DeepSeek has released the code for Super-Agent, an experimental AI agent that takes on coding work for a development team. The code is on GitHub as dsh-superagent under an MIT licence, and the company's Harness team, which builds its agent tooling, set out the design in a WeChat post on Saturday. DeepSeek calls the release a "research prototype" and not an official DeepSeek product.
The agent works on a single GitHub repository and can also listen to Feishu, ByteDance's workplace chat app. When an approved person asks it for work, it opens a task, handles it in a separate session and replies in the original thread, for example with a pull request. A web panel lists what is waiting for a human decision. According to Bloomberg, the agent can spot coding problems in developers' group chats and suggest fixes without being summoned; according to the project's philosophy document, it acts only when asked or when a rule set by the team allows it.
The goal, according to the South China Morning Post, is to spare people the work of coordinating several AI agents and monitoring them. The GitHub manual puts it this way: "Take as little human attention as possible, and give as much value as possible." According to the README, the public code differs from DeepSeek's internal version, and some of the ideas described have not been built yet.
DeepSeek set up the Harness team this year to compete with US labs' coding tools, and in August released a preview of its own agent app. According to Bloomberg, the company has raised about $15 billion ahead of a planned listing. In the US, civilian federal agencies were told last month to remove DeepSeek. The company tells users to review and adapt the code before running it, and how the agent performs against rival tools is not yet known.
