×
Community Blog HiClaw Joins AgentScope, Partnering with CoPaw to Build Multi-Agent Infrastructure

HiClaw Joins AgentScope, Partnering with CoPaw to Build Multi-Agent Infrastructure

This article introduces HiClaw joining AgentScope to partner with CoPaw in building enterprise multi-agent infrastructure.

Recently, the HiClaw GitHub repository was migrated under AgentScope, and it will work together with CoPaw to build multi-Agent infrastructure.

I. Background

Over the past month, Alibaba open-sourced two Claw-form projects, CoPaw and HiClaw. Both have received positive community feedback, and their star counts have grown rapidly. Specifically:

CoPaw: A personal intelligent assistant. It has been deeply optimized for small models, security mechanisms, multi-agent collaboration, and memory management, and is committed to becoming a truly controllable and stable personal intelligent assistant.

HiClaw: Directly enterprise-facing, acting as an enterprise Agent team. It adopts a Manager-Workers collaboration architecture, where the Manager centrally schedules multiple Workers, focusing on collaboration scenarios between people and Agents, and between Agents within enterprises.

After the migration, HiClaw, together with CoPaw, will become part of AgentScope and jointly advance the evolution of multi-Agent systems.

II. HiClaw + CoPaw

The collaboration between both sides will focus on multiple directions, aiming to jointly build more intelligent, secure, standardized, and user-friendly multi-agent infrastructure, including but not limited to:

Improve consistency across different Agent building methods: In the Manager-Workers collaboration architecture, HiClaw allows users to create Managers and Workers using diverse intelligence cores and model services, such as CoPaw, OpenClaw, or even ZeroClaw. Both sides will create a more consistent and smooth experience across personality definition (SOUL.md), memory systems (MEMORY.md), skill declarations (Skills.md), and runtime configuration.

Optimize collaboration experiences between people and Agents, and among multiple Agents: Single-Agent systems still have many collaboration experience issues, and multi-Agent collaboration faces even more complex engineering challenges. Both sides will improve experiences in long-horizon and complex tasks, as well as enterprise collaboration scenarios, including task assignment, state synchronization, context sharing, heartbeat monitoring, and fault isolation.

Strengthen system support during Agent runtime: Enterprise environments naturally impose higher requirements for Agent identity authentication, permission control, resource scheduling, observability, and lifecycle management. Both sides will transform practices in AI gateway credential isolation, Nacos AI registry, end-to-end tracing, and sandboxing into extensible framework-side capabilities.

In addition, the Claw-form Agent has accelerated market consensus around Harness engineering.

We will also actively translate Harness engineering practices into open-source capabilities, abstracting methodology into the final mile of productivity. For example:

From monolithic orchestration to networked collaboration: Through the Matrix protocol, orchestration is extended from a single instance to a cross-instance level, forming networked collaboration. Adding an Agent can be plug-and-play, just like “onboarding a new employee.”

From black-box execution to full transparency, and then to Agent self-evolution: Interactions between people and Agents, and among Agents, can be shared, traced, and audited, while humans can intervene and correct at any time.

From self-held credentials to zero-trust governance: Structural credential isolation is achieved through the AI gateway. Agents have “badges” but no “keys.” All calls to LLMs, MCP, and external services go through unified gateway authentication and control. Security does not depend on the reliability of the Agent framework itself, but is guaranteed by the infrastructure layer.

III. HiClaw + AgentScope Ecosystem

AgentScope is a production-ready, easy-to-use Agent framework with the necessary abstractions to work with continuously improving model capabilities, and it includes built-in fine-tuning support.

Based on the AgentScope framework, a complete ecosystem has been expanded, including:

1

Copaw: A personal intelligent assistant supporting both local and cloud deployment.

HiClaw: An enterprise Agent team focused on collaboration scenarios between people and Agents, and among Agents within enterprises..

AgentScope: A flexible programming framework for building multi-agent applications with self-describing APIs.

AgentScope-Runtime: A scalable deployment infrastructure for reliably running agents in production environments.

AgentScope-Studio: A visual development environment for rapid prototyping, debugging, and monitoring of agents.

AgentScope-Samples: A curated collection of ready-to-use agent examples and templates, ranging from simple command-line tools to full-stack production-ready applications.

Skills: A carefully selected collection of skills centered around the AgentScope ecosystem and applications of CoPaw &HiClaw.

Reme: ReMe is a memory management framework designed specifically for AI agents, providing file-based and vector-based memory systems.

HiClaw will also fully leverage the capabilities of the AgentScope ecosystem to continuously improve the collaboration experience of Agent teams.

IV. HiClaw + Alibaba Cloud

In addition to open-source evolution, HiClaw will actively leverage Alibaba Cloud’s mature infrastructure. Relying on cloud products such as Alibaba Cloud AI Gateway, MSE Nacos, ACS Container Compute Service, OSS Object Storage Service, and SLS Log Service, it will transform open-source capabilities into one-stop enterprise-grade, production-ready product capabilities.

The commercial version of HiClaw will launch soon. Stay tuned.

Any xxxClaw on Alibaba Cloud is not the commercial version of HiClaw.

● HiClaw GitHub: https://github.com/agentscope-ai/HiClaw

● HiClaw DingTalk group: 167365014834

0 0 0
Share on

You may also like

Comments