MCP e chamadas de ferramentas

QwenPaw

agentscope-ai/qwenpaw

A self-hostable personal AI assistant that runs locally or in the cloud, featuring multi-channel connectivity, extensible skills, and a three-layer memory system.

Abrir repositório oficial ↗ Página inicial do projeto ↗
★ 33,9KEstrelas
⑂ 3KForks
947Problemas em aberto
PythonIdioma
Apache-2.0Licença
Q@project.QualityScorePontuação editorial

Capturas de tela do projeto

Captura de tela de QwenPaw Captura de tela de QwenPaw Captura de tela de QwenPaw Captura de tela de QwenPaw Captura de tela de QwenPaw Captura de tela de QwenPaw

Visão geral

QwenPaw is an open-source personal AI assistant developed by agentscope-ai. Built on the AgentScope 2.0 architecture, it provides an Agent OS environment that manages resources, governance, and sandboxing for autonomous agents. It supports local execution via QwenPaw-Flash models or integration with over 14 cloud providers. The system is designed to be reachable across various messaging platforms and offers a unified file workspace, multi-agent orchestration, and continuous memory evolution.

Principais recursos

  • Three-layer memory with self-evolving personal knowledge base (ReMe)
  • Local or cloud deployment with no mandatory API keys
  • Kernel-level Sandbox, Tool Guard, File Guard, Skill Scanner, and Access Policy
  • Multi-agent and parallel execution with Agent Communication Protocol (ACP)
  • Unified file workspace with navigation, preview, editing, and diffs
  • Extensible Skills and Plugin architecture with marketplace and MCP integration
  • Multi-channel connectivity (DingTalk, Lark, WeChat, Discord, Telegram, iMessage, QQ)
  • Console, Terminal UI (TUI), and desktop application interfaces
  • Scroll Context for full conversation persistence without summarization

Requisitos, instalação e início rápido

Multiple installation methods are supported: 1) Pip install: `pip install qwenpaw`, then `qwenpaw init --defaults`, then `qwenpaw app`. 2) Script install for macOS/Linux: `curl -fsSL https://qwenpaw.agentscope.io/install.sh | bash`. 3) Script install for Windows (PowerShell): `irm https://qwenpaw.agentscope.io/install.ps1 | iex`. 4) Docker: `docker pull agentscope/qwenpaw:latest` followed by `docker run -p 127.0.0.1:8088:8088 -v qwenpaw-data:/app/working -v qwenpaw-secrets:/app/working.secret -v qwenpaw-backups:/app/working.backups agentscope/qwenpaw:latest`. 5) Desktop Application (Beta) available for Windows 10+ and macOS 14+. 6) Cloud deployment options available via Alibaba Cloud ECS, AgentScope Platform, and ModelScope Studio.

Uso

After installation, run `qwenpaw app` and open the Console at http://127.0.0.1:8088/ to configure models. For terminal users, run `qwenpaw` to open the TUI, `qwenpaw tui --resume <id>` to resume a session, or `qwenpaw .` to start in Coding Mode. API keys for cloud providers can be configured in Console under Settings -> Models, via `qwenpaw init`, or through environment variables.

Compatibilidade do modelo e casos de uso

Supports QwenPaw-Flash series models (2B, 4B, 9B with Q4 and Q8 quantizations) via the built-in QwenPaw Local runtime. Also compatible with Ollama, LM Studio, and cloud providers including DashScope/Qwen, OpenAI, Anthropic, Google Gemini, DeepSeek, Kimi, and OpenRouter.

Observações sobre licença e riscos

Licensed under the Apache-2.0 license.

Editorial verification 2026-08-09: repository URL, owner, description, license and repository statistics were reviewed. License metadata: Apache-2.0. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

Lançamento e manutenção

v2.0.1 (2026-07-24) introduced PawApp mini-app platform, user-editable Agent Modes, Oh-My-Paw plugins, and ReMe memory enhancements. v2.0.0 (2026-07-10) was a ground-up rewrite based on AgentScope 2.0, delivering Agent OS architecture, Loop Engineering, Scroll Context, ReMe v0.4, and a Terminal UI. v1.1.12 (2026-06-17) overhauled the Models page and added Simple Mode. v1.1.11 (2026-06-10) added Free Model OAuth, Plugin Market, and MCP Tool Whitelisting.

MemPalace

mempalace/mempalace

★ 58KPython

Composio

composiohq/composio

★ 29,5KTypeScript