DeepSeek open-sources Super-Agent, a coding agent that reads team chats

DeepSeek has released the code for Super-Agent, an experimental AI agent that takes on coding work for a software team. The Chinese AI lab put it on GitHub under the name dsh-superagent, with an MIT licence. The team that builds its agent tools set out the design in a WeChat post on Saturday.

DeepSeek is aiming the agent at Anthropic’s Claude Code, Debby Wu reported for Bloomberg. In her account, it can watch developers’ group chats for coding problems and suggest solutions without being summoned. DeepSeek itself calls the release a research prototype and “not an official DeepSeek product”.

Built to save human attention

The agent works on one 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 and handles it in its own session. It then replies in the original thread, for example with a pull request. A web panel lists what needs a human.

The aim is to spare people the job of coordinating several AI agents and checking on them. Ben Jiang reported that goal for the South China Morning Post, citing the team’s article. DeepSeek’s manual puts the rule in one line.

“Take as little human attention as possible, and give as much value as possible.” DeepSeek, in the agent’s GitHub manual

DeepSeek warns that an agent acting for a whole team can also waste the whole team’s attention. So it acts only when asked, or when a team’s own rule allows it, says the project’s philosophy document. The public code differs from DeepSeek’s internal version. Some ideas in the article are not built yet, the README says.

A race to run many agents

DeepSeek set up that team this year to take on coding tools from US labs, Bloomberg reported. In August it released a preview of a Claude Code rival, its own agent app. The open-source app, now in public preview, handles coding, documents and spreadsheets.

US companies are chasing the same idea. On Wednesday, at GitHub Universe, GitHub showed a Copilot app that runs agent work across several repositories in parallel sessions. Getting agents to share work is still hard, though. In January, researchers behind the CooperBench study found that coding agents did worse as a team. Their success rates were on average 30% lower than when each agent did the tasks alone. The study did not test DeepSeek’s agent.

The release also lands as DeepSeek faces limits in the US. Civilian federal agencies there were told to remove DeepSeek last month. The company has raised about $15bn ahead of a planned listing, Bloomberg reported. For now, the code on GitHub is a prototype. DeepSeek tells users to review and adapt it before they run it.