DeepSeek Open-Sources Super-Agent Coding Agent Prototype
AI assistants · AI coding
DeepSeek's Harness team open-sourced its Super-Agent coding agent prototype on GitHub under the MIT license. Designed for single repositories and capable of connecting to Feishu, the agent infers tasks but defaults to an observation mode that logs actions without executing them. The team emphasizes it is a research prototype.
DeepSeek's Harness team released an open-source coding agent research prototype named Super-Agent (dsh-superagent) on GitHub under the MIT license. According to project documentation and news reports, the agent is designed for a single GitHub repository and can monitor codebases or integrate with ByteDance's Feishu workspace to propose fixes. To avoid distracting development teams, the system incorporates off, observe, and auto acceptance modes, defaulting to observe, where it evaluates and logs candidate tasks without taking direct action. DeepSeek explicitly notes the release is a research prototype rather than an official commercial product, requiring external deployments to supply separate accounts and model providers.
Key takeaways
- DeepSeek's Harness team open-sourced the Super-Agent coding agent prototype on GitHub under the MIT license.
- The agent operates on a single GitHub repository, can connect to Feishu, and handles tasks by opening sessions and returning Pull Requests.
- The system incorporates off, observe, and auto acceptance modes, defaulting to observe mode to log decisions without taking action.
- Documentation highlights that the release is a research prototype rather than an official commercial product, requiring dedicated accounts and model backends.
Source：科技報橘 Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human. Translation checked by AI.