Skip to content

Home

Jump to bottom
beidald edited this page Feb 27, 2026 · 1 revision

Welcome to Lilium AI Wiki

Not just a chatbot, but your AI Co-pilot for total computer automation.

Lilium AI is a powerful personal AI agent framework designed for autonomous execution. It connects Large Language Models (LLMs) like OpenAI and Ollama to browser automation, system-level shell control, deep file integration, and omni-channel communication, enabling AI to perform complex end-to-end tasks just like a human.

🚀 Quick Navigation

🧠 Core Concepts

Lilium AI is built around several core components that define its "personality" and capabilities.

🐱 Soul

The Soul defines who Lilium is. It includes:

  • Personality: Helpful, friendly, concise, and curious.
  • Values: Accuracy over speed, user privacy, and transparency.
  • Communication Style: Clear, direct, and reasoning-oriented. Location: workspace/SOUL.md

🤖 Agents & Instructions

Instructions given to the AI to guide its behavior.

  • Guidelines: Explain actions, ask for clarification, use tools, and maintain memory.
  • Memory:
    • MEMORY.md: Long-term facts and context.
    • HISTORY.md: Event log for recalling past events. Location: workspace/AGENTS.md

🛠️ Tools

Lilium is equipped with a variety of tools to interact with the world:

  • File Operations: Read, write, edit, and list files.
  • Shell Execution: Run system commands (with safety restrictions).
  • Web Access: Search the web and fetch content from URLs.
  • Messaging: Communicate across different channels.
  • Background Tasks: Spawn sub-agents for asynchronous work. Location: workspace/TOOLS.md

💓 Heartbeat

A mechanism for periodic tasks. Lilium checks workspace/HEARTBEAT.md every 30 minutes to execute active tasks such as:

  • Checking calendars.
  • Scanning inboxes.
  • Monitoring weather or stock prices.

📥 Installation

Prerequisites

  • Node.js: v18 or higher.
  • pnpm: Recommended (npm install -g pnpm).
  • Python 3: (Optional) For macOS permission scripts.

Steps

  1. Clone: git clone https://github.com/beidald/liliumai.git
  2. Install: pnpm install
  3. Config: cp config.example.json config.json
  4. Run: pnpm start

⚙️ Configuration

Lilium is highly configurable via config.json. Key sections include:

LLM Providers

Supports OpenAI and Ollama (local).

"llm": {
  "provider": "openai",
  "apiKey": "your-key-here",
  "model": "gpt-4-turbo"
}

Knowledge Base (RAG)

Enable RAG using SQLite or LanceDB.

"knowledge": {
  "enabled": true,
  "provider": "sqlite",
  "embedding": {
    "provider": "ollama",
    "model": "nomic-embed-text"
  }
}

Security

Protect your system by restricting filesystem and shell access.

"security": {
  "restrict_fs_write": true,
  "restrict_shell_execution": true,
  "allowed_read_only_commands": ["ls", "cat", "grep"]
}

🔌 Omni-Channel

Lilium can live in your favorite messaging apps:

  • WeChat: Via Wechaty.
  • Telegram: Native bot support.
  • Discord: Native bot support.
  • Feishu / DingTalk: Webhook/App support.
  • WhatsApp: Bridge support.
  • Slack / Email: Integrated.

🚀 Advanced Features

🌐 Browser Automation

Lilium can navigate the web using Puppeteer. Just ask it to "Find the cheapest price for X" or "Summarize the latest news from Y".

🐣 Sub-Agents (Spawn)

For long-running tasks, use the spawn tool.

"Hey Lilium, spend the next hour researching X and write a summary in research.md."

🎙️ Voice Control

Built-in Whisper support (Local or Groq) allows you to control your agent via voice messages.

🗺️ Roadmap

  • Improved Multi-Agent Collaboration.
  • Visual UI for Task Management.
  • More local model optimizations.
  • Mobile App integration.

For more details, check the docs/ folder or join our community.

Clone this wiki locally