PokeClaw - An open-source, fully local, offline mobile AI agent application.

What is PokeClaw?

PokeClaw (full name PocketClaw) is an open-source Android AI automation application independently developed by Nicole. It is the first mobile AI agent to achieve fully local offline operation, using the Google Gemma 4 model and the LiteRT-LM runtime for direct on-device inference, autonomously controlling the phone without internet connection, account, or API keys. By reading the screen UI tree through Android Accessibility, the AI can perform actions such as clicking, swiping, typing, and sending messages, achieving a true closed-loop agent. All data processing is completed locally, completely protecting privacy; it can also be equipped with cloud models such as OpenAI and Anthropic to enhance its ability to handle complex tasks. It supports features such as WhatsApp context-aware auto-reply, notification aggregation, and cross-application operations, and only requires Android 9+ and 8GB of RAM to run.

PokeClaw's main functions

• Fully local offline AI operationIt enables on-device inference based on the Google Gemma 4 model, requiring no internet connection, account, or API key, and the data remains on the phone.
• Cloud-based model optional enhancementSupports connection to APIs such as OpenAI, Anthropic, and Google to obtain stronger inference capabilities, and displays token consumption and costs in real time.
• Mobile phone automation controlIt reads the screen UI tree through Android Accessibility Service and autonomously performs operations such as clicking, swiping, typing, sending messages, and opening applications.
• Context-aware auto-responseMonitor message notifications from apps like WhatsApp, automatically open the chat interface, read historical conversations, and generate intelligent replies.
• Skills SystemIt provides preset reusable workflows that combine common tools into specific task sequences, such as automatic replies, morning briefings, and smart forwarding.
• System Information QueryIt provides real-time reports on battery status, storage space, phone temperature, Bluetooth status, installed apps, Android version, and more.
• Multilingual and device compatibilityThe interface supports English, Hindi, Japanese, German, and Traditional Chinese, and is compatible with chips such as Tensor G3/G4, Snapdragon 8 Gen 2+, and Dimensity 9200+.

How to use PokeClaw

• Download and installInstall the APK file from GitHub Releases: https://github.com/agents-io/PokeClaw. It will automatically download approximately 2.6GB of local model on first launch.
• Grant permissionsEnable accessibility service permissions (required, for screen reading and operation) and notification access permissions (recommended, for background message monitoring).
• Select running modeLocal mode requires no configuration and can be used without any setup, while cloud mode requires you to enter the API key for OpenAI, Anthropic, or Google in the settings.
• Chat interactionSend commands in natural language in the input box at the bottom.
• Task CardIn the task interface, click the skill card (such as "Monitor WhatsApp Messages" or "Send Message"), fill in the contact name and content, and then execute.
• Running in the backgroundWhen automatic reply monitoring is enabled, keep the application running in the background, and the AI will automatically detect new messages and execute the reply process.

PokeClaw's core advantages

• Run completely offlineBased on the Google Gemma 4 model, it performs direct inference on the device, allowing users to access core functions without an internet connection, account, or API key. All data processing is completed within the phone, fundamentally eliminating the risk of privacy leaks.
• Privacy-first architectureIn local mode, the user's screen content, chat history, and personal data are not transmitted to any external servers. Sensitive information is always kept on the device, achieving true data sovereignty.
• Context-aware intelligenceThe auto-reply function not only monitors message notifications, but also proactively opens the chat interface to read the complete conversation history and generates an appropriate reply based on the preceding context, rather than simply triggering a fixed template.
• Skills SystemIt combines common tools (clicks, swipes, input, etc.) into reusable workflows (such as auto-response and morning briefings), enabling small models to reliably perform complex multi-step tasks, while also supporting user-defined skill extensions.
• Costs are transparent and controllableThe cloud mode displays token consumption and operating costs in real time, and supports setting software/hard budget limits, allowing users to monitor resource usage at any time during complex task execution.
• Zero-threshold deploymentNo root access or developer account is required. The APK can be used immediately after installation. It supports mainstream devices with Android 9+ and 8GB or more of RAM, and is compatible with various chip platforms such as Tensor, Snapdragon, and Dimensity.

PokeClaw Key Information and Usage Requirements

• Project natureThis is an open-source Android AI automation application, licensed under the Apache 2.0 license. It is independently developed and maintained by developer Nicole, and the code is hosted on GitHub (agents-io/PokeClaw).
• Core positioningThe world's first fully local offline mobile AI agent (on-device AI Agent) that can autonomously control Android devices to perform complex tasks without cloud connectivity.
• Technical foundationBased on the Google Gemma 4 model (inference via LiteRT-LM runtime), it supports native tool calling, has a model size of approximately 2.6GB, and performs pure local inference without GPU acceleration (CPU can run).
• Operating modeLocal-first architecture, default offline operation to ensure privacy; optional access to cloud APIs such as OpenAI, Anthropic, and Google to enhance complex task processing capabilities, and support seamless model switching within a session.
• Control methodBy obtaining screen UI tree information through Android Accessibility Service, AI autonomously makes decisions and executes operations such as clicking, swiping, inputting, and sending messages, forming a closed-loop proxy process.
• Core FunctionsContext-aware auto-response (supports WhatsApp), notification aggregation, clipboard analysis, system information query, cross-application operations, and a reusable workflow system based on the Skills architecture.
• System RequirementsAndroid 9.0 and above, arm64 architecture chip, minimum 8GB RAM (12GB+ recommended), at least 3GB free storage space (for initial model download), no root access required.
• Hardware performancePure CPU devices can run the program, but startup takes approximately 45 seconds; devices equipped with Tensor G3/G4 (Pixel 8/9), Snapdragon 8 Gen 2+, or Dimensity 9200+ chips are recommended for near-instantaneous response times.
• Required permissionsAccessibility service permissions must be granted (for screen reading and automated operations), and notification access permissions are recommended (for background message monitoring and automatic reply functionality).
• Initial configurationWhen you select local mode for the first time after installing the APK, approximately 2.6GB of model files will be automatically downloaded; when you select cloud mode, you need to configure the corresponding API key (OpenAI, Anthropic, Google, etc.) in the settings.
• Language supportThe interface supports English, Hindi, Japanese, German, and Traditional Chinese. The skill system supports command input and application operations in any language.
• Network requirementsLocal mode runs completely offline and requires no network connection; cloud mode requires an internet connection to the corresponding API service; model download requires a stable network connection (supports resuming interrupted downloads).

Analysis of similar competing products of PokeClaw

Application scenarios of PokeClaw

• Message automation managementIt can monitor messages from specific contacts on apps like WhatsApp and automatically reply with context-aware messages. It also supports automatically sending follow-up text messages after a missed call. It is suitable for scenarios where manual replies are not possible, such as when driving or in a meeting, or helps users manage high-frequency social communications.
• Intelligent monitoring of equipment statusIt can monitor battery health, storage space usage, phone temperature, Bluetooth status, and installed applications in real time, automatically analyze and suggest cleanup targets, and is suitable for device maintenance, performance optimization, and troubleshooting.
• Cross-application information retrieval and integrationIt allows users to search for videos within the YouTube app, view trending topics on Twitter/X, find content on Reddit, and copy email subjects for Google searches. It is suitable for research and information gathering scenarios that require quick access to information from multiple platforms.
• Personal Schedule and Information BriefingExecutes the morning briefing process, automatically opens the weather app to read the temperature, checks the calendar to get the day's events, counts the number of unread emails and generates a summary, suitable for information summarization and schedule planning at the beginning of the day.
• Privacy-sensitive data processingIt analyzes clipboard content, summarizes notification information, and processes sensitive personal data in a completely offline mode, making it suitable for business, medical, or legal scenarios with extremely high privacy requirements, ensuring that sensitive information does not leave the device.
• Accessibility assistanceBy using natural language commands, AI can autonomously perform operations such as clicking, swiping, and input, helping users with mobility impairments or visual impairments to control mobile applications more conveniently and lowering the barrier to entry for smartphone use.