QwenPaw: Run a Self-Hosted AI Assistant Across Chat Apps

Overview

QwenPaw is a self-hostable personal AI assistant for using agents across a web console, terminal, desktop app, and messaging channels. It combines model connections with tools, scheduled tasks, agent memory, and configurable access controls, addressing the gap between a chat interface and an assistant that can work with files and workflows over time.

It can use local models or cloud providers. Its data and configuration can remain on a machine you control, while integrations and extensions let teams shape the assistant for their own tasks.

Key Features

• Connects to messaging channels including DingTalk, Lark, WeChat, Discord, Telegram, iMessage, and QQ.
• Supports local inference through QwenPaw Local, Ollama, and LM Studio, as well as cloud model providers.
• Maintains working context, conversation history, and a searchable, editable Markdown knowledge base powered by ReMe.
• Extends capabilities with skills, plugins, MCP tools, and scheduled tasks.
• Supports multiple agents, sub-agents, and cross-system orchestration through ACP.
• Provides a file workspace for browsing, editing, previews, and diffs.
• Includes sandboxing, tool and file guards, skill scanning, and configurable access policies.
• Offers a browser console, terminal UI, and beta desktop application.

Use Cases

• Individuals can run a persistent assistant on their own machine and use it through the chat apps they already use.
• Developers can ask an agent to work with a project workspace, review or edit files, and use a coding-oriented mode.
• Small teams can schedule digests or reports and distribute them through one or more messaging channels.
• Users who prefer local inference can connect supported local model runtimes instead of relying on a cloud API.
• Operators can combine skills, plugins, and MCP tools into workflows, with approval and sandbox controls for tool access.

Getting Started

With Python 3.11 through 3.13 installed:

pip install qwenpaw
qwenpaw init --defaults
qwenpaw app

Open http://127.0.0.1:8088/ to configure a model. See the README for script, Docker, desktop, and cloud deployment options.

Alternatives

• cognithor: Cognithor emphasizes a locally controlled agent operating system, while QwenPaw highlights configurable agents with skills, plugins, and MCP tools.
• nanobot: nanobot focuses on a lightweight Python agent framework with scheduled workflows, while QwenPaw centers on a persistent, extensible personal assistant.
• core: CORE emphasizes task planning and coding agents integrated with apps, while QwenPaw focuses on configurable agents and extensions across chat, web, and terminal.

Considerations

• The project is changing quickly, with regular commits and releases and active issue and pull-request handling. That pace brings ongoing change, so check release notes and test upgrades against your deployment.
• Cloud model use requires a provider API key. Local use avoids that requirement but depends on setting up a compatible local model backend.
• The desktop app is explicitly beta, with incomplete compatibility testing and possible performance or feature gaps.
• The project is licensed under Apache-2.0. Review the license terms when distributing or modifying it, and assess how its sandbox and policy settings fit your threat model.
• The documented Python support is limited to versions 3.11 through below 3.14. The README also notes that the installer may not work in restricted network environments.