QwenPaw (formerly CoPaw) is an open source AI personal assistant developed by Alibaba’s AgentScope team. Released under the Apache License 2.0, it can run on your own computer, in Docker, or on a cloud server. Its defining feature is that you can reach the same assistant from the chat apps you already use — DingTalk, Lark, WeChat, Discord, Telegram, iMessage, and QQ. In April 2026, version 1.1.0 renamed the project from “CoPaw” to “QwenPaw,” formally folding it into Alibaba’s Qwen open source ecosystem. As of August 2026, the latest release is in the v2.1.0 line.
Key Features
- Three-layer memory: The assistant remembers through three layers — the live working context of recent exchanges, the full conversation history, and a self-evolving knowledge base powered by ReMe. The design means you need to explain your background less as you keep using it
- One instance across multiple chat apps: Supports DingTalk, Lark, WeChat, Discord, Telegram, iMessage, and QQ. Even if you use Lark for work and Discord privately, you get the same assistant with the same memory
- Deployment options from local to cloud: Install via pip or a setup script, use the Tauri-based desktop app, run it in Docker containers, deploy to Alibaba Cloud ECS with one click, or host it on the AgentScope Platform
- Extensible through skills and MCP: Add skills for PDF and Office document handling, browser operation, news retrieval, and more, and connect external tools that speak MCP (Model Context Protocol). A REST API and ACP (Agent Communication Protocol) are also included
- Scheduled task automation: A built-in Cron-based scheduler handles recurring jobs such as “summarize this information and send it to me every morning”
- Sandbox design that errs on the side of safety: Tool Guard, File Guard, and Skill Scanner control tool execution, file access, and the import of external skills — an important part of any tool meant to run on a personal machine
- QwenPaw-Flash lightweight models: Small models optimized for agent work in 2B / 4B / 9B sizes. Combined with local models via Ollama or LM Studio, you can run the whole setup without any API key
Pricing
| Item | Cost | Details |
|---|---|---|
| The software itself | Free | Open source under the Apache License 2.0. Install it on your own machine or server |
| LLM usage | Separate (usage-based) | Using cloud APIs such as DashScope (Qwen) or OpenAI incurs each provider’s fees. QwenPaw-Flash and local models require no API key at all |
| Cloud runtime | Check with the provider | Running on Alibaba Cloud ECS or the AgentScope Platform follows each platform’s own pricing |
Pricing reflects information as of August 2026. Check the official site and the GitHub repository for the latest details.
Pros & Cons
✅ Pros
- Being open source, you can keep your data entirely within your own machine or server
- The chat apps you already use become the entry point, so there is no separate app to open
- Memory accumulates across conversations, so you repeat your context far less often
- Local model support means you can run it with zero API spend
- Extensible through both skills and MCP, making it easy to add capabilities over time
⚠️ Cons
- Setup requires a Python environment (3.11–3.13) or container knowledge; this is not a one-click service
- Chat app integration requires registering bots and issuing tokens on each service, which takes effort up front
- Support for Chinese chat apps (DingTalk, Lark, WeChat, QQ) is thorough, but check ahead if you rely on Slack or Microsoft Teams
- Development moves fast and the structure can change between versions — the project itself was renamed from CoPaw to QwenPaw in April 2026
- Running local models calls for a reasonably capable machine
Comparison with Similar Services
| Criteria | QwenPaw | Open WebUI | LibreChat | Dify |
|---|---|---|---|---|
| Provider | Alibaba (AgentScope) | Community | Community | LangGenius |
| Main use | Personal AI assistant | Chat UI for various LLMs | Multi-model chat UI | LLM app development platform |
| Chat app integration | DingTalk, Discord, Telegram, iMessage, etc. | Its own web UI | Its own web UI | Embedded into your own app via API |
| Local execution | Yes | Yes | Yes | Yes |
| Main extension method | Skills + MCP | Functions and pipelines | Agents and MCP | Workflows and plugins |
| Long-term memory | Three-layer memory built in | Via settings or extensions | Conversation history focused | Designed on the app side |
Who Is It For
- People who do not want to hand business data or personal records to an external service and prefer to run an AI assistant in their own environment
- People who would rather talk to an AI from the chat app they already have open than reopen a browser chat window
- Anyone who wants to delegate scheduled work such as a morning information roundup or recurring reminders
- Users already running local models with Ollama or LM Studio who want a practical assistant layer on top
- Developers who want to build and connect their own MCP tools into a personal agent environment
Summary
QwenPaw (formerly CoPaw) is an open source AI personal assistant built around three ideas: it runs on your own hardware, it answers from the chat apps you already use, and it remembers more the more you use it. Setup requires a working knowledge of Python and containers, so it is not something to recommend to everyone. What it offers in return is hard to find in cloud-only chat services: an assistant with memory, scheduled tasks, and MCP extensibility that stays entirely under your control. A realistic path is to follow the GitHub quickstart to run it locally first, then expand into chat app integration and additional skills once you are comfortable.