HKUDS/AgentSpace

What it solves

AgentSpace addresses the gap between individual agent use and organizational collaboration. Most agent frameworks are designed for solo users, leaving powerful agents locked in private terminals and creating fragmented contexts, scattered execution paths, and a lack of of governance when teams try to scale their use of AI agents.

How it works

It provides an agent-native collaborative workspace where humans and agents are treated as a single team. The system uses an "AgentRouter" to normalize different agent runtimes (such as Claude Code, Codex, and OpenClaw) into a unified execution contract. This allows agents to maintain a stable identity and context while being routed to the best-fit runtime for a specific task. The platform includes a digital employee board for sharing agents across the organization and a centralized permission control plane to manage access to documents, tools, and credentials.

Who it’s for

It is designed for teams and organizations—particularly small founder teams—that want to integrate AI agents into their daily operations as "digital employees" with defined roles, accountability, and human-in-the-loop governance.

Highlights

• AgentRouter: A normalization layer that allows one agent to work across multiple provider runtimes without losing its identity or skills.
• Digital Employee Board: Turns private agents into shared organizational assets that can be discovered and borrowed by teammates.
• Governance Framework: Centralized control over permissions, approvals for sensitive actions, and full audit trails of agent decisions.
• Multi-Agent Coordination: Enables agents to collaborate across channels, documents, and task boards with human approval gates for high-impact actions.