# 4AIs

> 4AIs is an AI agent that lives on your Android phone. It taps, types, and runs apps on your behalf, the same way you would. It is not a chatbot. It is an agent that does work in the background and reaches out only when something needs you.

4AIs runs locally on the device using the OpenClaw gateway and free NVIDIA NIM models (Kimi K2, Llama 3.3 70B). It uses Android's AccessibilityService to control apps, and optionally Shizuku for advanced shell access. No root. No PC required. No cloud subscription.

## What 4AIs is

- An autonomous AI agent for Android, version targets SDK 34, supports Android 8.0+
- Runs entirely on-device. The model gateway, the agent loop, and tool execution all happen on the phone
- Controls your phone's UI through Android AccessibilityService: real-time UI tree reading, native gesture dispatch, tap by text, scroll to find element, wait for element
- Uses 985 third-party app integrations through Composio (Gmail, Slack, GitHub, Notion, Stripe, Salesforce, HubSpot, Discord, Figma, Asana, Trello, Linear, Zoom, LinkedIn, YouTube, and more)
- Asks for approval before doing anything destructive or irreversible (toggleable in Settings)
- Free. No subscription. Powered by NVIDIA NIM free-tier models

## What 4AIs is not

- Not a chatbot you open and talk to
- Not a cloud product. There is no server doing the thinking
- Not paid. There is no subscription
- Not rooted access. AccessibilityService level only, with optional Shizuku for ADB-level shell

## Key links

- [Landing page](https://4ais.in/)
- [GitHub — free & open source](https://github.com/8crsk/openclaw-android)
- [Instagram @try4ais](https://instagram.com/try4ais) — daily build updates, demos, behind-the-scenes

## Frequently asked

- **What is 4AIs?** An AI agent that lives on your Android phone. It taps, types, and runs apps on your behalf.
- **How is it different from ChatGPT or Gemini?** Those are chat assistants. 4AIs is an agent that actually controls your phone.
- **Is it free?** Yes. It runs on free NVIDIA NIM models.
- **Does it need a PC?** No. It is fully on-device.
- **How do I get it?** It is free and open source — download or build it at github.com/8crsk/openclaw-android. Follow @try4ais on Instagram for drops.

## Architecture (technical)

```
Compose UI → WsRpcClient → ws://127.0.0.1:3000/ → OpenClaw Gateway (on-device Node.js) → NVIDIA NIM
                                                              ↓
                                                  spawns MCP shim → ShizukuBridge → Android (uid 2000)
                                                              ↓
                                              AccessibilityService → UiRoutes → UI tree + gestures
```

- OpenClaw is WebSocket-RPC, not OpenAI REST. The client speaks JSON-RPC frames over a single long-lived WebSocket
- Chat uses the `agent` method. Streaming arrives as `event:"agent"` frames demuxed into TextDelta, ToolStart, ToolResult, Lifecycle, AgentError, and Done events
- Tool calls execute via curl to a local ShizukuBridge HTTP server on 127.0.0.1:3001
- The gateway runs as a Termux-style Node.js binary (`libnode.so`) placed in the app's nativeLibraryDir, executed directly via ProcessBuilder