https://github.com/ChenKuanSun/MobileClaw/compare/v1.2.4...v1.2.5/

MobileClaw is the Android port of OpenClaw. It turns your phone into an AI agent — tell it what to do in your own language and it taps, types, reads, calls, and navigates for you. Runs with cloud AI (Claude, GPT, Gemini, 10 providers) or entirely offline with Gemma 4 on-device. No server. No subscription. Your phone does everything. Why MobileClaw? MobileClaw: - Runs locally on your phone. - Controls any app through accessibility actions. - Supports on-device AI and offline workflows. - Open source and easy to fork. - Works with multiple providers and your own keys. - Extensible through skills you can create yourself. On-Device AI Run Gemma 4 directly on your phone via LiteRT-LM. No API key. No internet. No cost. - E2B (2.6 GB) — 6 GB+ RAM, ~9 tok/s on GPU - E4B (3.7 GB) — 8 GB+ RAM, higher quality responses - GPU accelerated with NPU/CPU fallback (auto-detected) - Tool calling works on-device via constrained decoding - Download models from Settings, switch between cloud and local anytime Tools 29 native tools that directly access Android APIs: - SMS - Call Log - Contacts - Phone Call - Calendar - Alarms - Notifications - App Launcher - Navigation - UI Automation - Screen Capture - Web Browser - HTTP API - File System - Photos - Clipboard - Media Control - Volume - Brightness - Flashlight - System Info - Scheduled Tasks - Skill Author - Memory - Session History - Sub-Agent - Channel - Telegram - OpenAI Every tool works with both cloud and on-device models. Skills 21 composable skills (Markdown + YAML frontmatter): Built-in: morning-routine, email, messaging-apps, navigation, notification-digest, phone-basics, photos, self-learning, social-media, telegram-bot, translation, ui-fallback, weather, web-research Vertical: finance, health, notion, real-estate, shopping, smart-home, telegram Create your own: describe what you want, and the AI writes the skill for you (/create command). License MobileClaw is licensed under the MIT License.