nanoMuse
Open-source personal agent that runs across Android, iOS, desktop and web, performing tasks with a shell, browser, MCP servers and screen control.
At a Glance
Free, open-source, non-profit personal agent (phone app, desktop app, web console and relay) under GPL-3.0-or-later; self-hostable. Sign in to the community relay for a free model allowance, or use your own key.
Engagement
Available On
Alternatives
Listed Oct 2026
About nanoMuse
nanoMuse is an open-source personal agent for every device you own, developed by the nano-muse organization as a non-profit community project under GPL-3.0-or-later. One agent with its own name and look runs on Android, iPhone and iPad, macOS, Windows, Linux and the web, sharing one account and the same conversations. The latest listed release is 0.1.41 "Choice", dated 2026-10-07.
What It Is
nanoMuse is a cross-device personal agent that does things rather than only answering questions. It has a Linux shell, a browser, MCP servers and skills. With "Hands" turned on, it can operate phone apps and computer windows through their screens, which covers software without an API. Its identity, what it knows about the user and when it wakes are stored as Markdown files the user can read and edit.
How the Multi-Device Setup Works
Each device runs its own agent: the phone app runs Alpine Linux under proot inside the APK, while the desktop app is built as a plugin of DeepSeek Harness with a Python runtime for the hands. When signed in, devices meet on a relay and can ask each other for things. A task started on the phone can run on a PC, and approvals return to the device in the user's hand. Conversation text travels through the relay, while files and screenshots stay where they were made.
Safety and Approvals
Every tool call goes through a component called the Sentinel, which stops to ask before deleting, sending or paying. Approvals can be remembered for once, for the chat, or always. Passwords and codes are typed by the user, and logins or CAPTCHAs are handed over to the user, who resumes the agent with Done.
Models, Self-Hosting and Chat Apps
The project documents a free allowance of model use on a community relay, your own key with one of eighteen providers (including Bailian, OpenRouter, OpenAI, Gemini and DeepSeek), or existing ChatGPT, Claude or Kimi plans. The same relay can run on a server you control, via a script or Docker. The agent can also answer in Feishu, DingTalk, WeCom and Telegram, run scheduled goals and routines while the app is closed, and generate a custom look with image and video models.
Current Status
The iPhone and iPad build is distributed through TestFlight as a beta, with the public link delivering the build once Apple's beta review passes. The browser demo runs on a simulated phone using MobileGym re-creations of apps. The phone app is based on OpenMinis 1.13. nanoMuse is not affiliated with Meta Platforms.
Community Discussions
Be the first to start a conversation about nanoMuse
Share your experience with nanoMuse, ask questions, or help others learn from your insights.
Pricing
Open Source
Free, open-source, non-profit personal agent (phone app, desktop app, web console and relay) under GPL-3.0-or-later; self-hostable. Sign in to the community relay for a free model allowance, or use your own key.
- GPL-3.0-or-later licensed phone app, desktop app, web console and relay in one repository
- Free allowance of model use on the community relay (paid by the developer); afterwards use your own key
- Runtime needs no account when using your own key
- Same relay can run on your own server
- Model provider costs with your own key are external
Capabilities
Key Features
- Linux shell, browser, MCP servers and skills
- Screen-based control of phone apps and desktop windows (Hands)
- Approval prompts before deleting, sending or paying
- Cross-device tasks over a shared relay and account
- Scheduled goals and routines that run while the app is closed
- Editable Markdown files for identity, memory and wake schedule
- Replies in Feishu, DingTalk, WeCom and Telegram
- Custom agent look generated with image and video models
- Bring your own model key from eighteen providers
- Self-hostable relay and web console via Docker
- Browser demo on a simulated phone
