Alibaba's new CoPaw 1.0 targets local AI deployment with enhanced security

  • Alibaba's Tongyi lab has released version 1.0 of CoPaw, a system focusing on localized AI deployment to significantly reduce users' token consumption costs.
  • HiClaw will continue to be jointly optimized with CoPaw in the future to focus on enhancing the enterprise user experience.
Alibaba's new CoPaw 1.0 targets local AI deployment with enhanced security
(Image credit: Tongyi)

Alibaba's Tongyi lab announced that version 1.0 of its personal AI assistant named CoPaw has been officially released to the public, according to the company's announcement on Tuesday.

Built on the AgentScope ecosystem, the model is designed for local device deployment and multi-agent collaboration without relying on the cloud.

CoPaw 1.0 introduces a series of customized small models known as CoPaw-Flash for execution in local environments, improving users' privacy and security, the company said.

The software system includes a layered defense architecture to mitigate potential security risks that may arise when agents operate files.

This security architecture features a tool guard for the real-time interception of dangerous commands, as well as a file protection mechanism that guards sensitive directory paths against unauthorized access.

Personal and enterprise users can concurrently operate multiple fully isolated agent workspaces within the same instance to collaboratively process different workflow nodes.

The system utilizes an underlying driving mechanism called ReMe to manage context and long-term memory by integrating vector retrieval and full-text search technologies, Alibaba stated.

CoPaw has now been integrated into HiClaw, an open-source collaborative multi-agent system based on the Matrix room communication protocol.

HiClaw adopts a manager-worker architecture, allowing enterprise users to transparently coordinate tasks without exposing real API credentials. Going forward, HiClaw will be deeply and jointly optimized with CoPaw to provide enterprise users with a smart, secure, standardized, and user-friendly experience, the company said.

Facing a surge in demand for large language model token calls, Alibaba is hoping to capture AI market share through local deployment tools that eliminate the need for cloud reliance.

Alibaba launches full-modal AI model Qwen3.5-Omni, supporting a 256k context window and 113 languages.
Mar 31, 2026
AI News Alert
Subscribe to receive email notifications immediately when new articles about AI are published.
AI
View more channels