DeepSeek SuperAgent

DeepSeek developer guide

DeepSeek SuperAgent: What It Is, How It Works, and How to Get Started

DeepSeek SuperAgent is an open-source research prototype built on DeepSeek Harness. It listens to GitHub activity for one repository, works on authorized requests or team rules, and returns results in the original conversation, such as a pull request.

Updated · DeepSeek V4 Hub

Use DeepSeek Now Get StartedView GitHub Repository

Research prototype · Not an official DeepSeek product

What is DeepSeek SuperAgent?

The public dsh-superagent prototype is a proactive coding agent packaged for DeepSeek Harness. It listens to GitHub and optionally Feishu, and can take on work when an authorized person requests it or a team rule permits it. It is an agent system, not a new model; a model provider must be configured separately.

The authors describe the release as research code to review and adapt, not production-ready software. The public prototype differs from their internal version, and some ideas in the design article are not implemented in the release. Read the project README

How does DeepSeek SuperAgent work?

A source delivers activity; the agent evaluates whether a request or configured rule authorizes work. Accepted work receives its own agent session and directory, and the agent reports back to the original conversation. Teams should review results and configure their own permissions.

  1. 01

    Receive

    A GitHub event or request reaches the agent.

  2. 02

    Decide

    Authorization and team rules determine whether to act.

  3. 03

    Work

    Accepted work runs in an agent session and directory.

  4. 04

    Report

    A result or pull request returns to the conversation.

Separate working directories help organize work; they are not a security boundary. See the execution documentation before granting access.

What do you need to get started?

Start with a test repository, a dedicated GitHub identity, a machine you control, DeepSeek Harness and a configured model provider. Decide who can request work and what permissions the agent needs. The official setup and deployment documents list the current software requirements and configuration steps.

Begin with a small task, review the resulting changes and keep human review in place. We have not verified that this prototype works with our Coding Plans; check compatibility before purchasing a plan for this use.

Read the setup and cost guide

How much does SuperAgent cost to run?

The prototype is MIT-licensed, but running it can still cost money. The total depends on the model provider and usage, the machine or server, any other services, and the time needed to configure and maintain it. Measure your own workload before estimating a monthly amount.

Explore available DeepSeek Coding Plans

How is SuperAgent different from DeepSeek Harness?

DeepSeek Harness is the agent runtime. SuperAgent is a bundle for Harness that organizes work from a GitHub repository, including authorization, execution and returning results. Configure a model provider separately for the agent to make model calls.

Explore DeepSeek Harness

SuperAgent guides and articles

Choose a starting point, from the basics to setup and the design of the agent system.

Frequently asked questions

What is DeepSeek SuperAgent?

dsh-superagent is an open-source research prototype for a proactive coding agent built on DeepSeek Harness. It works with one GitHub repository and responds to authorized requests or team rules.

Is DeepSeek SuperAgent an official DeepSeek product?

No. The project describes its public release as a research prototype for reference, not an official DeepSeek product or production-ready software.

Is DeepSeek SuperAgent free to use?

The source code is available under the MIT license, but running the agent can incur model, infrastructure and other service costs.

How is SuperAgent different from DeepSeek Harness?

DeepSeek Harness provides the agent runtime; SuperAgent is a bundle built on it that organizes authorized work for a GitHub repository.

Can SuperAgent create pull requests on GitHub?

Yes. The project describes returning results in the original conversation, for example as a pull request. Review the agent's changes and configure repository permissions before use.

Is DeepSeek SuperAgent ready for production?

The maintainers describe it as research code, not production-ready software. Review and adapt it for your environment before use.