nanoMuse Ships Open-Source Agent That Runs Across All Your Devices
nanoMuse is an open-source cross-device personal agent that runs on your phone and computer, with built-in approvals before any irreversible action.
- nanoMuse is an open-source GPL-3.0 personal agent that runs on phone, desktop, and web with one shared account.
- Devices coordinate through a relay so a phone can dispatch tasks to a Mac using
@Mac, with approvals returning to the user. - A Sentinel component intercepts every tool call and asks for confirmation before deleting, sending, or paying.
- GUI control works via screen hands on Android, Windows, macOS, and Linux, built on a port of ByteDance's UI-TARS-desktop.
- Supports 18 model providers including OpenAI, Gemini, DeepSeek, Claude, Kimi, plus a free community relay allowance.
- Current release is 1.0.1 Ballast, with a published arXiv paper from Zhejiang University researchers.
nanoMuse links one agent identity across phones, desktops, and the web
The GPL-3.0-licensed nanoMuse project combines device-specific agent runtimes under one account and shared conversation history. Inspired by Meta’s Muse personal agent, it adds self-hosting, model-provider choice, cross-device task routing, and approval prompts for consequential actions.
The repository publishes clients for Android, iPhone and iPad, Windows, macOS, Linux, and the browser. Its documentation describes an agent that can continue background tasks where operating-system policies permit, retain user memory, and request confirmation before deleting data, sending messages, or making payments.
One identity, several runtimes
Each signed-in device runs its own agent runtime while a relay coordinates identity, conversation history, task dispatch, and approvals. This design lets the system present one continuous agent even though execution happens on separate phones, computers, or servers.
On Android, nanoMuse packages Alpine Linux under proot, which provides a user-space Linux environment without requiring root access. The runtime includes a shell, browser access, and Model Context Protocol support. MCP gives agents a standard interface for calling external tools and services. The desktop client uses the DeepSeek Harness with a Python runtime for computer-control features.
Once devices join the same account, a request such as @Mac summarize the latest build logs can travel from a phone to a laptop. According to the project documentation, conversation text passes through the relay, while files and screenshots remain on the device that created them.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
