kevinzhow/Rungic

AgentOS in your hand - An Android phone, a full Linux desktop computer, and an assistant that does the work for you.

What it solves

Rungic transforms a standard Android phone into a full-featured Linux desktop environment while integrating an AI agent capable of performing tasks directly within that desktop. It solves the problem of needing a separate computer for complex tasks by allowing your phone to run desktop-class software (like Blender, Krita, or VS Code) and having an AI assistant operate those applications for you.

How it works

Rungic runs a full Ubuntu Linux container alongside the Android operating system. It uses an Android app to bridge the Linux desktop's display, touch input, and hardware access to the phone's screen. The AI assistant operates within this Linux environment, using screen-reading and input-simulation tools to interact with desktop applications. You can interact with the assistant via voice or text, and it can manage multiple "workspaces" where it performs tasks in the background while you continue using your phone.

Who it’s for

It is for users who want to turn their mobile hardware into a productive desktop workstation and for those interested in "AgentOS" concepts—where AI agents act as persistent, autonomous participants in a computer workspace rather than just simple chatbots.

Highlights

  • Runs a full Linux desktop (KDE Plasma) in a container with native GPU acceleration.
  • Features an AI assistant that can open apps, click buttons, and type to complete complex tasks like rendering 3D models or editing images.
  • Supports multi-agent teams that can collaborate on projects by sharing workspaces and project folders.
  • Provides proactive system care by detecting Linux crashes, storage issues, and software compatibility problems.
  • Maintains user control by requiring explicit confirmation for sensitive actions and keeping administrator passwords secure.

Related

  • Project
    delorenj/mcp-server-trello

    A Bun‑powered MCP server that wraps the Trello API, exposing 57 typed tools (e.g., get_card, get_acceptance_criteria, watch_card) for AI agents. Handles auth, rate‑limiting, dynamic board/workspace switching, and returns data as JSON or markdown. Install via npm and configure with Trello API key/token; then add to any MCP‑compatible client (Claude Desktop, Cursor, etc.).

  • Project
    Sayhi-bzb/CharDesk

    CharDesk is a shared visual medium built on Unicode text that lets humans and LLM-powered agents create, inspect, and edit diagrams and interfaces together on an infinite canvas. It solves the mismatch between how people read two-dimensional layouts and how language models edit token sequences, keeping everything selectable, searchable, and diffable.

  • Project
    SheberDavid/v4-flash-godmode-opencode-go

    An agent preset for DeepSeek V4 Flash via opencode-go that uses optimized guidance anchors to prevent shallow reasoning and improve output quality for complex tasks.

  • Project
    NVlabs/SpatialClaw

    SpatialClaw is a training‑free framework where a vision‑language model writes and executes Python code step‑by‑step in a persistent Jupyter kernel to solve 3‑D/4‑D reasoning tasks. It ships with perception tools (SAM‑3, Depth‑Anything‑3), benchmark loaders for 20 spatial datasets, and a three‑service runtime (vLLM, GPU tool server, agent kernel). The same prompt works across many VLM backbones, achieving 59.9% average accuracy. The repo includes full installation scripts, SLURM launch managers, and detailed docs.