qwenpaw

QwenPaw is an open-source personal AI assistant designed to be installed and run on your own machine or in the cloud. It is distributed via PyPI, an install script, Docker images, a desktop application (Tauri, beta), and several cloud deployment options including Alibaba Cloud ECS, the AgentScope Platform, and ModelScope Studio. Core capabilities described in the README: - Memory: a three-layer memory system combining live working context, full verbatim conversation history, and a self-evolving personal knowledge base powered by ReMe. Conversations and resources are turned into readable, editable, searchable, linked Markdown memory. - Model support: built-in QwenPaw Local runtime (llama.cpp) with QwenPaw-Flash 2B/4B/9B models trained for agent tasks, plus compatibility with Ollama, LM Studio, and 14+ cloud providers such as DashScope/Qwen, OpenAI, Anthropic, Google Gemini, DeepSeek, Kimi, and OpenRouter. - Security: kernel-level sandbox, Tool Guard, File Guard, Skill Scanner, and access policy that block dangerous commands before execution. - Multi-agent: independent agents with their own memory and skills, runtime sub-agents, and an Agent Communication Protocol (ACP) for cross-system orchestration. - Workspace: unified file navigation, preview, editing, diffs, upload, and download across project and agent files. - Extensibility: Skills for scheduling, documents, browser, news and more; a plugin architecture with a marketplace; and MCP integration for external tools. - Channels: DingTalk, Lark, WeChat, Discord, Telegram, iMessage, and QQ from one instance, plus a web Console, a full-screen terminal UI (TUI), and a desktop app. - Privacy: local deployment keeps data on the user's machine with no third-party hosting or data upload. Typical uses listed include automation and scheduled tasks (news digests, report generation, multi-channel broadcasting), code reading/editing/review/testing, document processing for PDF, Word, Excel and PowerPoint, web search and video summarization, multi-channel operations, and custom workflows combining built-in capabilities, plugins, and scheduled tasks. Installation paths: `pip install qwenpaw` (Python >= 3.11, < 3.14) followed by `qwenpaw init --defaults` and `qwenpaw app`; one-line install scripts for macOS/Linux and Windows; Docker images on Docker Hub and Alibaba Cloud ACR with volumes for data, secrets, and backups; and a desktop beta for Windows 10+ and macOS 14+. The Console runs at http://127.0.0.1:8088/ by default. Cloud LLM APIs require an API key, while local models do not. The project is licensed under Apache 2.0 and is built by AgentScope. The README also documents release notes for versions 2.0.0 through 2.2.0, covering the Agent OS architecture, Loop Engineering, Scroll Context, ReMe v0.4, a bundled TUI, a self-hosted multi-user Hub, and other features.