QwenPaw
ActiveDescription
Personal AI assistant from the AgentScope team with an Agent OS architecture, three-layer memory, kernel-level sandboxing, and one instance serving DingTalk, WeChat, Telegram and more.
Key Features
- Three-layer memory — live working context, full verbatim history, and a ReMe-powered self-evolving knowledge base that becomes searchable Markdown memory
- Local or cloud, runs free — bundled QwenPaw-Flash models (2B/4B/9B) need no API key, and Ollama plus 14+ cloud providers also work
- Security built in — kernel-level Sandbox, Tool Guard, File Guard, Skill Scanner, and Access Policy block dangerous commands before they run
- Multi-agent and parallel — spawn sub-agents with their own memory and skills, orchestrated across systems via the Agent Communication Protocol
- Reachable anywhere — one instance covers DingTalk, Lark, WeChat, Discord, Telegram, iMessage, and QQ, plus Console, TUI, and desktop apps
- Skills and plugin marketplace — scheduling, documents, browser skills built in, with a plugin marketplace and MCP integration
Use Cases
💡 Run a fully self-hosted assistant on your machine or server and reach it from WeChat, DingTalk, and other daily channels
💡 Configure recurring jobs like news digests, report generation, and multi-channel broadcasts to run on schedule
💡 Let the assistant read and edit code and Office documents through the unified file workspace
💡 Accumulate personal preferences and project context in the self-evolving knowledge base so the assistant improves with use
Strengths & Limitations
✅ Strengths
- • Actively maintained, recent updates
- • High community interest (34.8k stars)
- • Permissive open-source license (Apache-2.0)
Categories
Quick Start
# Option 1: pip install (Python 3.11+)
pip install qwenpaw
qwenpaw init --defaults
qwenpaw app
# Open http://127.0.0.1:8088 to configure your model, or use the script installer:
# curl -fsSL https://qwenpaw.agentscope.io/install.sh | bash