DeepSeek has published the code for Super-Agent, an experimental AI agent designed to take on coding work for software teams. The code is hosted on GitHub under the name dsh-superagent and released under an MIT licence. The team described the release as a research prototype and not an official DeepSeek product. The agent can monitor a single GitHub repository and also listen to Feishu, ByteDance’s workplace chat app. When an approved person requests work, it opens a task and processes it within its own session, replying in the original chat thread with updates such as a pull request. A web panel is provided to show what tasks still require human attention. The project emphasizes minimizing human attention while maximizing value, and notes that the agent can waste the team’s attention if not properly managed. The public code differs from an internal version, and some ideas described in related material are not yet built. The release follows other industry moves toward agent-based tools, including NASA-style demonstrations of cross-repository agent work and competing offerings from other firms. DeepSeek cautions users to review and adapt the code before running it. In other coverage, Bloomberg reported on the team’s Claude Code rival preview and broader ambitions; however, the release mentions regulatory limits in the US and indicates DeepSeek has raised significant funding with a planned listing. The code is positioned as a prototype with expectations that users will validate it for their own workflows.