██╗ ███████╗██╗ ██╗███╗ ███╗ █████╗ ██╗ ██╗██╗███████╗██████╗ ██╗ ██╗
██║ ██╔════╝╚██╗ ██╔╝████╗ ████║██╔══██╗ ██║ ██║██║██╔════╝██╔══██╗╚██╗ ██╔╝
██║ █████╗ ╚████╔╝ ██╔████╔██║███████║ ██║ ██║██║█████╗ ██████╔╝ ╚████╔╝
██║ ██╔══╝ ╚██╔╝ ██║╚██╔╝██║██╔══██║ ╚██╗ ██╔╝██║██╔══╝ ██╔══██╗ ╚██╔╝
███████╗███████╗ ██║ ██║ ╚═╝ ██║██║ ██║ ╚████╔╝ ██║███████╗██║ ██║ ██║
╚══════╝╚══════╝ ╚═╝ ╚═╝ ╚═╝╚═╝ ╚═╝ ╚═══╝ ╚═╝╚══════╝╚═╝ ╚═╝ ╚═╝
A fully autonomous AI agent that lives on your desktop, thinks in real-time, and controls your OS.
Features • Architecture • Installation • Hybrid LLM Setup • Development • Contributing • License
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 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.
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.
| 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 |
"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
- 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
┌─────────────────────┐ ┌─────────────────────┐
│ 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
| 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 |
- 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
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
- Node.js ≥ 18.0.0
- npm ≥ 9.0.0 (or pnpm/yarn)
- Git
# Clone the repository
git clone https://github.com/ManusAgents/ManusGPT.git
cd ManusGPT
# Install dependencies
npm install
# Start the application
npm run dev🪟 Windows
# Requires Windows 10+ (build 19041+)
# Ensure you have Build Tools installed:
npm install -g windows-build-tools
# Run the app
npm run devFor 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 devFor 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 devManusGPT supports both cloud-hosted and locally-running LLMs simultaneously. Configure multiple providers and switch between them on the fly.
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.
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 defaultIn ManusGPT Settings, select Ollama as the provider and enter:
- Base URL:
http://localhost:11434 - Model:
llama3(or any model you've pulled)
🖥️ LM Studio
- Download LM Studio
- Search and download any GGUF model
- Start the local server (default:
http://localhost:1234) - 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-cppIn ManusGPT Settings, select node-llama-cpp and point it to a GGUF model file.
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
}
}
}git clone https://github.com/ManusAgents/ManusGPT.git
cd ManusGPT
npm install
npm run dev| 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 |
| 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) |
🖼️ 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.
We love contributors! ManusGPT is built by a passionate community of AI enthusiasts and engineers.
Quick Start:
- 🍴 Fork the repository
- 📥 Clone your fork
- 🔧 Run
npm install && npm run dev - 🎉 Build something awesome
- 📤 Submit a Pull Request
See CONTRIBUTING.md for the full guide, code standards, and PR submission process.
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."