Mangi-11/Eta

What it solves

Eta addresses the limitations of built-in mobile AI assistants by providing a system-level AI Agent for Android that allows users to bring their own models (BYOK). It bridges the gap between simple chat interfaces and actual device control, enabling an AI to perform cross-app tasks, access local system data, and execute technical commands that are typically restricted in mobile sandboxes.

How it works

Eta uses an internal Agent Runtime that manages an "Agent Loop" to coordinate model calls, tool execution, and feedback. It employs several mechanisms to interact with the device:

• System Integration: Directly calls Android APIs and Intents for settings, media, and alarms. It can also hook into system entry points (like the power button or manufacturer assistants) via Xposed/LSPosed.
• GUI Agent: Uses accessibility UI trees and screenshots to perform clicks, scrolls, and text input in apps that lack APIs.
• Technical Environment: Includes a built-in browser (WebView) and a full Linux terminal (Alpine/Debian via PRoot or chroot) for running scripts, managing files, and executing code.
• Context & Memory: Retrieves local notifications, app usage, and location, and maintains a long-term memory file (MEMORY.md) for cross-conversation context.
• Extensibility: Supports "Skills" (task methods/scripts) and the Model Context Protocol (MCP) for remote tool integration.

Who it’s for

Android power users and developers who want a more capable, model-agnostic AI assistant that can actually control their device, run Linux commands, and integrate with their preferred LLM providers.

Highlights

• BYOK (Bring Your Own Key): Supports a wide range of providers including OpenAI, Anthropic, DeepSeek, and Kimi.
• System-Level Control: Ability to intercept power button presses and replace manufacturer assistants (like Breeno or Super Xiaoai).
• Integrated Linux Environment: Full terminal support with Python and Node.js installation options for "Vibe Coding" on mobile.
• Hybrid Execution: Combines direct API calls, GUI automation, and shell scripting to complete complex tasks.
• Personal Context: Accesses local data like notes, recordings, and health summaries (some features require Root).