AgentSpace

AgentSpace is an agent-native collaborative workspace for teams that want humans and AI agents to work in the same operating context. The GitHub repository describes the product as “Human + Agents. One Team. One Workspace.” Instead of treating agents as private terminal sessions or single chatbots, AgentSpace models them as digital teammates with roles, owners, responsibilities, permissions, approvals, and audit trails. The product is aimed at a problem that appears quickly once agents leave demos: collaboration and governance. Solo agent tools can write code or perform tasks, but teams need to know who owns the agent, what it is allowed to do, how sensitive actions are approved, what happened, and how work moves across people and departments. AgentSpace addresses that with a shared workspace, agent recruitment and assignment, multi-agent coordination, scheduling, permissions, approvals, audit logs, and transfer workflows. The public repository also matters for technical evaluation. It is a TypeScript-heavy project under the HKUDS organization and documents both hosted and self-hosted modes. The hosted option points to a platform for immediate use, while the self-hosted option lets teams run the system locally with more control over data, infrastructure, provider CLIs, runtime machines, and deployment policy. The repository timeline and README describe AgentRouter support for Claude Code, Codex, OpenCode, OpenClaw, Hermes, and related CLI-based agents. Pricing is recorded as source-backed context rather than a fixed public plan because the reviewed signal is the repository and hosted platform page, not a formal pricing page. Users should verify current hosted pricing, self-hosting requirements, provider model costs, and any enterprise terms before rolling it out. Self-hosting may reduce vendor dependency but still requires infrastructure, model accounts, authentication, storage, and operational maintenance. AgentSpace is a tool, not an LLM and not an MCP server. Its durable entity is the workspace and coordination layer around agents. It is most useful for builders who already believe agents can perform meaningful work but need a more accountable place to assign, observe, approve, and transfer that work. The caveat is maturity: teams should pilot with low-risk workflows, verify permission boundaries, and inspect audit logs before letting agents touch production systems. Source snapshot: GitHub reported 1003 stars, 135 forks, primary language TypeScript, license Apache-2.0, and latest push 2026-07-24. A sensible rollout starts with one non-production workflow, such as research, internal documentation, issue triage, or repository maintenance. That lets a team test the workspace model without handing agents broad production authority. During the pilot, reviewers should inspect how AgentSpace records actions, how approvals are requested, how agent ownership is represented, and how easy it is to pause or transfer work. The benefit is not just more automation; it is a clearer management layer for automation that would otherwise live in private chats, terminals, and undocumented scripts.