GitHub - agentscope-ai/CoPaw: Your Personal AI Assistant; easy to install, deploy on your own machine or on the cloud; supports multiple chat apps with easily extensible capabilities.

Source
ai-assistantself-hostedopen-sourceagentsllmautomationlocal-modelsmulti-channel

Summary

CoPaw is an open-source, self-hostable personal AI assistant from the AgentScope team (Alibaba). It runs locally or in the cloud, connects to messaging platforms (Discord, Telegram, DingTalk, iMessage, QQ, Feishu), and supports local LLMs via llama.cpp, MLX, or Ollama - no API key required. At v0.0.7, it is early-stage but iterating quickly with 13 new contributors in a single release cycle.

Key Insight

  • No lock-in by design: custom skills live in your own workspace and auto-load; cron is built-in, so you can build agentic workflows without depending on a SaaS platform.
  • Apple Silicon support via MLX: bash install.sh --extras mlx gives native on-device inference on M-series Macs.
  • 3-command install: pip install copaw && copaw init --defaults && copaw app - genuinely low friction; Docker image also available for server deployment on tank.
  • v0.0.7 added a Tool Guard security layer - blocks risky tool calls until user approval, which is a meaningful safety feature for an agent with file and shell access.
  • Telemetry caveat: --defaults silently opts in to telemetry. Use copaw init (interactive) to decline.
  • China-focused channel integrations (DingTalk, Feishu, QQ, Xiaohongshu, Zhihu, Bilibili) suggest the primary user base and contributor pool is Chinese - worth factoring in when assessing long-term maintenance risk.
  • Roadmap items “multi-model routing” and “experience distillation & skill extraction” are still planned, not shipped - don’t evaluate it as if those exist yet.