nanoMuse, an open-source AI agent, aims to run one assistant across your phone, computer and browser

nanoMuse is a new open-source personal agent that is meant to run across all the devices you own. The details here come from the project's own page, and I have not tested the software.

The idea is one agent with a name and a look of its own. It is modelled on Meta's Muse. The project says it does things instead of just answering questions, keeps working while the app is closed, remembers you, and stops to ask before anything you could not undo. By default the agent appears as a small dragon, and users can describe a different look and have an image model draw it.

The code is free and released under GPL-3.0-or-later, and the project describes itself as non-profit. It says it is built on OpenMinis, another open-source project. The repository holds the phone app, the desktop app, the web console and the relay, which is the server that joins them. The word "nano" refers to this whole set being small enough for a person to run and deploy on their own.

On what it can do, the project lists a Linux shell, a browser, MCP servers (a standard way to plug tools into AI agents) and skills. With a setting called Hands on, it can also work the apps on your phone and the windows on your computer through their screens. This is meant for apps that have no API, which is the official way for software to talk to other software. The project says it tries APIs first and hands logins and CAPTCHAs over to the user. This screen control does not work on iOS, because the system does not allow it.

The agent asks before deleting, sending or paying. You can choose to approve once, for the current chat, or always. Passwords and codes are left for you to type.

The devices can also work together. According to the project, you can give a task on your phone and have it run on your PC. A message that starts with @Mac sends the task to that machine, and approval requests come back to the device in your hand. The text of conversations travels through the relay, while files and screenshots stay on the device where they were made.

The agent can also run goals on a schedule, run routines while the app is closed, and write a feed for you each morning. What it knows about you, and when it wakes up, are stored as Markdown files that you can read and edit. It can answer in the chat apps Feishu, DingTalk, WeCom and Telegram.

On models, users can pick the relay's free allowance, their own key from one of eighteen providers, or a plan they already pay for, such as ChatGPT, Claude or Kimi. The project says the developer pays for the free allowance on the community relay. When it runs out, users bring their own key. The same relay can run on a user's own server, so nothing has to leave their home. If a provider has no image or video models, those features are switched off and the app says so.

Installing depends on the device. There is an Android APK (version 0.1.40, Android 8.0 or newer, arm64) and a Windows installer. There are also macOS builds for Apple Silicon and Intel, plus Linux packages. The iPhone and iPad version is a TestFlight beta that depends on Apple's beta review. The macOS build is not notarised, so users need to right-click and choose Open the first time. A browser demo runs a nanoMuse on a simulated phone, and the project says the apps are the real thing. Docker scripts are provided for self-hosting.

The project compares itself with Meta Muse and OpenMinis. It says Muse runs in a cloud VM per user and is closed, while OpenMinis runs only on the phone it is installed on. nanoMuse says it can run on your phone, your computer or your own server under one account. These comparisons are the project's own claims.

The source material was limited, so names, dates and figures should be checked on the project's pages before relying on them.