QwenPaw

// Overview

QwenPaw is an open-source personal AI assistant from the AgentScope team that you deploy on your own machine or your own cloud. It is built around a single long-lived assistant rather than a chat window per task: one instance keeps your memory, skills and scheduled work, and you reach it from a web console, a terminal UI, a desktop app, or from the messaging apps you already use — DingTalk, Lark, WeChat, Discord, Telegram, iMessage and QQ.

Memory is the part the project leads with. QwenPaw keeps three layers: the live working context, a full verbatim history, and a self-evolving personal knowledge base powered by ReMe that turns conversations and files into readable, editable, searchable Markdown you can inspect and correct. Its Scroll Context design persists every turn and indexes evicted turns for recall on demand instead of summarising them away.

Because the assistant runs with real tool access, security is enforced at the runtime rather than by prompt: a kernel-level sandbox on macOS, Linux and Windows, plus Tool Guard, File Guard, a Skill Scanner and an access policy that blocks dangerous commands before they execute. It can spawn sub-agents with their own memory and skills, extends through a skill and plugin marketplace and MCP, and runs either on the bundled QwenPaw-Flash local models (2B, 4B and 9B, no API key needed) or against Ollama, LM Studio and cloud providers.

// What it does

• Three-layer memory — working context, verbatim history and a ReMe-backed personal knowledge base stored as editable Markdown
• Runs fully local with the bundled QwenPaw-Flash models and local runtime, or against Ollama, LM Studio and cloud providers
• Runtime security: kernel-level sandbox, Tool Guard, File Guard, Skill Scanner and an access policy applied before a command runs
• Sub-agents at runtime with their own memory and skills, plus the Agent Communication Protocol for cross-system orchestration
• One instance reachable from DingTalk, Lark, WeChat, Discord, Telegram, iMessage and QQ, as well as a console, TUI and desktop app
• Skills, a plugin marketplace and MCP integration for scheduling, documents, browsing and other external tools

// Getting started

There are two supported paths: a normal pip install if you manage Python yourself (3.11 up to but not including 3.14), or a one-command script that downloads uv and sets everything up for you.

pip install qwenpaw
qwenpaw init --defaults
qwenpaw app

curl -fsSL https://qwenpaw.agentscope.io/install.sh | bash

# then, in a new terminal
qwenpaw init --defaults
qwenpaw app

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Run a private assistant that keeps years of context on your own hardware, with nothing uploaded to a third party
• Schedule recurring work — news digests, report generation, multi-channel broadcasts — and have the results pushed to your chat apps
• Give an assistant guarded access to real files and shell commands, with a sandbox and policy gate deciding what it may run
• Reach the same assistant, with the same memory, from a terminal, a desktop app and a phone messaging client

// How QwenPaw compares

QwenPaw alongside other open-source assistants & chatbots tools AI/TLDR tracks, ranked by GitHub stars.