nutshellai-tech/mobius 预览 preview

nutshellai-tech/mobius

插件Plugin ⭐ 286 NOASSERTION cat_label.dsh-pluginscat_label.dsh-plugins

The first self-evolving open-source Agent OS, connecting your team, AI agents, devices, and compute

项目介绍Project Overview

该插件为 Mobius,已迁移至 mobius-system/mobius,原仓库不再维护。其核心能力是提供原仓库的镜像与迁移指引,便于用户更新远程地址。适用于仍引用旧仓库的项目接入。建议尽快切换至新仓库以获取后续更新与支持。

This plugin is Mobius, which has migrated to mobius-system/mobius, and the original repository is no longer maintained. Its core capability is providing a mirror and migration guide for the original repo, helping users update their remote URL. Use it when integrating projects still referencing the old repository. Note: switch to the new repository promptly to receive future updates and support.

或使用命令行安装(适合开发者)Or use CLI install (for developers)

命令行安装CLI Install

dsh plugin --profile web add github:nutshellai-tech/mobius

nutshellai-tech/mobius 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

English · 简体中文

Mobius logo Mobius

The first self-evolving open-source Agent OS
One system to connect your team, AI agents, devices, and compute

Stars Status Website Docs

Mobius GitHub cover

Building a perfect AI system once and for all is like finding the end of a Mobius strip — impossible by design.

Mobius is the world's first self-evolving open-source Agent OS. Not a fixed toolbox — a growing productivity system that connects projects, teams, models, devices, compute, and apps into one traceable workspace.

News

  • 2026-08-09Windows one-key install: a single PowerShell command installs the Mobius TUI on a fresh Windows machine.
  • 2026-08-02Easy Mode: an optional clutter-free layout (cross-project recent sessions + JSONL + floating input). First-time users choose between Easy and Normal mode and can switch anytime.
  • 2026-08-02TUI released: connect to Mobius from any terminal with the TUI, and use Mobius the way you use Codex.
  • 2026-07-26Search improvements: results stream in via SSE with case/whole-word matching; clicking a result jumps to the exact JSONL card.
  • 2026-07-14Code Conversation v2 workspace mode: a 3-pane layout (file browser + built-in CodeMirror editor with syntax highlighting and in-place saving + chat).

Self-Evolving

Mobius rewrites itself from your input. Send a change request, a screenshot, or a reference link — Mobius turns them into real code, UI, plugins, or workflow updates, without interrupting your work. Each iteration quietly replaces another plank on the Ship of Theseus, in the background.

Self-evolving Agent OS demo

View self-evolution examples

Serve Every Scenario and Project

  • Serve every scenario — switch to the most convenient interface for the task at hand

Interface layouts: regular, copilot, server co-op, and auto research

  • Manage all projects — one overview where you can watch every agent working for you

Project overview dashboard with agent status and task progress

Auto Research

Mobius orchestrates multiple agents into an autonomous research pipeline — reading papers, extracting methods, running experiments, and surfacing results. A research goal becomes a multi-agent system, not a single Q&A.

Auto Research demo

XiaoMo

XiaoMo is the natural-language interface to the entire system. Talk to it: create projects, split tasks, launch agents, track progress. Anything clickable, XiaoMo can do; things the UI cannot do, XiaoMo handles too. It supports voice input, multiple devices (Web, PC, mobile), and configurable reminders.

XiaoMo assistant interface

On the web. Zero install — open a browser on any device and the full Mobius workspace is ready instantly.

XiaoMo on mobile

On your phone. XiaoMo goes wherever you go — chat with your agents, track progress, and approve decisions from anywhere. The iOS and Android apps are now fully available.

XiaoMo desktop app

On your desktop. A native app that turns your PC into a Mobius workstation — read and write local project files directly, enroll this machine as a controllable node, and run multi-tab workflows. Available now on Windows, macOS, and Linux.

The demos on this page were produced by XiaoMo itself, recorded with zero human involvement.

Any Model, Any Agent

Mobius is model-agnostic. GPT, Claude, GLM-5.2, Codex — all serve as execution engines inside the same project. Choose by task type, cost, or performance.

Connect Everything

Mobius schedules browsers, terminals, GPU clusters, embedded boards, cloud servers, and workstations — all inside the same task network.

Reach your resources through SSH, AIMUX, and controllable proxies:

SSH and AIMUX access paths

Team Collaboration

Human members, AI agents, tasks, and deliverables in one view. Leads see who is doing what, where each agent is, what needs confirmation, and where risks exist — no more fragmented communication.

Team collaboration demo

Self-Incubating Extensions

Mobius ships with built-in extensions and grows new ones from your needs — financial dashboards, PPT generators, research workbenches, live portals. Each extension comes with a frontend, backend handler, data directory, and an invocation entry point, ready to keep evolving.

Extensions demo

Immersive Web Experiences
Turn visual ideas into runnable extension apps.
Matrix-style extension
Financial News Wall
Track live market narratives.
Financial news wall
World Cup Portal
Data-rich sports portal.
World Cup extension
PPT Maker
Presentations from topics and materials.
PPT maker extension

Quick Start

Full deployment guide at Docs.

Containers (recommended)

# 1. Clone the repo (tip: fork first, then clone — after self-evolution you can commit directly to your own repo)
git clone https://github.com/mobius-system/mobius.git && cd mobius

# 2. Generate config (random keys/passwords; you may configure manually to skip this)
python3 conf_prepare.py --docker && python3 conf_check.py --docker

# 3. Build the images (the base image is environment only, no code)
docker build -t mobius-system-base:latest -f deploy/Dockerfile .
docker build -t mobius-system-exe:latest .

# 4. Launch
docker compose up

Direct (Linux / macOS)

# 1. Install prerequisites (tmux, git, etc.)
sudo apt install tmux python3 git curl proxychains openssh-server build-essential

# 2. Install coding agents (either works; installing both is recommended)
npm install -g @anthropic-ai/claude-code @openai/codex

# 3. Clone the repo (tip: fork first, then clone — after self-evolution you can commit directly to your own repo)
git clone https://github.com/mobius-system/mobius.git && cd mobius

# 4. Generate and check config (copies .env.default to .env with random passwords)
python3 conf_prepare.py && python3 conf_check.py

# 5. Install dependencies (frontend + backend)
cd ./mobius && npm install && cd ./frontend && npm install && cd ../..

# 6. Run
python3 start.py

Roadmap

What we are building next:

  • Mobile App — XiaoMo and full Agent control on iOS and Android
  • Desktop App — a native connector that brings PC devices (Windows, macOS, Linux) into Mobius
  • Extension Market — discover, share, and install community-built extensions
  • i18n & Multi-language — localize the interface and documentation into more languages

Contribution

Issues, plugins, docs, bug reports, use cases — all welcome. If you believe AI systems should keep evolving instead of remaining static tools, join us.

GitHub · Website · Docs

上一个 Prev easyeda-agent 下一个 Next modsearch