DeepSeek Releases Open-Source SuperAgent Coding Prototype

DeepSeek’s Harness team has released SuperAgent, an open-source research prototype designed to monitor software teams’ repository activity and chats, identify coding tasks and dispatch work to separate AI sessions. The team published the code on GitHub under the MIT licence and described it as a research prototype, not an official DeepSeek product. The system is intended to work around a single GitHub repository and can also listen to Feishu, a workplace chat app. When an approved team member requests work, SuperAgent opens a task, assigns it to an isolated session and returns results to the original conversation, potentially as a pull request. Project documentation also describes the agent as monitoring activity so it can identify work without waiting for a direct prompt, subject to team rules about when it may act. DeepSeek’s design separates coordination from coding. A persistent Router monitors messages and repository activity, maintains shared context and decides which work to send onward. Worker sessions carry out coding tasks in separate directories and can run in parallel. The team’s design write-up says workers can subscribe to relevant pull requests or chat topics so...