Skip to content

README_EN.md

Latest commit

Β 

History

History
345 lines (249 loc) Β· 12.4 KB

README_EN.md

File metadata and controls

345 lines (249 loc) Β· 12.4 KB

πŸ“± AndroidForClaw

Release Android License: MIT

Let AI truly control your Android phone.

Architecture aligned with OpenClaw (280k+ Stars), bringing full AI Agent capabilities to your phone β€” see the screen, tap apps, run code, connect platforms.

πŸ“– Docs (Chinese) Β· πŸš€ Quick Start Β· πŸ’¬ Community Β· δΈ­ζ–‡ζ–‡ζ‘£

πŸ”₯ What Can AI Do for You

πŸ“± Control Any App

WeChat, Alipay, TikTok, Taobao, Maps… Anything you can do manually, AI can do too.

You: Open WeChat and send "See you tomorrow" to John
AI:  β†’ Open WeChat β†’ Search John β†’ Type message β†’ Send βœ…

πŸ”— Cross-App Workflows

You: Got an address in WeChat, navigate me there
AI:  β†’ Copy address from WeChat β†’ Open Maps β†’ Search β†’ Start navigation

🐧 Run Code

Python, Node.js, Shell β€” run directly on your phone:

You: Use Python to analyze the CSV in my Downloads folder
AI:  β†’ exec("python3 analyze.py") β†’ Return analysis results

🌐 Web Search & Fetch

You: Search for today's tech news
AI:  β†’ web_search("tech news") β†’ Return titles + links + summaries

πŸ’¬ Multi-Platform Messaging

Control your phone AI remotely via Feishu, Discord, Telegram, Slack and more:

Channel Status
Feishu βœ… Available
Discord βœ… Available
Telegram πŸ”§ Ready (config aligned with OpenClaw)
Slack πŸ”§ Ready (Socket / HTTP dual mode)
Signal πŸ”§ Ready (signal-cli integration)
WhatsApp πŸ”§ Ready

Each channel supports per-channel model override β€” pick a dedicated model from your configured providers.

πŸ€– MCP Server (For External Agents)

Built-in MCP Server (port 8399) exposes the phone's accessibility and screenshot capabilities to external agents via the standard MCP protocol:

Tools: get_view_tree / screenshot / tap / swipe / input_text / press_home / press_back / get_current_app

This is NOT used by AndroidForClaw itself β€” it's for external agents like Claude Desktop, Cursor, etc.

🧩 Skill Extensions

Search and install new capabilities from ClawHub, or create your own Skills:

You: What skills are available on ClawHub?
AI:  β†’ skills_search("") β†’ Show available skill list

⚑ Quick Start

Download & Install

Download from the Release page:

APK Description Required?
AndroidForClaw Main app (Accessibility Service, Agent, Gateway) βœ… Required
BrowserForClaw AI Browser (web automation) Optional
termux-app + termux-api Terminal (run Python/Node.js) Optional

3 Steps to Get Started

  1. Install β€” Download and install AndroidForClaw
  2. Configure β€” Open the app, enter an API Key (or skip to use built-in Key), enable Accessibility + Screen Capture permissions
  3. Chat β€” Talk directly in the app, or send messages via Feishu/Discord

πŸ’‘ First launch opens a setup wizard automatically. Default: OpenRouter + MiMo V2 Pro. One-click skip supported.

Termux Setup (Optional)

With Termux installed, AI can run Python/Node.js/Shell. Built-in one-click setup wizard:

Settings β†’ Termux Config β†’ Copy command β†’ Paste into Termux β†’ Done

πŸ—οΈ Architecture

324 source files Β· 62,000+ lines of code Β· 10 modules

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  Channels                                 β”‚
β”‚  Feishu Β· Discord Β· Telegram Β· Slack Β·    β”‚
β”‚  Signal Β· WhatsApp Β· In-app chat          β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Agent Runtime                            β”‚
β”‚  AgentLoop Β· 20 Tools Β· 20 Skills Β·       β”‚
β”‚  Context Management (4-layer) Β· Memory    β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Providers                                β”‚
β”‚  OpenRouter Β· MiMo Β· Gemini Β· Anthropic Β· β”‚
β”‚  OpenAI Β· Custom                          β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Android Platform                         β”‚
β”‚  Accessibility Β· Termux SSH Β· device tool β”‚
β”‚  MediaProjection Β· BrowserForClaw         β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Core Features

Feature Description
Playwright Mode Screen ops aligned with Playwright β€” snapshot gets UI tree + ref β†’ act operates elements
Unified exec Auto-routes to Termux (SSH) or built-in Shell, transparent to the model
Context Management 4-layer protection aligned with OpenClaw: limitHistoryTurns + tool result trimming + budget guard
Model Smart Routing Model ID normalization + Fallback Chain (multi-candidate degradation) + API Key rotation + Allowlist/Blocklist
Session Maintenance 30-day auto-cleanup, 500 entry cap, 10MB rotation, disk budget auto-reclaim
Skill System 20 built-in Skills editable on device, ClawHub online installation
Multi-model MiMo V2 Pro Β· DeepSeek R1 Β· Claude Sonnet 4 Β· Gemini 2.5 Β· GPT-4.1
MCP Server Expose accessibility/screenshot to external agents (port 8399, Streamable HTTP)
Per-channel Model Each messaging channel can independently select a model, fields aligned with OpenClaw types
Steer Injection Inject messages into a running Agent Loop mid-run via Channel (mid-run steering)

πŸ“‹ Full Capability Table

πŸ”§ 20 Tools

Tool Function Alignment
device Screen ops: snapshot/tap/type/scroll/press/open Playwright
read_file Read file contents OpenClaw
write_file Create or overwrite files OpenClaw
edit_file Precise file editing OpenClaw
list_dir List directory contents OpenClaw
exec Execute commands (Termux SSH / built-in Shell) OpenClaw
web_search Brave search engine OpenClaw
web_fetch Fetch web page content OpenClaw
javascript Execute JavaScript (QuickJS) OpenClaw
tts Text-to-speech (device TTS engine) OpenClaw
skills_search Search ClawHub skills OpenClaw
skills_install Install skills from ClawHub OpenClaw
memory_search Semantic memory search OpenClaw
memory_get Read memory snippets OpenClaw
config_get Read config entries OpenClaw
config_set Write config entries OpenClaw
list_installed_apps List installed apps Android-specific
install_app Install APK Android-specific
start_activity Launch Activity Android-specific
stop Stop the Agent Android-specific

🧩 20 Skills

Category Skills
Feishu Suite feishu Β· feishu-doc Β· feishu-wiki Β· feishu-drive Β· feishu-bitable Β· feishu-chat Β· feishu-task Β· feishu-perm Β· feishu-urgent
Search & Web browser Β· weather
Skill Management clawhub Β· skill-creator
Dev & Debug debugging Β· data-processing Β· session-logs
Config Management model-config Β· channel-config Β· install-app Β· model-usage

Skills are stored at /sdcard/.androidforclaw/skills/ β€” freely editable, addable, and removable.

πŸ’¬ Messaging Channels

Channel Status Features
Feishu βœ… Available WebSocket real-time, group/DM, 32 Feishu tools
Discord βœ… Available Gateway connection, group/DM
Telegram πŸ”§ Ready Bot API polling/webhook, model override, streaming
Slack πŸ”§ Ready Socket Mode / HTTP Mode, model override, streaming
Signal πŸ”§ Ready signal-cli daemon integration, model override
WhatsApp πŸ”§ Ready WhatsApp Business API, model override
In-app Chat βœ… Available Built-in chat UI

All channel config fields are aligned with OpenClaw TypeScript type definitions (types.slack.ts, types.telegram.ts, etc.).

πŸ€– Supported Models

Provider Models Notes
OpenRouter MiMo V2 Pro, Hunter Alpha, DeepSeek R1, Claude Sonnet 4, GPT-4.1 Recommended, built-in Key
Xiaomi MiMo MiMo V2 Pro, MiMo V2 Flash, MiMo V2 Omni Direct Xiaomi API
Google Gemini 2.5 Pro, Gemini 2.5 Flash Direct
Anthropic Claude Sonnet 4, Claude Opus 4 Direct
OpenAI GPT-4.1, GPT-4.1 Mini, o3 Direct
Custom Any OpenAI-compatible API Ollama, vLLM, etc.

Default: OpenRouter + MiMo V2 Pro (1M context + reasoning). Skip the wizard to auto-use built-in Key.

πŸ› οΈ Configuration

/sdcard/.androidforclaw/openclaw.json

{
  "models": {
    "providers": {
      "openrouter": {
        "baseUrl": "https://openrouter.ai/api/v1",
        "apiKey": "sk-or-v1-your-key",
        "models": [{"id": "xiaomi/mimo-v2-pro", "reasoning": true, "contextWindow": 1048576}]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": { "primary": "openrouter/xiaomi/mimo-v2-pro" }
    }
  },
  "channels": {
    "feishu": { "enabled": true, "appId": "cli_xxx", "appSecret": "xxx" },
    "slack": {
      "enabled": true,
      "botToken": "xoxb-...",
      "appToken": "xapp-...",
      "mode": "socket",
      "streaming": "partial",
      "model": "openrouter/xiaomi/mimo-v2-pro"
    },
    "telegram": {
      "enabled": true,
      "botToken": "123456:ABC-...",
      "streaming": "partial"
    }
  }
}

Each channel supports per-channel model override β€” pick a specific model from your configured providers, or leave empty to use the global default.

See πŸ“– Feishu Docs for detailed configuration reference.

πŸ”¨ Build from Source

git clone https://github.com/SelectXn00b/AndroidForClaw.git
cd AndroidForClaw
export JAVA_HOME=/path/to/jdk17
./gradlew assembleRelease
adb install releases/AndroidForClaw-v1.3.0-release.apk

πŸ”— Related Projects

Project Description
OpenClaw AI Agent framework (Desktop)
iOSForClaw OpenClaw iOS client
AndroidForClaw OpenClaw Android client (this project)

πŸ“ž Community

Feishu Group

Join Feishu Group

Click to join Feishu Group

Discord

Discord

Join Discord

WeChat Group

WeChat Group QR Code

Scan to join WeChat group β€” Valid for 7 days

πŸ”— Links

πŸ“„ License

MIT β€” LICENSE

πŸ™ Acknowledgments

  • OpenClaw β€” Architecture inspiration
  • Claude β€” AI reasoning capabilities

⭐ If this project helps you, please give it a Star! ⭐