DeepSeek has released the source code for Super-Agent, an experimental AI agent designed to handle coding tasks for software development teams.
The Chinese AI company published the project on GitHub under the name dsh-superagent, making it available under an MIT licence.
The team responsible for developing its agent tools explained the system’s design in a WeChat post on Saturday.
DeepSeek is positioning the agent as a competitor to Anthropic’s Claude Code. It can monitor developers’ group chats for coding issues and proactively suggest solutions without waiting for users to prompt it, according to a Bloomberg report.
However, DeepSeek described the release as a research prototype, stressing that it was “not an official DeepSeek product”.
Super-Agent operates on a single GitHub repository and can also monitor conversations on Feishu, ByteDance’s workplace messaging platform.
When an authorised user assigns it a task, Super-Agent creates a job and executes it in a separate session. It then responds in the original chat thread with the results, which could include a pull request. A web dashboard also displays tasks that require human attention.
The system is designed to reduce the burden on developers of coordinating multiple AI agents and constantly monitoring their progress.
“Take as little human attention as possible, and give as much value as possible.” DeepSeek, stated in the agent’s GitHub manual.
DeepSeek cautioned that an AI agent operating on behalf of an entire team could also end up consuming the attention of the whole team if not properly managed.
