CoPaw

What CoPaw is

CoPaw is an open-source personal agent workstation from the AgentScope team in Alibaba's Tongyi lab, released in early 2026 and opened up shortly after. It is aimed at people who want a desktop agent they can run and inspect themselves. The official site could not be reached when checked, so the details come from the team's developer community write-ups and third-party listings.

The distinction from the hosted products in this group is ownership: you run it, you read the code, and you decide what it may touch. That is a different relationship with an agent than subscribing to one.

What it does

It provides a personal assistant that can run locally or in the cloud, built on the team's agent framework and supporting custom skills. The loop is install, configure, extend. A first major release is described as adding support for small custom models, security mechanisms and multi-agent capability, with the core agent implementing tool calling and memory management on top of the framework's reasoning agent.

Because the primary site was unreachable, the specific tool set, supported platforms and deployment requirements should be confirmed from the repository and the documentation rather than taken from a listing. An open-source project's own repository is the reliable source, and it is also where you can see how active the project is.

Who it is for

It is aimed at developers and power users who prefer an open, inspectable agent they can deploy and modify, and at researchers who want to work with a framework rather than a black box. Technically minded users who would rather own the stack than rent it fit the same audience, and they are also the people most likely to read the security code.

What to keep in mind

It is free under an open-source model, so the cost is your time plus any model or hosting fees. Because it is open source you can read and change the code, which is the main advantage, and you also inherit the responsibility that comes with it.

An agent that runs shell commands and touches files needs sandboxing and least-privilege credentials before it is pointed at anything real. The security mechanisms added in the first release should be reviewed in the code rather than assumed to work as described, because a young security layer is exactly where careful readers find problems early.

Check the licence terms, the update cadence and the issue tracker before adopting it in a serious workflow. Commit frequency and how quickly issues are answered tell you more about the project's health than a feature list does, and an agent with filesystem access is not something to adopt from a project that has gone quiet.

Confirm the repository and the documentation, and plan for the operational work that self-hosting implies.