Skip to content

v1.0.0 - The AGI Awakening

Pre-release
Pre-release

Choose a tag to compare

View all tags
@ManusClawAI ManusClawAI released this 10 Jun 07:40
v1.0.0-beta
61f8ab9

🌋 v1.0.0-beta — The AGI Awakening

The God-Tier Desktop AGI has been unleashed upon the world.

🔥 What's Inside

🧠 Dual-Mode AGI Engine

  • Command Mode — Event-driven, precise instruction execution with instant responses
  • AGI Autonomous Mode — Always-on background intelligence that monitors, plans, and acts independently
  • PAORR Loop — Plan → Act → Observe → Reflect → Retry — the cognitive engine powering every thought
  • DAG Orchestration — Multi-agent pipelines with topological sorting (Product Manager → Architect → Engineer → QA)

🛠️ 14-Power Tool Arsenal

Tool Capability
Bash Shell Persistent command-line execution
Python Engine Sandboxed Python subprocess (2GB limit)
Browser Playwright web automation
WebSearch DuckDuckGo + Bing fallback
Crawl URL content extraction
Editor File CRUD with str_replace
ImageGen DALL-E / FAL.ai image creation
Memory FTS5 dual-layer memory system
SkillManager YAML-based skill engine
CrossSessionSearch Cross-session knowledge retrieval
Delegate Sub-agent spawning
AskHuman IPC human-in-the-loop bridge
DataViz Real-time data visualization
Terminate Graceful session shutdown

🖥️ Premium Glassmorphism GUI

  • Frameless Electron window with custom titlebar
  • PAORR real-time cycle visualization
  • DAG pipeline panel (ReactFlow)
  • Interactive terminal emulator
  • FTS5 Memory Vault browser
  • System tray with status-aware icons (idle/running/error/AGI)

🗣️ STT/TTS Voice System

  • Speech-to-Text: Real-time voice input for hands-free control
  • Text-to-Speech: Verbal AI explanations and responses
  • Wake word support for instant activation

📂 Omni File Processing Engine

  • Drag-and-drop file ingestion
  • ZIP extraction and analysis
  • PDF parsing with full content extraction
  • CSV/Excel data analysis
  • Image OCR and metadata extraction

🔲 Dual-Pane Live Preview

  • Split-screen layout: Logs on left, Chromium preview on right
  • Swapable panes for maximum productivity
  • Real-time HTML/CSS/JS preview rendering

🕹️ God Mode — Real OS Control

  • Mouse Controller: Click, drag, scroll, move automation
  • Keyboard Controller: Type, hotkey, key-combo automation
  • Screen Controller: Screenshot, pixel analysis, window management
  • Full system-level control at the AI's fingertips

🤖 Hybrid LLM Support

  • Online: OpenAI, Anthropic Claude, Google Gemini, Mistral
  • Offline: Ollama, LM Studio, node-llama-cpp
  • Smart Routing: Automatic online/offline failover
  • Credential Pool: Multi-key rotation with cooldown
  • Token Budget: Per-session spend tracking

🛡️ Security Stack

  • 3-Layer Permission Gate (ALLOW / ASK / DENY)
  • Catastrophic mode rejection (rm -rf, format, etc.)
  • Identity Guard: 30+ injection/jailbreak detection patterns
  • SHA-256 approval caching for trusted operations
  • FTS5 SQLite encrypted session storage

📦 Installation

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

# Install dependencies
npm install

# Configure your LLM (see .env.example)
cp .env.example .env

# Run in development
npm run dev

# Build for production
npm run build

Hybrid LLM Setup

Online (OpenAI):

OPENAI_API_KEY=sk-...
LLM_PROVIDER=openai

Offline (Ollama):

curl -fsSL https://ollama.ai/install.sh | sh
ollama pull llama3.1
LLM_PROVIDER=ollama
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=llama3.1

Offline (LM Studio):

LLM_PROVIDER=openai-compatible
OPENAI_BASE_URL=http://localhost:1234/v1

⚠️ God Mode Disclaimer

ManusGPT possesses real OS-level control. Please read NDA.md and SECURITY.md before enabling God Mode. You assume full responsibility for all autonomous actions performed by the agent.

🏗️ Built With

  • Electron 31 — Desktop framework
  • React 18 + Vite 5 — Renderer
  • TypeScript 5.5 — Language
  • Zustand — State management
  • Tailwind CSS — Glassmorphism styling
  • Framer Motion — Animations
  • ReactFlow — DAG visualization
  • Playwright — Browser automation
  • better-sqlite3 + FTS5 — Memory & sessions
  • node-llama-cpp — Offline LLM inference

📄 Stats

  • 141 source files across 2 processes (Main + Renderer)
  • 39,000+ lines of production TypeScript
  • 14 built-in tools with permission gating
  • 6 major feature modules (Voice, God Mode, File Processing, Dual-Pane, Hybrid LLM, OS Integration)

Built with 🔥 by ManusAgents

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

Assets 4
Loading