DeepSeek Open-Sources dsh-superagent Router-Worker Coding Agent
DeepSeek releases dsh-superagent under MIT license, turning GitHub and Feishu requests into parallel worker sessions while cutting p95 task time.
Stock photo for illustration only, not from the actual event
- DeepSeek released dsh-superagent as an MIT-licensed research prototype
- The tool transforms requests from GitHub and Feishu into parallel worker sessions
- DeepSeek simulation shows p95 task completion dropped to 15.7 minutes
- The previous completion time in simulation baseline stood at 29.5 minutes
DeepSeek has expanded its open-source offerings with the release of dsh-superagent, a research prototype distributed under an MIT license. Designed to streamline software development workflows, this new tool aims to bridge the gap between platform requests and automated task execution.
At its core, dsh-superagent acts as an intelligent system that converts incoming requests from GitHub and Feishu into parallel worker sessions. By routing tasks efficiently, the system allows multiple sub-processes to handle complex coding assignments concurrently rather than sequentially.
Stock photo for illustration only, not from the actual event
To demonstrate its capability, DeepSeek ran internal simulations measuring performance outcomes. The results highlight a dramatic improvement, with the p95 task completion time dropping from 29.5 minutes down to just 15.7 minutes.
The router-worker architecture utilized in dsh-superagent represents an evolving trend in AI engineering: shifting away from monolithic prompt executions toward coordinated multi-agent pipelines. By delegating specific tasks from a central router to parallel workers, frameworks can bypass traditional latency bottlenecks in automated software engineering.
Developers and researchers interested in experimenting with the code or inspecting the repository can access the MIT-licensed prototype through DeepSeek's official development channels and open-source code repositories.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment