QwenPaw

QwenPaw (Qwen Personal Agent Workstation) is an open-source personal AI assistant built for local or cloud deployment, giving users full control over their data and memory. It supports multiple messaging channels including DingTalk, Feishu, WeChat, Discord, and Telegram, and can be extended with custom skills and plugins. The agent learns from interactions, evolves its memory over time, and proactively serves users — getting smarter the more it is used. It includes multi-layer security features including tool guards, file access controls, and skill security scanning.

• Local or cloud deployment — Install via pip, script, Docker, desktop app, or ModelScope Studio; data stays on your machine when deployed locally.
• Multi-channel support — Connect to DingTalk, Feishu, WeChat, Discord, Telegram, QQ, iMessage, and more from a single agent instance.
• Skills extension — Built-in scheduling, PDF/Office processing, and news digest; add custom skills that are auto-loaded with no vendor lock-in.
• Multi-agent collaboration — Create multiple independent agents with distinct roles and enable inter-agent communication for complex task execution.
• Memory-evolving & proactive — Agent learns from interactions, reflects on experience, and proactively delivers relevant information.
• Local model support — Run LLMs entirely on-device via llama.cpp, Ollama, or LM Studio — no API keys required.
• MCP integration — Manage MCP clients directly from the console for extended tool connectivity.
• Multi-layer security — Tool guard intercepts dangerous shell commands; file access guard restricts sensitive paths; skill security scanning detects prompt injection and data exfiltration risks.
• Web Console UI — Browser-based interface at http://127.0.0.1:8088/ for chat, agent configuration, model settings, and environment variables.
• Cross-platform install — Supports Windows, macOS, and Linux via pip, shell script, Docker, or desktop application (Beta).