Reference for core types and data structures used in the mobile-use SDK.
AgentProfile Represents a profile for the mobile-use agent with LLM configuration.
from minitap.mobile_use.sdk.types import AgentProfile Constructor AgentProfile( * , name: str , llm_config: LLMConfig | None = None , from_file: str | None = None , ) LLM configuration for the agent
Path to a file containing LLM configuration (JSONC format)
llm_config and from_file are mutually exclusive - use only one.
Examples # From file profile = AgentProfile( name = "default" , from_file = "llm-config.defaults.jsonc" ) # Programmatic from minitap.mobile_use.config import LLM , LLMConfig profile = AgentProfile( name = "custom" , llm_config = LLMConfig( planner = LLM( provider = "openai" , model = "gpt-5-nano" ), cortex = LLM( provider = "openai" , model = "gpt-5" ), # ... other components ) ) TaskRequest Represents a mobile automation task request.
from minitap.mobile_use.sdk.types import TaskRequest Attributes Natural language description of the task goal
Name of the agent profile to use
Name of the task for logging
Description of the expected output format
Pydantic model class for typed output
Maximum number of steps the agent can take
Whether to record execution traces
Directory to save trace data
Path to save agent thoughts
TaskRequest objects are typically created via TaskRequestBuilder:
task_request = ( agent.new_task( "Your goal" ) .with_name( "task_name" ) .build() ) Task request for execution via the Minitap Platform.
from minitap.mobile_use.sdk.types import PlatformTaskRequest With PlatformTaskRequest, you only reference a task by name. The SDK
automatically fetches the task configuration (goal, max_steps, output format)
and LLM profile from the platform, then executes the task and streams
observability data back.
It also automatically upload animated GIF traces at the end of the task to
cloud storage for easy viewing. See Observability &
Tracing for details. Constructor PlatformTaskRequest( task: str , profile: str | None = None , api_key: str | None = None , record_trace: bool = True , trace_path: Path = Path(tempfile.gettempdir()) / "mobile-use-traces" , llm_output_path: Path | None = None , thoughts_output_path: Path | None = None , max_steps: int = 400 ) Attributes Name of the LLM profile to use for this task. If not specified, uses the Minitap-managed default profile.
API key for authentication with the Minitap Platform. If not provided, uses the MINITAP_API_KEY environment variable.
When enabled, an animated GIF of the task execution is saved locally. In case
of PlatformTaskRequest, the GIF is also uploaded to Minitap cloud storage
for viewing in the platform.
Directory to save local trace files (steps and GIF run trajectory). Defaults
to system temp directory.
Path to save the final LLM output locally.
Path to save agent thoughts/reasoning locally.
Maximum number of steps (overridden by platform configuration).
PlatformTaskRequest does not support locked_app_package as a parameter. The app lock must be configured on the platform task itself at platform.minitap.ai/tasks .from minitap.mobile_use.sdk import Agent from minitap.mobile_use.sdk.types import PlatformTaskRequest agent = Agent() agent.init() # Simple task execution result = await agent.run_task( request = PlatformTaskRequest( task = "check-notifications" ) ) agent.clean() AgentConfig Configuration for the agent.
from minitap.mobile_use.sdk.types import AgentConfig Created via AgentConfigBuilder:
config = ( Builders.AgentConfig .with_default_profile(profile) .build() ) agent = Agent( config = config) Enum for device platforms.
from minitap.mobile_use.sdk.types import DevicePlatform # Available values DevicePlatform. ANDROID DevicePlatform. IOS config = ( Builders.AgentConfig .for_device( platform = DevicePlatform. ANDROID , device_id = "emulator-5554" ) .build() ) ServerConfig Configuration for agent servers.
from minitap.mobile_use.sdk.types import ServerConfig Attributes Hardware Bridge server URL
servers = ServerConfig( hw_bridge_base_url = "http://localhost:8001" , screen_api_base_url = "http://localhost:8000" , adb_server_host = "localhost" , adb_server_port = 5037 ) config = Builders.AgentConfig.with_servers(servers).build() LLMConfig Configuration for LLM models used by different agent components.
from minitap.mobile_use.config import LLM , LLMConfig, LLMConfigUtils, LLMWithFallback Structure llm_config = LLMConfig( planner = LLM( provider = "openai" , model = "gpt-5-nano" ), orchestrator = LLM( provider = "openai" , model = "gpt-5-nano" ), cortex = LLMWithFallback( provider = "openai" , model = "gpt-5" , fallback = LLM( provider = "openai" , model = "gpt-5" ) ), executor = LLM( provider = "openai" , model = "gpt-5-nano" ), utils = LLMConfigUtils( hopper = LLM( provider = "openai" , model = "gpt-5-nano" ), outputter = LLM( provider = "openai" , model = "gpt-5-nano" ) ) ) Components
planner Creates high-level plans from goals
orchestrator Coordinates execution steps
cortex Visual understanding and decision-making
executor Performs specific actions
hopper Extracts relevant information from large data batches
outputter Extracts structured output
Basic LLM configuration.
from minitap.mobile_use.config import LLM llm = LLM( provider = "openai" , model = "gpt-5" ) Provider name: openai, google, xai, openrouter
Model identifier (e.g., gpt-5, gemini-2.5-flash)
LLMWithFallback LLM configuration with a fallback model.
from minitap.mobile_use.config import LLM , LLMWithFallback llm = LLMWithFallback( provider = "openai" , model = "o4-mini" , fallback = LLM( provider = "openai" , model = "gpt-5" ) ) If the primary model fails, the fallback is used automatically.
TaskRequestCommon Common configuration shared across tasks.
from minitap.mobile_use.sdk.types import TaskRequestCommon Created via TaskDefaults builder:
defaults = ( Builders.TaskDefaults .with_max_steps( 500 ) .build() ) Next Steps 