Skip to content

ManusAgents/ManusGPT

██╗     ███████╗██╗   ██╗███╗   ███╗ █████╗     ██╗   ██╗██╗███████╗██████╗ ██╗   ██╗
██║     ██╔════╝╚██╗ ██╔╝████╗ ████║██╔══██╗    ██║   ██║██║██╔════╝██╔══██╗╚██╗ ██╔╝
██║     █████╗   ╚████╔╝ ██╔████╔██║███████║    ██║   ██║██║█████╗  ██████╔╝ ╚████╔╝
██║     ██╔══╝    ╚██╔╝  ██║╚██╔╝██║██╔══██║    ╚██╗ ██╔╝██║██╔══╝  ██╔══██╗  ╚██╔╝
███████╗███████╗   ██║   ██║ ╚═╝ ██║██║  ██║     ╚████╔╝ ██║███████╗██║  ██║   ██║
╚══════╝╚══════╝   ╚═╝   ╚═╝     ╚═╝╚═╝  ╚═╝      ╚═══╝  ╚═╝╚══════╝╚═╝  ╚═╝   ╚═╝

God-Tier Desktop AGI Assistant

A fully autonomous AI agent that lives on your desktop, thinks in real-time, and controls your OS.

TypeScript Electron React License Version PRs Welcome

FeaturesArchitectureInstallationHybrid LLM SetupDevelopmentContributingLicense


✨ Features

🧠 Dual-Mode AGI Engine

ManusGPT isn't just a chatbot — it's a living, breathing autonomous agent with two distinct modes:

Mode Description
🎮 Command Mode Event-driven, on-demand execution. Ask a question, get an answer. Execute a task, get results. You're in the driver's seat.
🤖 AGI Autonomous Mode Always-on background agent that plans, executes, observes, reflects, and retries — all on its own. Set it loose on a goal and watch the magic happen.

🔄 PAORR Loop — The Thinking Engine

   ┌──────────────────────────────────────────────────┐
   │              PAORR AUTONOMOUS LOOP               │
   │                                                  │
   │   📋 PLAN  ──►  ⚡ ACT  ──►  👁️ OBSERVE          │
   │      ▲                                │           │
   │      │         ◄──  🔄 RETRY  ◄─────┘           │
   │      │               │                           │
   │      └── 🪞 REFLECT ◄─┘                          │
   │                                                  │
   └──────────────────────────────────────────────────┘

Every autonomous task flows through Plan → Act → Observe → Reflect → Retry. The entire loop is visualized in real-time in the ThoughtProcess panel. Watch the agent think, fail, learn, and adapt.

🕸️ DAG Multi-Agent Orchestration

Complex tasks are decomposed into a Directed Acyclic Graph of specialized agents:

  📊 Product Manager  ──►  🏗️ Architect  ──►  💻 Engineer  ──►  ✅ QA
       (planning)            (design)           (code)          (verify)

Each agent has its own PAORR loop, toolset, and context window. They communicate through the orchestrator to build, test, and ship — autonomously.

🛠️ 14 Built-In Tools

Tool Capability
bash Execute shell commands on your OS
python Run Python scripts with full stdlib access
browser Headless Chromium automation via Playwright
web_search Real-time web search across engines
crawl Scrape and extract content from any URL
editor Read, write, and edit files on disk
image_gen Generate images from text descriptions
memory Persistent dual-layer memory (short + long term)
skill_manager Load, create, and manage dynamic skills
cross_session_search Search across all past sessions
delegate Spawn sub-agents for parallel execution
ask_human Pause and ask the user for input
terminate Gracefully end a task or session
data_viz Generate charts, graphs, and visualizations

🎭 God Mode — Real OS Control

"The cursor moves. Keys are pressed. Windows open and close. The machine obeys."

God Mode grants ManusGPT actual mouse and keyboard control over your operating system. It can:

  • Click anywhere on screen with pixel precision
  • Type text into any application
  • Read screen content via OCR / accessibility APIs
  • Navigate file explorers, browsers, and native apps
  • Automate workflows across multiple applications

⚠️ Protected by 3-Layer Permission Gate — nothing executes without explicit user consent.

🗣️ Voice Interface

  • STT (Speech-to-Text) — Talk to ManusGPT using your microphone
  • TTS (Text-to-Speech) — Hear ManusGPT respond in natural speech
  • Wake word support for hands-free activation

🧠 FTS5 Dual-Layer Memory

  ┌─────────────────────┐     ┌─────────────────────┐
  │   SHORT-TERM        │     │   LONG-TERM         │
  │   Sliding Window    │────►│   BM25 Full-Text    │
  │   (recent context)  │     │   (all sessions)    │
  └─────────────────────┘     └─────────────────────┘
           SQLite FTS5 Engine — Instant semantic search across everything you've ever discussed

🛡️ Security Stack

Layer Protection
Permission Gate 3-tier system: ALLOW / ASK / DENY per tool and action
Catastrophic Mode Rejection Automatically blocks destructive commands (rm -rf /, format, etc.)
Identity Guard 30+ injection/jailbreak detection patterns protect the agent's core personality
Sandboxed Execution Tool outputs are validated and sanitized before rendering

🎨 Glassmorphism Premium GUI

  • Frosted glass panels with backdrop blur
  • Real-time animated PAORR loop visualization
  • Dual-pane layout: terminal logs + live Chromium preview
  • System tray integration (minimize to tray, background operation)
  • Drag-and-drop file processing for ZIP, PDF, CSV, and images

🏗️ Architecture

manusgpt/
├── electron/                     # Main process (Node.js)
│   ├── main.ts                   # Electron entry point
│   ├── preload.ts                # IPC bridge (secure context)
│   ├── engine/                   # 🧠 AGI Core Engine
│   │   ├── agent/                # Multi-agent system
│   │   │   ├── manus.ts          # 🏛️ Manus — the orchestrator
│   │   │   ├── orchestrator.ts   # DAG pipeline controller
│   │   │   ├── react-agent.ts    # ReAct reasoning agent
│   │   │   ├── toolcall-agent.ts # Direct tool-call agent
│   │   │   └── roles/            # PM, Architect, Engineer, QA
│   │   ├── modes/                # Dual-mode engine
│   │   │   ├── command-mode.ts   # 🎮 Event-driven mode
│   │   │   └── agi-mode.ts       # 🤖 Autonomous mode
│   │   ├── flow/                 # PAORR loop implementation
│   │   ├── tool/                 # 14 tool implementations
│   │   ├── llm/                  # Hybrid LLM router
│   │   ├── memory/               # FTS5 dual-layer memory
│   │   ├── permissions/          # 3-layer permission gate
│   │   ├── identity-guard.ts     # Injection detection
│   │   ├── skills/               # Dynamic skill engine
│   │   ├── voice/                # STT/TTS engine
│   │   ├── system-control/       # God Mode (mouse/keyboard)
│   │   └── file-processing/      # Omni file parser
│   ├── database/                 # SQLite session storage
│   ├── ipc/                      # IPC handler registry
│   └── system/                   # OS integration (tray, cron, watcher)
├── src/                          # Renderer process (React)
│   ├── components/               # UI components
│   │   ├── Chat/                 # Chat panel + input
│   │   ├── Dashboard/            # Main dashboard + sidebar
│   │   ├── ThoughtProcess/       # PAORR visualization
│   │   ├── DAGVisualization/     # Agent pipeline graph
│   │   ├── DualPane/             # Split-screen layout
│   │   ├── Terminal/             # Log terminal
│   │   ├── GodMode/              # God Mode controls
│   │   ├── VoiceControl/         # Voice input/output
│   │   ├── MemoryVault/          # Memory browser
│   │   ├── Settings/             # LLM + config panel
│   │   ├── FileDrop/             # Drag-and-drop zone
│   │   └── Canvas/               # Freeform canvas
│   ├── store/                    # Zustand state management
│   ├── hooks/                    # Custom React hooks
│   └── styles/                   # Global styles + glassmorphism
├── package.json
├── vite.config.ts
├── tsconfig.json
└── tailwind.config.js

📦 Installation

Prerequisites

  • Node.js ≥ 18.0.0
  • npm ≥ 9.0.0 (or pnpm/yarn)
  • Git

Quick Install

# Clone the repository
git clone https://github.com/ManusAgents/ManusGPT.git
cd ManusGPT

# Install dependencies
npm install

# Start the application
npm run dev

Platform-Specific Notes

🪟 Windows
# Requires Windows 10+ (build 19041+)
# Ensure you have Build Tools installed:
npm install -g windows-build-tools

# Run the app
npm run dev

For packaged builds, download the latest .exe installer from Releases.

🍎 macOS
# Requires macOS 11+ (Big Sur or later)
# No additional dependencies needed
npm install
npm run dev

For packaged builds, download the latest .dmg from Releases.

🐧 Linux
# Debian/Ubuntu — install required system libraries:
sudo apt-get install -y libgtk-3-0 libnotify4 libnss3 libxss1 \
  libasound2 libatspi2.0-0 libsecret-1-0

# Fedora:
sudo dnf install -y gtk3 libnotify nss libXScrnSaver alsa-lib \
  at-spi2-atk libsecret

# Arch:
sudo pacman -S gtk3 libnotify nss libxss alsa-lib at-spi2-atk libsecret

npm install
npm run dev

🧠 Hybrid LLM Setup

ManusGPT supports both cloud-hosted and locally-running LLMs simultaneously. Configure multiple providers and switch between them on the fly.

☁️ Online Providers

Supported providers and their configuration:

{
  "llm": {
    "online": {
      "openai": {
        "apiKey": "sk-...",
        "model": "gpt-4o",
        "baseURL": "https://api.openai.com/v1"
      },
      "anthropic": {
        "apiKey": "sk-ant-...",
        "model": "claude-sonnet-4-20250514"
      },
      "google": {
        "apiKey": "...",
        "model": "gemini-2.5-pro"
      },
      "mistral": {
        "apiKey": "...",
        "model": "mistral-large-latest"
      }
    }
  }
}

All providers can be configured through the Settings Panel in the GUI, or by editing the config file directly.

🏠 Local / Offline Providers

Run powerful models directly on your machine — zero API costs, full privacy.

📦 Ollama
# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull a model
ollama pull llama3
ollama pull codellama

# Ollama runs on http://localhost:11434 by default

In ManusGPT Settings, select Ollama as the provider and enter:

  • Base URL: http://localhost:11434
  • Model: llama3 (or any model you've pulled)
🖥️ LM Studio
  1. Download LM Studio
  2. Search and download any GGUF model
  3. Start the local server (default: http://localhost:1234)
  4. In ManusGPT Settings, select LM Studio as the provider
⚡ node-llama-cpp

For the lightest possible setup — run models directly in-process via Node.js bindings.

# Installed automatically as an optional dependency
# No external server needed
npm install node-llama-cpp

In ManusGPT Settings, select node-llama-cpp and point it to a GGUF model file.

🔀 Hybrid Routing

Configure a router strategy to intelligently distribute requests:

{
  "llm": {
    "router": {
      "strategy": "smart",        // "smart" | "round-robin" | "priority" | "fallback"
      "primary": "openai",        // First choice
      "fallback": "ollama",       // Used if primary fails
      "localFor": ["code", "file"], // Route these tasks to local model
      "cloudFor": ["web", "complex"] // Route these to cloud
    }
  }
}

🚀 Development

Setup

git clone https://github.com/ManusAgents/ManusGPT.git
cd ManusGPT
npm install
npm run dev

Available Scripts

Command Description
npm run dev Start Electron app in development mode with HMR
npm run build Build the production bundle
npm run lint Run ESLint on all source files
npm run typecheck Run TypeScript type checking
npm run test Run the test suite
npm run package Build platform-specific installers

Tech Stack

Layer Technology
Desktop Shell Electron 31
UI Framework React 18 + TypeScript 5.4
State Management Zustand
Styling Tailwind CSS + CSS Glassmorphism
Build Tool Vite
Database SQLite (better-sqlite3 + FTS5)
Browser Automation Playwright
LLM Integration OpenAI SDK, Anthropic SDK, Ollama API
Voice Web Speech API (STT/TTS)

📸 Screenshots

🖼️ Dual-Pane Interface — Terminal logs on the left, live Chromium preview on the right.

🖼️ PAORR Loop Visualization — Watch the agent plan, act, observe, reflect, and retry in real-time.

🖼️ DAG Multi-Agent Pipeline — See the PM → Architect → Engineer → QA flow in action.

🖼️ God Mode — Real OS control with mouse/keyboard automation visualized.

🖼️ Settings Panel — Configure hybrid LLM providers, permissions, and voice settings.


🤝 Contributing

We love contributors! ManusGPT is built by a passionate community of AI enthusiasts and engineers.

Quick Start:

  1. 🍴 Fork the repository
  2. 📥 Clone your fork
  3. 🔧 Run npm install && npm run dev
  4. 🎉 Build something awesome
  5. 📤 Submit a Pull Request

See CONTRIBUTING.md for the full guide, code standards, and PR submission process.


⚖️ License

This project is licensed under the Apache License 2.0 — see the LICENSE file for full details.

Copyright 2024 ManusAgents

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

Built with 🧠 by ManusAgents

"The future of AI is not a chat window. It's an agent that thinks, acts, and builds."

About

The Ultimate God-Tier Desktop AGI Assistant. Dual-mode, Omni-file processing, Live Preview, Local LLM support, and Deep OS Control.

Topics

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages