DeepSeek Releases the Code for an Experimental Coding Agent Anyone Can Run
DeepSeek has published the source code of a prototype coding agent that works inside one GitHub repository, taking requests made in a conversation and answering them with a pull request, a proposed change to the code. The code is public under the MIT license, which lets anyone run, change and redistribute it.
The repository's README describes the program, dsh-superagent, as a proactive coding agent packaged as a bundle for DeepSeek Harness, the company's own agent runtime. It listens to GitHub and, optionally, to Feishu. When someone with permission asks it for work, it opens a work item, handles that item in its own session and its own directory, and replies in the original conversation, for example with a pull request. A web panel shows the items that need a person. The stated aim is to "take as little human attention as possible, and give as much value as possible," and the agent acts only when it is asked or when a rule the team has written allows it.
GitHub's record for the repository puts its creation on Oct. 31, 2026, with one commit on the main branch and TypeScript as the language.
The README calls the project a "Research prototype for reference only, not an official DeepSeek product," says much of the engineering was done by coding agents, and asks anyone using it to review and adapt the code first. DeepSeek also says it adapted the project for publication, so the public version is not identical to the one the company runs internally or to the implementation described in the article the prototype accompanies. It also says some designs in that article are future directions rather than working features. No benchmark, evaluation or performance figure is attached to the release.
In September 2026, the company also published downloadable weights for a V4 model that reads images.
The README says setup takes many steps and asks the user's own coding agent to read the repository's SETUP.md file and run the installation with them. The documentation ships as numbered documents in the repository.
