n os.environ[\"PATH\"] = current_path\n return android_home\n \n \ndef start_emulator(avd_name=\"AI_Agent_Phone\"):\n setup_android_env()\n \n try:\n res = subprocess.run([\"adb\", \"devices\"], capture_output=True, text=True)\n if \"emulator\" in res.stdout:\n return\n except FileNotFoundError:\n pass\n \n print(f\"Starting emulator '{avd_name}'...\")\n log_file = open(os.path.join(BASE_DIR, \"emulator.log\"), \"w\")\n \n subprocess.Popen(\n [\"emulator\", \"-avd\", avd_name, \"-delay-adb\"],\n stdout=log_file,\n stderr=log_file,\n start_new_session=True,\n )\n \n print(\"Waiting for emulator to boot...\")\n for _ in range(60):\n try:\n res = subprocess.run([\"adb\", \"devices\"], capture_output=True, text=True)\n if \"emulator\" in res.stdout:\n boot_res = subprocess.run(\n [\"adb\", \"shell\", \"getprop\", \"sys.boot_completed\"],\n capture_output=True,\n text=True,\n )\n if boot_res.stdout.strip() == \"1\":\n print(\"Emulator ready.\")\n return\n except Exception:\n pass\n time.sleep(2)\n \n print(\"Error: Emulator failed to boot.\", file=sys.stderr)\n sys.exit(1)\n \n \nclass ADBBridge:\n def __init__(self, device_id=None):\n self.prefix = [\"adb\"] + ([\"-s\", device_id] if device_id else [])\n self.width, self.height = self._screen_size()\n \n def _run(self, args, check=True):\n result = subprocess.run(self.prefix + args, capture_output=True, text=True)\n if check and result.returncode != 0:\n raise RuntimeError(f\"ADB error: {result.stderr.strip()}\")\n return result.stdout\n \n def _screen_size(self):\n output = self._run([\"shell\", \"wm\", \"size\"])\n match = re.search(r\"Physical size: (\\d+)x(\\d+)\", output)\n return (int(match.group(1)), int(match.group(2))) if match else (1080, 1920)\n \n def _px(self, x, y):\n return int(x / 1000 * self.width), int(y / 1000 * self.height)\n \n def click(self, y, x, **_):\n px, py = self._px(x, y)\n self._run([\"shell\", \"input\", \"tap\", str(px), str(py)])\n \n def type(self, text, press_enter=False, **_):\n self._run([\"shell\", \"input\", \"text\", text.replace(\" \", \"%s\")])\n if press_enter:\n self._run([\"shell\", \"input\", \"keyevent\", \"66\"])\n \n def open_app(self, app_name=None, package_name=None, **_):\n pkg = app_name or package_name\n if not pkg:\n raise ValueError(\"open_app requires app_name or package_name\")\n \n stdout = self._run([\"shell\", \"monkey\", \"--pct-syskeys\", \"0\", \"-p\", pkg,\n \"-c\", \"android.intent.category.LAUNCHER\", \"1\"], check=False)\n if \"No activities found\" in stdout or \"monkey aborted\" in stdout:\n raise RuntimeError(f\"App {pkg} is not installed or has no launcher activity.\")\n \n def scroll(self, y, x, direction, magnitude=800, **_):\n px, py = self._px(x, y)\n dist = int(magnitude / 1000 * self.height)\n dx, dy = {\"up\": (0, -dist), \"down\": (0, dist),\n \"left\": (-dist, 0), \"right\": (dist, 0)}.get(direction, (0, 0))\n self._run([\"shell\", \"input\", \"swipe\", str(px), str(py),\n str(px + dx), str(py + dy), \"300\"])\n \n def long_press(self, y, x, seconds=2, **_):\n px, py = self._px(x, y)\n self._run([\"shell\", \"input\", \"swipe\", str(px), str(py),\n str(px), str(py), str(seconds * 1000)])\n \n def drag_and_drop(self, start_y, start_x, end_y, end_x, **_):\n sx, sy = self._px(start_x, start_y)\n ex, ey = self._px(end_x, end_y)\n self._run([\"shell\", \"input\", \"swipe\", str(sx), str(sy),\n str(ex), str(ey), \"300\"])\n \n def press_key(self, key, **_):\n keymap = {\"home\": \"3\", \"back\": \"4\", \"enter\": \"66\",\n \"app_switch\": \"187\", \"menu\": \"82\"}\n self._run([\"shell\", \"input\", \"keyevent\", keymap.get(key.lower(), key)])\n \n def go_back(self, **_):\n self._run([\"shell\", \"input\", \"keyevent\", \"4\"])\n \n def wait(self, seconds=1, **_):\n time.sleep(seconds)\n \n def list_apps(self, **_):\n output = self._run([\"shell\", \"pm\", \"list\", \"packages\", \"-3\"])\n apps = [l.split(\":\")[1] for l in output.splitlines() if l.startswith(\"package:\")]\n if not apps:\n return {\"apps\": \"No third-party apps installed on this device.\"}\n return {\"apps\": apps}\n \n def take_screenshot(self, **_):\n return None\n \n def screenshot(self) -> bytes:\n result = subprocess.run(\n self.prefix + [\"exec-out\", \"screencap\", \"-p\"], capture_output=True\n )\n return result.stdout\n \n \nSYSTEM_PROMPT = \"\"\"You are operating an Android phone.\n* Use the provided tools to complete the task.\n* Scroll down to inspect the full screen before assuming an element is missing.\n* You can open apps by package name from anywhere.\n* Type text only using the `type` tool. Do not use the virtual keyboard.\n* If the task is already complete, state that directly.\n\"\"\"\n \n \ndef run_agent(task: str, device_id: str = None, max_turns: int = 100):\n start_emulator()\n client = genai.Client()\n bridge = ADBBridge(device_id)\n \n print(f\"\\nTask: {task}\")\n print(\"-\" * 40)\n \n screenshot_bytes = bridge.screenshot()\n user_input = [\n {\"type\": \"text\", \"text\": task},\n {\n \"type\": \"image\",\n \"data\": base64.b64encode(screenshot_bytes).decode(),\n \"mime_type\": \"image/png\",\n },\n ]\n \n previous_interaction_id = None\n turn = 0\n \n while turn < max_turns:\n turn += 1\n \n interaction = client.interactions.create(\n model=\"gemini-3.5-flash\",\n system_instruction=SYSTEM_PROMPT,\n input=user_input,\n tools=[{\"type\": \"computer_use\", \"environment\": \"mobile\"}],\n previous_interaction_id=previous_interaction_id,\n )\n \n function_responses = []\n for step in interaction.steps:\n if step.type == \"function_call\":\n print(f\"[function_call] {step.name}({step.arguments})\")\n handler = getattr(bridge, step.name, None)\n result_text = {\"status\": \"ok\"}\n \n if handler:\n try:\n res = handler(**step.arguments)\n if isinstance(res, dict):\n result_text.update(res)\n except Exception as e:\n result_text = {\"status\": \"error\", \"error\": str(e)}\n else:\n result_text = {\"status\": \"error\", \"error\": f\"Unknown action: {step.name}\"}\n \n print(f\"[function_result] {result_text}\")\n \n if \"safety_decision\" in step.arguments:\n # auto approve safety decisions for demo\n result_text[\"safety_acknowledgement\"] = True\n \n screenshot_bytes = bridge.screenshot()\n \n fr = {\n \"type\": \"function_result\",\n \"name\": step.name,\n \"call_id\": step.id,\n \"result\": [\n {\"type\": \"text\", \"text\": json.dumps(result_text)},\n {\n \"type\": \"image\",\n \"data\": base64.b64encode(screenshot_bytes).decode(),\n \"mime_type\": \"image/png\",\n },\n ],\n }\n function_responses.append(fr)\n else:\n print(f\"\\nResult: {interaction.output_text}\")\n break\n \n user_input = function_responses\n previous_interaction_id = interaction.id\n \n if not function_responses:\n break\n \n return interaction\n \n \nif __name__ == \"__main__\":\n task_desc = \"Find the latest blog post from philipp schmid and summarize it.\"\n if len(sys.argv) > 1:\n task_desc = \" \".join(sys.argv[1:])\n run_agent(task_desc)\n```\n\n### How to run it\n\nBash\n\n```bash\n# Set your API key\nexport GEMINI_API_KEY=\"your-key\"\n \n# Run the agent (starts the emulator automatically if needed)\npython agent.py \"Open Settings and enable dark mode\"\n```\n\n## Connecting to a Remote Device\n\nYou can also target physical Android devices or remote cloud emulators instead of a local virtual device.\n\n1. **Enable USB/Wireless Debugging**: On your target device, enable Developer Options and turn on USB Debugging or Wireless Debugging.\n2. **Connect via ADB**: Use the `adb connect` command to link to a remote device over TCP/IP:\n\n Bash\n\n ```bash\n adb connect :5555\n ```\n\n For details on setting up wireless debugging, refer to the official [Android developer documentation on ADB over Wi-Fi](https://developer.android.com/tools/adb#wireless).\n3. **Pass the Device ID to the Agent**: Pass the remote device's connection string as the `device_id` parameter to target it in the agent loop:\n\n Python\n\n ```python\n # Target a remote or cloud-hosted emulator\n run_agent(\"Check the weather\", device_id=\"35.200.100.10:5555\")\n ```\n\n## Next Steps & Developer Tips\n\nI hope this gets you started with Gemini 3.5 Flash Computer Use on mobile. Next steps can be:\n\n- **Supporting iOS / iPhone**: The Gemini API's `mobile` environment is platform-agnostic. The model outputs actions (clicks, swipes, types) on the same normalized `0-999` grid regardless of whether the device is Android or iOS. To target an iPhone or iOS Simulator, you only need to swap out `ADBBridge` with an iOS-compatible tool—such as Apple's `simctl` CLI for simulator control, Appium, or [go-ios](https://github.com/danielpaulus/go-ios) for physical devices.\n- **Production Robustness**: The Python bridge code provided here is synchronous and optimized for demonstration. For production use, you should implement robust retry logic for network drops, handle ADB disconnects gracefully, and execute operations asynchronously.\n- **Handling Safety Decisions**: In real-world tasks (especially those modifying state or making payments), the model might flag action steps with a `safety_decision` requesting confirmation. Ensure your production loop inspects `step.arguments` for safety flags and prompts the user before executing the action. See the [Gemini API Computer Use Safety Guidelines](https://ai.google.dev/gemini-api/docs/computer-use#safety-decisions) for details.\n\n* * *\n\nThanks for reading! If you have any questions or feedback, please let me know on [Twitter](https://twitter.com/_philschmid) or [LinkedIn](https://www.linkedin.com/in/philipp-schmid-a6a2bb196/).\n", "authors": [], "captured_at": "2026-09-27T19:29:15.293681927+00:00", "category": "programming", "content_html": "

June 25, 2026 9 minute read View Code

\n

Gemini 3.5 Flash has built-in Computer Use. The model looks at a screenshot, decides what to do, and returns a function call like click(y=300, x=500). You execute that action on the device via ADB, take a new screenshot, and send it back. Repeat until the task is done.

\n

This guide walks you through controlling an Android emulator using the mobile environment and the Python SDK.

\n

Github Repository: https://github.com/google-gemini/gemini-android-computer-use-quickstart

\n

What is Computer Use?

\n

Computer Use is a native tool in Gemini 3.5 Flash. You give the model a screenshot and a goal ("open Settings and turn on dark mode"). The model returns structured actions: taps, text input, swipes, app launches. Your code executes those actions on the target device.

\n

It works like function calling. The model proposes actions, you execute them, you send back the result. The model stays in the loop through screenshots.

\n

Gemini supports three Computer Use environments: browser for desktop web automation, mobile for mobile devices/emulators, and desktop for operating system-level control. This guide uses mobile.

\n
  Screenshot ──> Gemini 3.5 Flash ──> function_call\n      ▲                                    │\n      │                                    ▼\n      └──────── Execute via ADB ◄──────────┘\n
\n

Pseudocode

\n

Python

\n
from google import genai\n \nclient = genai.Client()\nbridge = ADBBridge()\n \n# Take initial screenshot and send first request\nscreenshot = bridge.screenshot()\n \ninteraction = client.interactions.create(\n    model="gemini-3.5-flash",\n    input=[\n        {"type": "text", "text": "Open Settings and enable dark mode"},\n        {"type": "image", "data": b64(screenshot), "mime_type": "image/png"},\n    ],\n    tools=[{"type": "computer_use", "environment": "mobile"}],\n)\n \n# Agent loop: execute actions, send results back\nwhile interaction has function_calls:\n    for call in interaction.function_calls:\n        bridge.execute(call.name, call.args)  # click, type, open_app...\n \n    screenshot = bridge.screenshot()\n    interaction = client.interactions.create(\n        model="gemini-3.5-flash",\n        previous_interaction_id=interaction.id,\n        input=[function_results + screenshot],\n        tools=[{"type": "computer_use", "environment": "mobile"}],\n    )\n \nprint(interaction.output_text)\n
\n

Supported actions (mobile environment)

\n

These are the actions the model can request when using environment: "mobile". Your code must handle each one.

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
ActionDescriptionArguments
open_appOpens an app by nameapp_name: str, intent: str (optional)
clickTap a screen coordinatey: int (0-999), x: int (0-999)
typeType texttext: str, press_enter: bool (default false)
long_pressLong-press a coordinatey: int (0-999), x: int (0-999), seconds: int (default 2)
drag_and_dropDrag between two pointsstart_y, start_x, end_y, end_x
press_keyPress a key (home, back, enter...)key: str
go_backNavigate back(none)
waitPause executionseconds: int (default 1)
list_appsList installed apps(none)
take_screenshotCapture the screen(none)
\n

Coordinates are normalized to a 0-999 grid. (0, 0) is top-left, (999, 999) is bottom-right. Your bridge converts these to actual pixel coordinates based on the device resolution.

\n

Setup

\n

No Android Studio GUI required. Run the setup script on your Mac to install the Android SDK, emulator, and create a virtual device from the terminal.

\n

1. Run the setup script

\n

Install script link: https://github.com/google-gemini/gemini-android-computer-use-quickstart/blob/main/setup_emulator.sh

\n

Bash

\n
chmod +x setup_emulator.sh\n./setup_emulator.sh\n
\n

2. Install Python dependencies

\n

Bash

\n
pip install google-genai\n
\n

The Python agent script handles locating SDK paths and starting the emulator in the background automatically, so no additional manual environment variable exporting or terminal setup is required.

\n

The agent loop

\n

Full working script agent.py. It automatically sets up the environment variables, launches the emulator in the background (if it isn't already running), waits for it to boot, and begins the Computer Use loop.

\n

Python

\n
import base64\nimport json\nimport os\nimport re\nimport subprocess\nimport sys\nimport time\n \nfrom google import genai\n \nBASE_DIR = os.path.dirname(os.path.abspath(__file__))\n \n \ndef setup_android_env():\n    paths_to_check = [\n        os.environ.get("ANDROID_HOME"),\n        "/opt/homebrew/share/android-commandlinetools",\n        "/usr/local/share/android-commandlinetools",\n    ]\n    \n    android_home = None\n    for p in paths_to_check:\n        if p and os.path.exists(p):\n            android_home = p\n            break\n            \n    if not android_home:\n        print("Error: ANDROID_HOME not found. Run setup_emulator.sh first.", file=sys.stderr)\n        sys.exit(1)\n        \n    os.environ["ANDROID_HOME"] = android_home\n    sdk_paths = [\n        os.path.join(android_home, "cmdline-tools", "latest", "bin"),\n        os.path.join(android_home, "emulator"),\n        os.path.join(android_home, "platform-tools"),\n    ]\n    \n    current_path = os.environ.get("PATH", "")\n    for p in sdk_paths:\n        if p not in current_path:\n            current_path = p + os.pathsep + current_path\n            \n    os.environ["PATH"] = current_path\n    return android_home\n \n \ndef start_emulator(avd_name="AI_Agent_Phone"):\n    setup_android_env()\n    \n    try:\n        res = subprocess.run(["adb", "devices"], capture_output=True, text=True)\n        if "emulator" in res.stdout:\n            return\n    except FileNotFoundError:\n        pass\n \n    print(f"Starting emulator '{avd_name}'...")\n    log_file = open(os.path.join(BASE_DIR, "emulator.log"), "w")\n    \n    subprocess.Popen(\n        ["emulator", "-avd", avd_name, "-delay-adb"],\n        stdout=log_file,\n        stderr=log_file,\n        start_new_session=True,\n    )\n    \n    print("Waiting for emulator to boot...")\n    for _ in range(60):\n        try:\n            res = subprocess.run(["adb", "devices"], capture_output=True, text=True)\n            if "emulator" in res.stdout:\n                boot_res = subprocess.run(\n                    ["adb", "shell", "getprop", "sys.boot_completed"],\n                    capture_output=True,\n                    text=True,\n                )\n                if boot_res.stdout.strip() == "1":\n                    print("Emulator ready.")\n                    return\n        except Exception:\n            pass\n        time.sleep(2)\n        \n    print("Error: Emulator failed to boot.", file=sys.stderr)\n    sys.exit(1)\n \n \nclass ADBBridge:\n    def __init__(self, device_id=None):\n        self.prefix = ["adb"] + (["-s", device_id] if device_id else [])\n        self.width, self.height = self._screen_size()\n \n    def _run(self, args, check=True):\n        result = subprocess.run(self.prefix + args, capture_output=True, text=True)\n        if check and result.returncode != 0:\n            raise RuntimeError(f"ADB error: {result.stderr.strip()}")\n        return result.stdout\n \n    def _screen_size(self):\n        output = self._run(["shell", "wm", "size"])\n        match = re.search(r"Physical size: (\\d+)x(\\d+)", output)\n        return (int(match.group(1)), int(match.group(2))) if match else (1080, 1920)\n \n    def _px(self, x, y):\n        return int(x / 1000 * self.width), int(y / 1000 * self.height)\n \n    def click(self, y, x, **_):\n        px, py = self._px(x, y)\n        self._run(["shell", "input", "tap", str(px), str(py)])\n \n    def type(self, text, press_enter=False, **_):\n        self._run(["shell", "input", "text", text.replace(" ", "%s")])\n        if press_enter:\n            self._run(["shell", "input", "keyevent", "66"])\n \n    def open_app(self, app_name=None, package_name=None, **_):\n        pkg = app_name or package_name\n        if not pkg:\n            raise ValueError("open_app requires app_name or package_name")\n            \n        stdout = self._run(["shell", "monkey", "--pct-syskeys", "0", "-p", pkg,\n                            "-c", "android.intent.category.LAUNCHER", "1"], check=False)\n        if "No activities found" in stdout or "monkey aborted" in stdout:\n            raise RuntimeError(f"App {pkg} is not installed or has no launcher activity.")\n \n    def scroll(self, y, x, direction, magnitude=800, **_):\n        px, py = self._px(x, y)\n        dist = int(magnitude / 1000 * self.height)\n        dx, dy = {"up": (0, -dist), "down": (0, dist),\n                  "left": (-dist, 0), "right": (dist, 0)}.get(direction, (0, 0))\n        self._run(["shell", "input", "swipe", str(px), str(py),\n                   str(px + dx), str(py + dy), "300"])\n \n    def long_press(self, y, x, seconds=2, **_):\n        px, py = self._px(x, y)\n        self._run(["shell", "input", "swipe", str(px), str(py),\n                   str(px), str(py), str(seconds * 1000)])\n \n    def drag_and_drop(self, start_y, start_x, end_y, end_x, **_):\n        sx, sy = self._px(start_x, start_y)\n        ex, ey = self._px(end_x, end_y)\n        self._run(["shell", "input", "swipe", str(sx), str(sy),\n                   str(ex), str(ey), "300"])\n \n    def press_key(self, key, **_):\n        keymap = {"home": "3", "back": "4", "enter": "66",\n                  "app_switch": "187", "menu": "82"}\n        self._run(["shell", "input", "keyevent", keymap.get(key.lower(), key)])\n \n    def go_back(self, **_):\n        self._run(["shell", "input", "keyevent", "4"])\n \n    def wait(self, seconds=1, **_):\n        time.sleep(seconds)\n \n    def list_apps(self, **_):\n        output = self._run(["shell", "pm", "list", "packages", "-3"])\n        apps = [l.split(":")[1] for l in output.splitlines() if l.startswith("package:")]\n        if not apps:\n            return {"apps": "No third-party apps installed on this device."}\n        return {"apps": apps}\n \n    def take_screenshot(self, **_):\n        return None\n \n    def screenshot(self) -> bytes:\n        result = subprocess.run(\n            self.prefix + ["exec-out", "screencap", "-p"], capture_output=True\n        )\n        return result.stdout\n \n \nSYSTEM_PROMPT = """You are operating an Android phone.\n* Use the provided tools to complete the task.\n* Scroll down to inspect the full screen before assuming an element is missing.\n* You can open apps by package name from anywhere.\n* Type text only using the `type` tool. Do not use the virtual keyboard.\n* If the task is already complete, state that directly.\n"""\n \n \ndef run_agent(task: str, device_id: str = None, max_turns: int = 100):\n    start_emulator()\n    client = genai.Client()\n    bridge = ADBBridge(device_id)\n \n    print(f"\\nTask: {task}")\n    print("-" * 40)\n \n    screenshot_bytes = bridge.screenshot()\n    user_input = [\n        {"type": "text", "text": task},\n        {\n            "type": "image",\n            "data": base64.b64encode(screenshot_bytes).decode(),\n            "mime_type": "image/png",\n        },\n    ]\n    \n    previous_interaction_id = None\n    turn = 0\n    \n    while turn < max_turns:\n        turn += 1\n        \n        interaction = client.interactions.create(\n            model="gemini-3.5-flash",\n            system_instruction=SYSTEM_PROMPT,\n            input=user_input,\n            tools=[{"type": "computer_use", "environment": "mobile"}],\n            previous_interaction_id=previous_interaction_id,\n        )\n        \n        function_responses = []\n        for step in interaction.steps:\n            if step.type == "function_call":\n                print(f"[function_call] {step.name}({step.arguments})")\n                handler = getattr(bridge, step.name, None)\n                result_text = {"status": "ok"}\n                \n                if handler:\n                    try:\n                        res = handler(**step.arguments)\n                        if isinstance(res, dict):\n                            result_text.update(res)\n                    except Exception as e:\n                        result_text = {"status": "error", "error": str(e)}\n                else:\n                    result_text = {"status": "error", "error": f"Unknown action: {step.name}"}\n                    \n                print(f"[function_result] {result_text}")\n                \n                if "safety_decision" in step.arguments:\n                    # auto approve safety decisions for demo\n                    result_text["safety_acknowledgement"] = True\n \n                screenshot_bytes = bridge.screenshot()\n                \n                fr = {\n                    "type": "function_result",\n                    "name": step.name,\n                    "call_id": step.id,\n                    "result": [\n                        {"type": "text", "text": json.dumps(result_text)},\n                        {\n                            "type": "image",\n                            "data": base64.b64encode(screenshot_bytes).decode(),\n                            "mime_type": "image/png",\n                        },\n                    ],\n                }\n                function_responses.append(fr)\n            else:\n                print(f"\\nResult: {interaction.output_text}")\n                break\n                \n        user_input = function_responses\n        previous_interaction_id = interaction.id\n        \n        if not function_responses:\n            break\n            \n    return interaction\n \n \nif __name__ == "__main__":\n    task_desc = "Find the latest blog post from philipp schmid and summarize it."\n    if len(sys.argv) > 1:\n        task_desc = " ".join(sys.argv[1:])\n    run_agent(task_desc)\n
\n

How to run it

\n

Bash

\n
# Set your API key\nexport GEMINI_API_KEY="your-key"\n \n# Run the agent (starts the emulator automatically if needed)\npython agent.py "Open Settings and enable dark mode"\n
\n

Connecting to a Remote Device

\n

You can also target physical Android devices or remote cloud emulators instead of a local virtual device.

\n
    \n
  1. \n

    Enable USB/Wireless Debugging: On your target device, enable Developer Options and turn on USB Debugging or Wireless Debugging.

    \n
    2. \n
  2. \n

    Connect via ADB: Use the adb connect command to link to a remote device over TCP/IP:

    \n

    Bash

    \n
    adb connect <device-ip-address>:5555\n
    \n

    For details on setting up wireless debugging, refer to the official Android developer documentation on ADB over Wi-Fi.

    \n
    3. \n
  3. \n

    Pass the Device ID to the Agent: Pass the remote device's connection string as the device_id parameter to target it in the agent loop:

    \n

    Python

    \n
    # Target a remote or cloud-hosted emulator\nrun_agent("Check the weather", device_id="35.200.100.10:5555")\n
    \n
    4. \n
\n

Next Steps & Developer Tips

\n

I hope this gets you started with Gemini 3.5 Flash Computer Use on mobile. Next steps can be:

\n\n
\n

Thanks for reading! If you have any questions or feedback, please let me know on Twitter or LinkedIn.

\n", "content_markdown": "June 25, 2026 9 minute read [View Code](https://github.com/google-gemini/gemini-android-computer-use-quickstart)\n\nGemini 3.5 Flash has [built-in Computer Use](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-computer-use-gemini-3-5-flash/). The model looks at a screenshot, decides what to do, and returns a function call like `click(y=300, x=500)`. You execute that action on the device via ADB, take a new screenshot, and send it back. Repeat until the task is done.\n\nThis guide walks you through controlling an Android emulator using the `mobile` environment and the Python SDK.\n\nGithub Repository: [https://github.com/google-gemini/gemini-android-computer-use-quickstart](https://github.com/google-gemini/gemini-android-computer-use-quickstart)\n\n## What is Computer Use?\n\n[Computer Use](https://ai.google.dev/gemini-api/docs/computer-use) is a native tool in Gemini 3.5 Flash. You give the model a screenshot and a goal (\"open Settings and turn on dark mode\"). The model returns structured actions: taps, text input, swipes, app launches. Your code executes those actions on the target device.\n\nIt works like function calling. The model proposes actions, you execute them, you send back the result. The model stays in the loop through screenshots.\n\nGemini supports three Computer Use environments: `browser` for desktop web automation, `mobile` for mobile devices/emulators, and `desktop` for operating system-level control. This guide uses `mobile`.\n\n```plaintext\n Screenshot ──> Gemini 3.5 Flash ──> function_call\n ▲ │\n │ ▼\n └──────── Execute via ADB ◄──────────┘\n```\n\n### Pseudocode\n\nPython\n\n```python\nfrom google import genai\n \nclient = genai.Client()\nbridge = ADBBridge()\n \n# Take initial screenshot and send first request\nscreenshot = bridge.screenshot()\n \ninteraction = client.interactions.create(\n model=\"gemini-3.5-flash\",\n input=[\n {\"type\": \"text\", \"text\": \"Open Settings and enable dark mode\"},\n {\"type\": \"image\", \"data\": b64(screenshot), \"mime_type\": \"image/png\"},\n ],\n tools=[{\"type\": \"computer_use\", \"environment\": \"mobile\"}],\n)\n \n# Agent loop: execute actions, send results back\nwhile interaction has function_calls:\n for call in interaction.function_calls:\n bridge.execute(call.name, call.args) # click, type, open_app...\n \n screenshot = bridge.screenshot()\n interaction = client.interactions.create(\n model=\"gemini-3.5-flash\",\n previous_interaction_id=interaction.id,\n input=[function_results + screenshot],\n tools=[{\"type\": \"computer_use\", \"environment\": \"mobile\"}],\n )\n \nprint(interaction.output_text)\n```\n\n### Supported actions (mobile environment)\n\nThese are the actions the model can request when using `environment: \"mobile\"`. Your code must handle each one.\n\n| Action | Description | Arguments |\n| ----------------- | ---------------------------------- | -------------------------------------------------------------- |\n| `open_app` | Opens an app by name | `app_name: str`, `intent: str` (optional) |\n| `click` | Tap a screen coordinate | `y: int (0-999)`, `x: int (0-999)` |\n| `type` | Type text | `text: str`, `press_enter: bool` (default false) |\n| `long_press` | Long-press a coordinate | `y: int (0-999)`, `x: int (0-999)`, `seconds: int` (default 2) |\n| `drag_and_drop` | Drag between two points | `start_y`, `start_x`, `end_y`, `end_x` |\n| `press_key` | Press a key (home, back, enter...) | `key: str` |\n| `go_back` | Navigate back | (none) |\n| `wait` | Pause execution | `seconds: int` (default 1) |\n| `list_apps` | List installed apps | (none) |\n| `take_screenshot` | Capture the screen | (none) |\n\nCoordinates are normalized to a 0-999 grid. `(0, 0)` is top-left, `(999, 999)` is bottom-right. Your bridge converts these to actual pixel coordinates based on the device resolution.\n\n## Setup\n\nNo Android Studio GUI required. Run the [setup script](https://github.com/google-gemini/gemini-android-computer-use-quickstart/blob/main/setup_emulator.sh) on your Mac to install the Android SDK, emulator, and create a virtual device from the terminal.\n\n### 1\\. Run the setup script\n\nInstall script link: [https://github.com/google-gemini/gemini-android-computer-use-quickstart/blob/main/setup_emulator.sh](https://github.com/google-gemini/gemini-android-computer-use-quickstart/blob/main/setup_emulator.sh)\n\nBash\n\n```bash\nchmod +x setup_emulator.sh\n./setup_emulator.sh\n```\n\n### 2\\. Install Python dependencies\n\nBash\n\n```bash\npip install google-genai\n```\n\nThe Python agent script handles locating SDK paths and starting the emulator in the background automatically, so no additional manual environment variable exporting or terminal setup is required.\n\n## The agent loop\n\nFull working script [agent.py](https://github.com/google-gemini/gemini-android-computer-use-quickstart/blob/main/agent.py). It automatically sets up the environment variables, launches the emulator in the background (if it isn't already running), waits for it to boot, and begins the Computer Use loop.\n\nPython\n\n```python\nimport base64\nimport json\nimport os\nimport re\nimport subprocess\nimport sys\nimport time\n \nfrom google import genai\n \nBASE_DIR = os.path.dirname(os.path.abspath(__file__))\n \n \ndef setup_android_env():\n paths_to_check = [\n os.environ.get(\"ANDROID_HOME\"),\n \"/opt/homebrew/share/android-commandlinetools\",\n \"/usr/local/share/android-commandlinetools\",\n ]\n \n android_home = None\n for p in paths_to_check:\n if p and os.path.exists(p):\n android_home = p\n break\n \n if not android_home:\n print(\"Error: ANDROID_HOME not found. Run setup_emulator.sh first.\", file=sys.stderr)\n sys.exit(1)\n \n os.environ[\"ANDROID_HOME\"] = android_home\n sdk_paths = [\n os.path.join(android_home, \"cmdline-tools\", \"latest\", \"bin\"),\n os.path.join(android_home, \"emulator\"),\n os.path.join(android_home, \"platform-tools\"),\n ]\n \n current_path = os.environ.get(\"PATH\", \"\")\n for p in sdk_paths:\n if p not in current_path:\n current_path = p + os.pathsep + current_path\n \n os.environ[\"PATH\"] = current_path\n return android_home\n \n \ndef start_emulator(avd_name=\"AI_Agent_Phone\"):\n setup_android_env()\n \n try:\n res = subprocess.run([\"adb\", \"devices\"], capture_output=True, text=True)\n if \"emulator\" in res.stdout:\n return\n except FileNotFoundError:\n pass\n \n print(f\"Starting emulator '{avd_name}'...\")\n log_file = open(os.path.join(BASE_DIR, \"emulator.log\"), \"w\")\n \n subprocess.Popen(\n [\"emulator\", \"-avd\", avd_name, \"-delay-adb\"],\n stdout=log_file,\n stderr=log_file,\n start_new_session=True,\n )\n \n print(\"Waiting for emulator to boot...\")\n for _ in range(60):\n try:\n res = subprocess.run([\"adb\", \"devices\"], capture_output=True, text=True)\n if \"emulator\" in res.stdout:\n boot_res = subprocess.run(\n [\"adb\", \"shell\", \"getprop\", \"sys.boot_completed\"],\n capture_output=True,\n text=True,\n )\n if boot_res.stdout.strip() == \"1\":\n print(\"Emulator ready.\")\n return\n except Exception:\n pass\n time.sleep(2)\n \n print(\"Error: Emulator failed to boot.\", file=sys.stderr)\n sys.exit(1)\n \n \nclass ADBBridge:\n def __init__(self, device_id=None):\n self.prefix = [\"adb\"] + ([\"-s\", device_id] if device_id else [])\n self.width, self.height = self._screen_size()\n \n def _run(self, args, check=True):\n result = subprocess.run(self.prefix + args, capture_output=True, text=True)\n if check and result.returncode != 0:\n raise RuntimeError(f\"ADB error: {result.stderr.strip()}\")\n return result.stdout\n \n def _screen_size(self):\n output = self._run([\"shell\", \"wm\", \"size\"])\n match = re.search(r\"Physical size: (\\d+)x(\\d+)\", output)\n return (int(match.group(1)), int(match.group(2))) if match else (1080, 1920)\n \n def _px(self, x, y):\n return int(x / 1000 * self.width), int(y / 1000 * self.height)\n \n def click(self, y, x, **_):\n px, py = self._px(x, y)\n self._run([\"shell\", \"input\", \"tap\", str(px), str(py)])\n \n def type(self, text, press_enter=False, **_):\n self._run([\"shell\", \"input\", \"text\", text.replace(\" \", \"%s\")])\n if press_enter:\n self._run([\"shell\", \"input\", \"keyevent\", \"66\"])\n \n def open_app(self, app_name=None, package_name=None, **_):\n pkg = app_name or package_name\n if not pkg:\n raise ValueError(\"open_app requires app_name or package_name\")\n \n stdout = self._run([\"shell\", \"monkey\", \"--pct-syskeys\", \"0\", \"-p\", pkg,\n \"-c\", \"android.intent.category.LAUNCHER\", \"1\"], check=False)\n if \"No activities found\" in stdout or \"monkey aborted\" in stdout:\n raise RuntimeError(f\"App {pkg} is not installed or has no launcher activity.\")\n \n def scroll(self, y, x, direction, magnitude=800, **_):\n px, py = self._px(x, y)\n dist = int(magnitude / 1000 * self.height)\n dx, dy = {\"up\": (0, -dist), \"down\": (0, dist),\n \"left\": (-dist, 0), \"right\": (dist, 0)}.get(direction, (0, 0))\n self._run([\"shell\", \"input\", \"swipe\", str(px), str(py),\n str(px + dx), str(py + dy), \"300\"])\n \n def long_press(self, y, x, seconds=2, **_):\n px, py = self._px(x, y)\n self._run([\"shell\", \"input\", \"swipe\", str(px), str(py),\n str(px), str(py), str(seconds * 1000)])\n \n def drag_and_drop(self, start_y, start_x, end_y, end_x, **_):\n sx, sy = self._px(start_x, start_y)\n ex, ey = self._px(end_x, end_y)\n self._run([\"shell\", \"input\", \"swipe\", str(sx), str(sy),\n str(ex), str(ey), \"300\"])\n \n def press_key(self, key, **_):\n keymap = {\"home\": \"3\", \"back\": \"4\", \"enter\": \"66\",\n \"app_switch\": \"187\", \"menu\": \"82\"}\n self._run([\"shell\", \"input\", \"keyevent\", keymap.get(key.lower(), key)])\n \n def go_back(self, **_):\n self._run([\"shell\", \"input\", \"keyevent\", \"4\"])\n \n def wait(self, seconds=1, **_):\n time.sleep(seconds)\n \n def list_apps(self, **_):\n output = self._run([\"shell\", \"pm\", \"list\", \"packages\", \"-3\"])\n apps = [l.split(\":\")[1] for l in output.splitlines() if l.startswith(\"package:\")]\n if not apps:\n return {\"apps\": \"No third-party apps installed on this device.\"}\n return {\"apps\": apps}\n \n def take_screenshot(self, **_):\n return None\n \n def screenshot(self) -> bytes:\n result = subprocess.run(\n self.prefix + [\"exec-out\", \"screencap\", \"-p\"], capture_output=True\n )\n return result.stdout\n \n \nSYSTEM_PROMPT = \"\"\"You are operating an Android phone.\n* Use the provided tools to complete the task.\n* Scroll down to inspect the full screen before assuming an element is missing.\n* You can open apps by package name from anywhere.\n* Type text only using the `type` tool. Do not use the virtual keyboard.\n* If the task is already complete, state that directly.\n\"\"\"\n \n \ndef run_agent(task: str, device_id: str = None, max_turns: int = 100):\n start_emulator()\n client = genai.Client()\n bridge = ADBBridge(device_id)\n \n print(f\"\\nTask: {task}\")\n print(\"-\" * 40)\n \n screenshot_bytes = bridge.screenshot()\n user_input = [\n {\"type\": \"text\", \"text\": task},\n {\n \"type\": \"image\",\n \"data\": base64.b64encode(screenshot_bytes).decode(),\n \"mime_type\": \"image/png\",\n },\n ]\n \n previous_interaction_id = None\n turn = 0\n \n while turn < max_turns:\n turn += 1\n \n interaction = client.interactions.create(\n model=\"gemini-3.5-flash\",\n system_instruction=SYSTEM_PROMPT,\n input=user_input,\n tools=[{\"type\": \"computer_use\", \"environment\": \"mobile\"}],\n previous_interaction_id=previous_interaction_id,\n )\n \n function_responses = []\n for step in interaction.steps:\n if step.type == \"function_call\":\n print(f\"[function_call] {step.name}({step.arguments})\")\n handler = getattr(bridge, step.name, None)\n result_text = {\"status\": \"ok\"}\n \n if handler:\n try:\n res = handler(**step.arguments)\n if isinstance(res, dict):\n result_text.update(res)\n except Exception as e:\n result_text = {\"status\": \"error\", \"error\": str(e)}\n else:\n result_text = {\"status\": \"error\", \"error\": f\"Unknown action: {step.name}\"}\n \n print(f\"[function_result] {result_text}\")\n \n if \"safety_decision\" in step.arguments:\n # auto approve safety decisions for demo\n result_text[\"safety_acknowledgement\"] = True\n \n screenshot_bytes = bridge.screenshot()\n \n fr = {\n \"type\": \"function_result\",\n \"name\": step.name,\n \"call_id\": step.id,\n \"result\": [\n {\"type\": \"text\", \"text\": json.dumps(result_text)},\n {\n \"type\": \"image\",\n \"data\": base64.b64encode(screenshot_bytes).decode(),\n \"mime_type\": \"image/png\",\n },\n ],\n }\n function_responses.append(fr)\n else:\n print(f\"\\nResult: {interaction.output_text}\")\n break\n \n user_input = function_responses\n previous_interaction_id = interaction.id\n \n if not function_responses:\n break\n \n return interaction\n \n \nif __name__ == \"__main__\":\n task_desc = \"Find the latest blog post from philipp schmid and summarize it.\"\n if len(sys.argv) > 1:\n task_desc = \" \".join(sys.argv[1:])\n run_agent(task_desc)\n```\n\n### How to run it\n\nBash\n\n```bash\n# Set your API key\nexport GEMINI_API_KEY=\"your-key\"\n \n# Run the agent (starts the emulator automatically if needed)\npython agent.py \"Open Settings and enable dark mode\"\n```\n\n## Connecting to a Remote Device\n\nYou can also target physical Android devices or remote cloud emulators instead of a local virtual device.\n\n1. **Enable USB/Wireless Debugging**: On your target device, enable Developer Options and turn on USB Debugging or Wireless Debugging.\n2. **Connect via ADB**: Use the `adb connect` command to link to a remote device over TCP/IP:\n\n Bash\n\n ```bash\n adb connect :5555\n ```\n\n For details on setting up wireless debugging, refer to the official [Android developer documentation on ADB over Wi-Fi](https://developer.android.com/tools/adb#wireless).\n3. **Pass the Device ID to the Agent**: Pass the remote device's connection string as the `device_id` parameter to target it in the agent loop:\n\n Python\n\n ```python\n # Target a remote or cloud-hosted emulator\n run_agent(\"Check the weather\", device_id=\"35.200.100.10:5555\")\n ```\n\n## Next Steps & Developer Tips\n\nI hope this gets you started with Gemini 3.5 Flash Computer Use on mobile. Next steps can be:\n\n- **Supporting iOS / iPhone**: The Gemini API's `mobile` environment is platform-agnostic. The model outputs actions (clicks, swipes, types) on the same normalized `0-999` grid regardless of whether the device is Android or iOS. To target an iPhone or iOS Simulator, you only need to swap out `ADBBridge` with an iOS-compatible tool—such as Apple's `simctl` CLI for simulator control, Appium, or [go-ios](https://github.com/danielpaulus/go-ios) for physical devices.\n- **Production Robustness**: The Python bridge code provided here is synchronous and optimized for demonstration. For production use, you should implement robust retry logic for network drops, handle ADB disconnects gracefully, and execute operations asynchronously.\n- **Handling Safety Decisions**: In real-world tasks (especially those modifying state or making payments), the model might flag action steps with a `safety_decision` requesting confirmation. Ensure your production loop inspects `step.arguments` for safety flags and prompts the user before executing the action. See the [Gemini API Computer Use Safety Guidelines](https://ai.google.dev/gemini-api/docs/computer-use#safety-decisions) for details.\n\n* * *\n\nThanks for reading! If you have any questions or feedback, please let me know on [Twitter](https://twitter.com/_philschmid) or [LinkedIn](https://www.linkedin.com/in/philipp-schmid-a6a2bb196/).\n", "dateCreated": "2026-09-27T19:29:15.293681927+00:00", "date_created": "2026-06-25T00:00:00+00:00", "date_modified": "2026-06-25T00:00:00+00:00", "date_published": "2026-06-25T00:00:00+00:00", "description": "Control an Android emulator using Gemini 3.5 Flash Computer Use. Connect the Google GenAI SDK interactions loop with ADB to control a virtual device from your terminal.", "external_url": "https://www.philschmid.de/gemini-android-use", "git": { "commit": "ebb0941c5c06a2ab2377342d607535262a3e969e", "path": "items/philschmid-de/2026/06/2026-06-25-control-an-android-phone-with-gemini-3-5-flash-computer-use.md", "permalink": "https://github.com/aymericbeaumet/aggr-instance/blob/ebb0941c5c06a2ab2377342d607535262a3e969e/items/philschmid-de/2026/06/2026-06-25-control-an-android-phone-with-gemini-3-5-flash-computer-use.md" }, "id": "urn:aggr:item:fbfce7e022024d5e22915e897e6085c3dc08beb5", "isBasedOn": { "@id": "https://www.philschmid.de/gemini-android-use" }, "mainEntity": { "@id": "https://www.philschmid.de/gemini-android-use", "@type": "CreativeWork", "archivedAt": { "@id": "https://aggr.aymericbeaumet.com/items/philschmid-de/2026-06-25-control-an-android-phone-with-gemini-3-5-flash-computer-use/#webpage" }, "dateModified": "2026-06-25T00:00:00+00:00", "datePublished": "2026-06-25T00:00:00+00:00", "headline": "Control an Android Phone with Gemini 3.5 Flash Computer Use", "timeRequired": "PT8M", "url": "https://www.philschmid.de/gemini-android-use", "wordCount": 1771 }, "name": "Archived snapshot: Control an Android Phone with Gemini 3.5 Flash Computer Use", "preview": null, "reading_minutes": 8, "replicated_at": null, "source": "philschmid-de", "source_name": "philschmid.de - RSS feed", "summary": "Control an Android emulator using Gemini 3.5 Flash Computer Use. Connect the Google GenAI SDK interactions loop with ADB to control a virtual device from your terminal.", "tags": [], "temporalCoverage": "2026-09-27T19:29:15.293681927+00:00", "title": "Control an Android Phone with Gemini 3.5 Flash Computer Use", "url": "https://aggr.aymericbeaumet.com/items/philschmid-de/2026-06-25-control-an-android-phone-with-gemini-3-5-flash-computer-use/", "word_count": 1771 }