AI NewsOpen sourceAnnouncement

NanoMuse publishes an open-source personal agent for phones, desktops and web

NanoMuse publishes a GPL-3.0 personal agent that runs on phone, desktop, and the web, with a relay you can self-host so no data has to leave your network.

AI News

Editorial2 min read

LinkedInX
The nanoMuse GitHub repository, an open-source personal AI agent for every device

Image: GitHub

Why it mattersA reader who wanted a Meta Muse style agent without a vendor cloud can download the signed Android build, the macOS and Windows apps, or run the whole relay on one VPS.

A personal AI agent that keeps working while the app is closed and remembers you between sessions usually comes with a cloud account you cannot inspect. NanoMuse is an open-source project that publishes the whole stack: the phone app, the desktop app, the web console and the relay that joins them, under GPL-3.0-or-later. The maintainers describe it as modelled on Meta Muse.

The Show HN at news.ycombinator.com/item?id=49987765 has 58 points. The repository has 328 stars. Version 0.1.41 is the current release and the maintainers posted a paper on arXiv at 2610.08699.

The point of the agent is that it does things rather than answering questions. A Linux shell, a browser, MCP servers and skills are built in, and with "Hands" turned on the agent can drive apps on the phone and windows on the computer through their screens, for anything that has no API. Before deleting, sending or paying, it stops and asks, with the option to remember the answer once, in a chat, or always. A sign-in or a CAPTCHA is handed back to the user, and a "Done" button resumes the task.

Devices share one account. Say something on the phone and it can run on the PC; @Mac at the start of a message routes the task there. Approvals come back to the device in the user's hand. Goals are checked on a schedule, routines keep running while the app is closed, and a feed is written each morning.

Builds exist for every current platform. Android 8.0+ arm64 ships as a signed APK that installs over the last build. iPhone and iPad are on TestFlight. macOS 12+ has Apple Silicon and Intel builds. Windows 10+ has an .exe. Linux ships as an AppImage, a .deb and a tar.gz. A browser demo runs on demo.nanomuse.dev. Docker Compose runs the web app and a self-host script sets up a relay of your own.

The model is swappable. The community relay has a free allowance paid by the developer; after that, users can plug in a key for one of eighteen providers (Bailian, OpenRouter, OpenAI, Gemini, DeepSeek and more) or sign in with a plan they already pay for at ChatGPT, Claude or Kimi. A provider without picture or video models leaves those two off, and the app says so.

A table in the README compares NanoMuse with Meta Muse and the open-source project OpenMinis. Muse runs in a cloud VM per user and never touches the devices. OpenMinis runs on the Android it is installed on. NanoMuse runs on the user's phone, computer, or a server they rent, with one account across all of them.

The Hacker News thread flagged two things worth checking before installing. One commenter said a similar tool had written config files for eleven other coding assistants and started a port 4400 daemon when its state directory was redirected, and recommended auditing the installer. Another asked how security boundaries are handled when the agent can also take actions on the phone it runs on. The project is early enough that a reader should start on a device where the Hands permission has a clear boundary from anything sensitive.

Source

nano-muse/nanoMuse on GitHub. Discussion on Hacker News. Paper on arXiv.

SourceGitHub

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX