OpenCyvis transforms Android phones into open-source AI assistants, enabling natural language control of apps via a transparent, user-controlled interface. It addresses the black-box nature of commercial AI phones by offering model choice, data privacy, and full auditability, benefiting users who prioritize control and transparency.

How It Works

The core innovation is a virtual display where AI agents operate apps in the background, unseen by the user's primary screen. This allows complex, multi-app tasks (e.g., booking, messaging, scheduling) to execute without interrupting the user's current activity, providing a seamless and transparent AI interaction.

Quick Start & Requirements

• Standard Mode: Install opencyvis-standard-release.apk on Android 11+. Requires ADB wireless pairing (completed on-device) and LLM backend configuration (local via Ollama or cloud API). No root or computer needed.
• System App Mode: For developers; requires flashing a custom AOSP system image for elevated privileges.
• Docs: Roadmap, Contributing.

Highlighted Details

• Remote Control: Manage phone AI via IM (Feishu, Telegram), including commands, progress updates, and screenshots.
• Routines & Geofencing: Automate tasks based on schedules or location (e.g., daily summaries, clock-in).
• Provider Profiles: Easily switch between different LLM configurations and API keys.
• Model Agnostic: Supports numerous cloud (Qwen, Claude, GPT-4o) and local (Gemma, Qwen via Ollama) models, recommending Gemma 4 26B-A4B for local use.
• Multi-ROM Support: Standard mode accommodates various Android vendor ROMs (MIUI, ColorOS).

Maintenance & Community

Active development is indicated by a roadmap and contribution guidelines. Community interaction primarily occurs through GitHub issues and contributions.

Licensing & Compatibility

Released under the permissive Apache 2.0 license, allowing for commercial use and integration into closed-source projects without copyleft obligations.

Limitations & Caveats

System App mode requires advanced AOSP knowledge. Standard mode relies on ADB wireless debugging setup. Performance is dependent on the chosen LLM backend and device hardware.