Verified project record

agentscope-ai/QwenPaw

QwenPaw is a personal AI assistant designed for private deployment across user-owned machines or the cloud, featuring persistent three-layer memory, multi-agent execution, and multi-channel connectivity. It supports local execution via QwenPaw-Flash models without requiring an API key, though cloud LLM APIs are required if local models are not used.

Project overview

The project provides a kernel-level security sandbox and a three-layer memory architecture that keeps older conversation turns recallable on demand rather than summarizing them away.

Project type: AI Agent · Workflow · Model Runtime
Use cases: Chat Assistants · Knowledge Q&A · Coding & Development
Deployment: Refer to project documentation
License: Apache-2.0

Best for

Developers and general users who want to deploy a personal AI assistant on their own machines or the cloud with persistent memory and local execution capabilities.

Key capabilities

Maintains live working context, full verbatim history, and distilled knowledge. Older turns evict but stay recallable on demand.
Provides local execution using QwenPaw-Flash models (2B / 4B / 9B) trained for agent tasks without requiring an API key or cloud dependency.
Spawn independent agents with their own memory and skills, including sub-agents at runtime and Agent Communication Protocol (ACP) for cross-system orchestration.
Three-panel Web IDE with file tree, diff preview, and chat. Features jump-to-definition, find-references, and structural code search built in.
Kernel-level Sandbox isolates shell commands. Includes Tool Guard for command injection, File Guard to protect sensitive directories, and Skill Scanner for pre-activation scanning.
Connects to DingTalk, Lark, WeChat, Discord, Telegram, iMessage, QQ from one instance.
Full-screen terminal chat supporting streaming replies, slash commands, file pasting, and inline tool-permission prompts with the same agent backend.
Cross-platform desktop application supporting Windows 10+ and macOS 14+ that runs without manually configuring Python environments.

Limitations and risks

The desktop application is in Beta testing phase with incomplete compatibility testing and potential performance issues.
There is a risk of running dangerous shell commands or prompt injection from external skills, which is mitigated by the Sandbox, Tool Guard, and Skill Scanner.

Getting started

Install using pip (pip install qwenpaw), initialize defaults (qwenpaw init --defaults), and run the application (qwenpaw app). A GPU is optional.

Alternatives and comparisons

A local-first desktop AI assistant that orchestrates various model providers, durable memory, and extensive tools for automated work without requiring an account.
A private local-first desktop AI agent that does real work autonomously with one-click install, no terminal required, and full offline capability with BYOK support.
A cross-platform desktop AI assistant application that provides task planning, tool execution, and local model support with complete privacy and no vendor lock-in.

Project comparisons

Evidence and sources

GitHub project description: Your Personal AI Assistant; easy to install, deploy on your own machine or on the cloud; supports multiple chat apps with easily extensible capabilities.
README: Your personal AI assistant — deploy locally or in the cloud, extend with Skills & Plugins, connect across every channel.
README: Automation & scheduling: Set up recurring tasks — news digests, report generation, multi-channel broadcasting — all on your schedule.
README: Three-layer memory — live working context, full verbatim history, and distilled knowledge. Older turns evict but stay recallable on demand; nothing is summarized away or lost.
README: DingTalk, Lark, WeChat, Discord, Telegram, iMessage, QQ — one instance, all channels. Console, TUI, and desktop app for direct access.

Reviewed at: 2026-07-27T19:50:30Z

Open original GitHub project