AgentScope's QwenPaw 2.0 rebuilds the personal AI assistant on an Agent OS — GitHub
The Apache-2.0 personal assistant, past 22,000 stars, shipped a ground-up 2.0 rewrite with an Agent OS architecture and long-term memory, running locally across DingTalk, Lark, WeChat, Telegram, and iMessage.
Positive-sum angle: A local-first, multi-channel assistant becomes a shared substrate that smaller builders extend rather than rebuild. Every connector it ships and every self-hosted deployment widens common infrastructure, turning a private cross-app agent into a base the whole ecosystem draws on.
What's the impact: SEA operators wiring workflows across WeChat, Lark, and Telegram, the region's cross-border business rails, should evaluate local-first agent frameworks before committing to a single cloud vendor. Own the gateway and the data-residency story regulators keep tightening.