Summary

PalmClaw offers a private, secure, and always-available AI assistant deployable directly on Android phones. It targets mobile users and developers seeking to run AI agents locally, enhancing privacy and simplifying daily use without a PC.

How It Works

This project enables mobile-native deployment and operation on Android, utilizing a local-first runtime for enhanced safety and privacy. It integrates a full AI agent stack (memory, skills, tools, channels) within a single mobile runtime. The core loop involves input processing, LLM decision-making with tool calls, guided by context and memory, and outputting responses to sessions and channels.

Quick Start & Requirements

• Normal Users: Download the latest APK from Releases, install on Android, and follow in-app onboarding to configure a provider API key.
• Developers: Requires Android Studio and JDK 17. Clone the repo, open in Android Studio, sync Gradle, and run on a device/emulator.
• Prerequisites: Android device/emulator. Users must provide their own AI model API keys.

Highlighted Details

• Full mobile-native deployment and operation on Android.
• Local-first runtime ensures data privacy and security.
• Supports integration with Telegram, Discord, Slack, Feishu, Email, and WeCom channels.
• Includes a comprehensive agent stack: memory, skills, tools, and channels.
• Active development with frequent updates, recently focusing on channel stability, UI, and auto-updates.

Maintenance & Community

The project welcomes researchers, builders, and mobile AI practitioners. Active development is indicated by frequent updates and a roadmap including SkillHub integration, enhanced web search, deeper Android capabilities (screen reading, interaction), and multimodal support.

Licensing & Compatibility

PalmClaw uses a dual licensing model: AGPLv3 for open-source use (requiring network service modifications to be AGPLv3) and a separate Commercial License for proprietary integration without copyleft requirements.

Limitations & Caveats

The application requires users to provide their own AI model API keys, as it does not include hosted model access. Advanced features like screen reading, interaction, and multimodal I/O are still under development. The primary deployment target is Android.