MobileClaw for Cloud Phone is an AI automation framework based on the AgentBay platform. It combines the Qwen Large Language Model (LLM) with high-performance, cloud-based Android container technology. MobileClaw transforms Cloud Phone from a remote mobile window that executes commands into an intelligent agent that can think and complete tasks independently. MobileClaw runs AI tasks in an isolated cloud sandbox, which prevents the use of on-premises device resources and avoids screen interruptions. Tasks continue to run stably during events such as incoming calls or application switching. You can order food, plan travel, or perform cross-application operations. All these tasks are completed silently in the cloud. This provides a seamless local interaction experience and ensures highly reliable, uninterrupted task execution.
Service architecture
Core features
Multi-agent support: Empowers phones with execution capabilities
MobileClaw is an AI Agent framework, purpose-built and fine-tuned for the Cloud Phone Android environment. It supports multiple agent types, including MobileUseAgent and OpenClaw. Running within a cloud-based container, MobileClaw enables autonomous, closed-loop operations and simulates human-like interactions with high fidelity.
Cross-application flow: The AI can move beyond application borders. It automatically completes the end-to-end flow, from information extraction and data processing to the final operation.
24/7 operation: The AI agent in the cloud continues to execute key tasks based on preset logic, even if the on-premises device is disconnected.
Native OpenClaw integration: Brings Android Agents to the Open AI ecosystem
MobileClaw is deeply integrated with the OpenClaw open-source agent ecosystem. It pioneers the seamless integration of MobileUseAgent—an Android agent fine-tuned for Cloud Phone—into OpenClaw's unified collaborative framework. This elevates automation from isolated, single-point tasks to a distributed, collaborative intelligence model.
High-precision UI parsing: Enhances agent accuracy by precisely parsing the native Android UI tree, enabling reliable interaction even with applications that do not expose public APIs.
Low-latency state sync: Achieves millisecond-level synchronization between the cloud-based agent and the local device state, ensuring operational consistency and integrity.
Deep integration with Qwen: Empowers phones with thinking capabilities
MobileClaw natively integrates the capabilities of the Qwen LLM. Backed by large-scale computing power in the cloud, the AI can perceive screen content in real time and understand complex natural language instructions.
Intent recognition: The AI can directly understand user goals from natural language instructions without relying on complex hierarchical menus.
Intelligent task decomposition: The AI automatically breaks down complex tasks into executable Android system commands.
Excellent Android container technology: Provides a highly native user experience
Based on the container and virtual machine architecture of Cloud Phone, MobileClaw provides a near-native user experience in the cloud. It also offers highly customized Android services for the AI agent.
High-performance graphics rendering: Supports high frame rate and low-latency image transmission to ensure smooth interaction.
Elastic scaling: Compute resources are dynamically adjusted on demand. This allows for responses in seconds to the needs of different scenarios, from lightweight office work to high-payload AI computing.
Strict security administration: Provides reliable security and control
Built on the solid security foundation of Cloud Phone, MobileClaw provides end-to-end security control from the underlying network to the upper-layer applications.
Closed-loop data processing in the cloud: All AI computing and business logic are completed within a controlled cloud container. The on-premises device only receives encrypted instruction streams and pixel screens. This ensures that sensitive data never leaves the cloud, eliminating the risk of data leaks at the source.
Fine-grained custom images: Supports custom application installation and network access policies using blacklists and whitelists. This ensures that enterprise-level access borders remain compliant.
Use cases
Enterprise process automation
Business scenario: Many manual operations within a company are cross-application and highly repetitive, such as exporting data from a business system and organizing it into a DingTalk report every day, or regularly checking if an app's online features are available. These operations are time-consuming and prone to errors.
Solution: Use MobileClaw to automate these flows. For example, you can issue a command such as "At 9:00 AM every day, log on to the business system and send a screenshot of the sales data to the DingTalk group." The agent then automatically executes this task 24/7 in the cloud without using the on-premises device. This improves work efficiency and reduces labor costs and operational errors.
Automated testing and inspection
Business scenario: Mobile application versions iterate quickly, which requires frequent regression testing to ensure core features are stable. Traditional automated testing scripts are costly to maintain and struggle to cover complex user interaction scenarios.
Solution: Use MobileClaw as an automated testing tool. You can describe test use cases in natural language, such as "Test the complete flow from registration and logon to placing an order." The agent automatically executes the test and records the results. You can also use it for 24/7 online business network probes to promptly detect failures in the production environment. This reduces the cost of developing and maintaining test scripts, expands test coverage, and enables continuous online quality monitoring.
Authentic social interaction
Business scenario: User engagement on social media apps depends on high-frequency, natural interactions (such as replying to direct messages and participating in group chats). However, managing this manually is costly and prone to omissions. Traditional scripted bots are easily detected as non-human, leading to rate-limiting or account suspension, and they are incapable of handling complex scenarios like cross-app interactions and multi-turn conversations.
Solution: Use MobileClaw as an intelligent social media app operations agent. Operations staff can issue natural language instructions, such as "Proactively greet five active fans every day." The agent autonomously understands the semantics, invokes the social app, and simulates human operations such as swiping, clicking, typing, and selecting emojis. The entire process requires no intervention from the on-premises device and is stable and reliable. You do not need to write complex UI scripts. The agent simulates human behavior in terms of pacing, response delay, and text-and-image interaction. This helps avoid platform risk control mechanisms. It runs independently in the cloud, unaffected by the user's device status, to maintain social popularity 24/7.
FAQ
Q: What is the difference between MobileClaw and traditional automated testing frameworks, such as Appium or UIAutomator?
A: The main differences are the driving method and the degree of intelligence.
Traditional frameworks: These rely on developers to write precise code scripts based on control IDs or XPath. If the application UI changes, the scripts can easily fail, which leads to high maintenance costs.
MobileClaw: This is driven by an LLM. It understands task intent through natural language and uses visual recognition to adapt to UI changes. This provides better generalization and robustness, which lower the barrier to creating and maintaining automated tasks.
Q: How does MobileClaw ensure the security of tasks executed in the cloud?
A: MobileClaw is designed to provide multiple security guarantees:
Data fencing: Each MobileClaw instance runs in an independent cloud container, and the environments are isolated from each other.
Cloud-based processing and storage: All task logic and data processing are completed in the cloud. Sensitive information, such as account passwords, does not leave the cloud environment. The on-premises device only receives encrypted video streams and sends operation commands.
Granular control: Supports fine-grained control through custom images and network allowlists/denylists. This enables you to strictly limit the applications an Agent can install and its network access permissions, ensuring its behavior fully complies with corporate security policies.
Q: Does using MobileClaw violate the user agreements of third-party applications?
A: MobileClaw is a neutral technical tool, and its use is determined by the user. Using it for scenarios such as automated testing and internal process automation is usually compliant. However, if it is used for purposes that may disrupt the normal order of a target application, such as for game plug-ins, malicious crawlers, batch registrations, or flash sales, it is highly likely to violate that application's terms of service. This may result in consequences such as account suspension. Before use, you must read and comply with the relevant regulations of the target application.