DeepSeek has open sourced a coding agent that is designed to work for a whole development team without demanding constant attention. The prototype, called Super-agent, is available on GitHub under the MIT licence as a bundle for DeepSeek Harness.
What Super-agent does
Super-agent is a proactive coding agent built for a single GitHub repository. It listens to GitHub and, optionally, to Feishu, the workplace messaging app, and it picks up work when a person on an allowed list asks for it.
Each request becomes a separate work item with its own agent session and directory. The agent then works on that item and answers in the original conversation, for example by submitting a pull request for the team to review.
A web panel shows what still needs a person, so the team can see which items are waiting for human input. The design is meant to keep the agent useful without flooding developers with messages about work they cannot yet use.
A different way to think about attention
The project sets out a simple working principle for agents that serve a team. It aims to take as little human attention as possible, and give as much value as possible, a goal stated in its own user manual.
That matters because an agent acting for a whole team can also waste a whole team’s attention. Super-agent therefore acts only when it is asked, or when a rule written by the team allows it, and it speaks only when someone must act or can use a result.
What it is not
DeepSeek is clear about the limits of this release. The code is a research prototype for reference only, not an official DeepSeek product, and it is not production-ready software.
The team warns that the prototype may still contain bugs and that its coding style may be inconsistent, because much of the engineering was done by coding agents. Anyone who wants to use it should review and adapt the code first.
The published version is also not identical to the company’s internal version. Some designs discussed in the accompanying design article are directions for future work and are not yet implemented in the prototype.
Why it matters
The release puts DeepSeek more directly into coding agent work, a market where Anthropic’s Claude Code is already widely used by developers. Open sourcing the prototype lets other developers inspect how it handles team requests, permissions and sessions.
The company has been in the news before for its fundraising, with an earlier report covering its rise to become China’s most valuable AI startup.
For now, the value of Super-agent is as a reference design rather than a finished tool. Teams that try it will need to judge for themselves whether its careful approach to attention is worth the work of setting it up and reviewing its code.
