Blog

DeepSeek Super-agent (dsh-superagent): What It Is, How to Install It, and How to Use an API Key

TL;DR

  • What it is: dsh-superagent, the "Super-agent" DeepSeek open-sourced on October 31, 2026, is a coding agent that works on one GitHub repository. You @-mention it in an issue, it does the work in its own session and answers there, for example with a pull request. DeepSeek calls it a research prototype, "not an official DeepSeek product".
  • What it runs on: it is a plugin bundle for DeepSeek Harness (dsh), not a separate terminal tool like Claude Code. It uses the model of its DeepSeek Harness profile; the shipped default is DeepSeek's official API with deepseek-flash (DeepSeek-V4.1-Flash).
  • Your own API key: DeepSeek Harness accepts any OpenAI-compatible provider, so Super-agent can run on a gateway key. The config is below; we ran it and the requests reached the gateway.

Short answer

Install DeepSeek Harness, build the Super-agent bundle from deepseek-ai/dsh-superagent-prototype and add it to a separate superagent profile:

bash
npm install -g @deepseek-ai/dsh@0.2.1-rc.3
git clone https://github.com/deepseek-ai/dsh-superagent-prototype.git dsh-superagent && cd dsh-superagent
pnpm install && pnpm build
dsh superagent --from-default-profile web --dump-config > /dev/null
dsh plugin --profile superagent add "link:$PWD"

Then give it a model, a GitHub repository and a dedicated GitHub account for the agent, and start it with scripts/dsh-superagent start. Its panel runs at http://127.0.0.1:3082.

To run it on a Moyi API key instead of a DeepSeek account, add a provider before the first start (full steps below):

yaml
# ~/.dsh/settings.yaml
llm-pi-ai:
  providers:
    moyiapi:
      displayName: Moyi API
      apiKeyEnv: MOYIAPI_API_KEY
      api: openai-completions
      baseURL: https://api.moyi-api.com/v1
      models:
        - id: deepseek-v4-pro
          name: DeepSeek V4 Pro
        - id: deepseek-flash
          name: DeepSeek V4.1 Flash
agent-default-model:
  provider: moyiapi
  model: deepseek-v4-pro

What Super-agent is, and what it isn't

According to the README, dsh-superagent is "a proactive coding agent for one GitHub repository". It listens to GitHub and, optionally, Feishu chats:

  1. Someone on the allowed list asks for work, for example by commenting @<agent login> <request> on an issue.
  2. Super-agent opens a work item, called a Ticket, and works on it in its own agent session and directory.
  3. It answers in the original conversation, for example with a pull request. A web panel shows what needs a person.

Its design rule is to "take as little human attention as possible, and give as much value as possible": it acts only when asked or when a rule the team wrote allows it. Out of the box it runs in observe mode, which records proactive work without doing it.

What it isn't:

  • Not an official product. The README calls it research code that "may still contain bugs", adapted for publication and not identical to DeepSeek's internal version. Some ideas in DeepSeek's Super-agent article are not implemented in it.
  • Not a terminal assistant. Claude Code and Codex work with you in your terminal. Super-agent runs as a service on your machine and takes requests from your team on GitHub. For hands-on coding in DeepSeek Harness itself, see the dsh install guide.
  • Not sandboxed. Its worker and router presets run with danger-full-access as your OS user, so a worker can run programs and read your home directory. DeepSeek recommends a dedicated OS account that holds only the agent's credentials if you want isolation.

What you need

From docs/006-deployment.md:

Requirement Version or note
Node.js 22.19+ or 24+
pnpm, git, Python, tmux git 2.32+, Python 3.8+
GitHub CLI (gh) 2.48 or later
DeepSeek Harness ^0.2.1-alpha.2; the guide pins 0.2.1-rc.3
Repository On github.com; GitHub Enterprise Server isn't supported
Agent account A separate GitHub user with write access, signed in through its own gh wrapper. A GitHub App can't be the agent
Model A model provider configured for DeepSeek Harness
Feishu (optional) lark-cli and a Feishu or Lark bot app

Install

DeepSeek's recommended route is to let a coding agent do the setup with you. Open the cloned repository in Claude Code, Codex or DeepSeek Harness and ask: "Read SETUP.md in this repository and set up dsh-superagent with me." It asks one question at a time (repository, agent account, who may ask for work, checkout directory, commit identity, auto-merge), shows the whole plan once and then installs everything.

The manual route:

  1. Run the five install commands from the short answer above. The fourth one creates ~/.dsh/profiles/superagent from the Web template. Run it only once: on an existing profile it replaces the profile and deletes its configuration.
  2. Copy examples/profile.cordis.patch.yml to ~/.dsh/profiles/superagent/cordis.patch.yml and set stateDir to an absolute path, for example /home/me/.dsh/state/superagent.
  3. Create the agent's GitHub account and its gh-agent wrapper (the Accounts section of the deployment guide).
  4. Start it:
bash
scripts/dsh-superagent start
scripts/dsh-superagent status

On a fresh install status exits with code 3 and says what is missing, for example "The GitHub Source names no repository". Open the panel at http://127.0.0.1:3082 and fill in Sources, Policy and Execution there.

Which model it uses

Super-agent uses the default model of its DeepSeek Harness profile unless the Eval or Execution settings name another. In DeepSeek Harness 0.2.1-rc.3 the shipped default is provider deepseek-official with model deepseek-flash, which DeepSeek Harness lists as DeepSeek-V4.1-Flash with a 1M-token context. The other official model is deepseek-v4-pro.

Official DeepSeek API prices per million tokens, from DeepSeek's pricing page on November 2, 2026:

Model Input, cache miss Input, cache hit Output
deepseek-flash (V4.1 Flash) $0.30 peak / $0.15 off-peak $0.006 / $0.003 $1.20 / $0.60
deepseek-v4-pro $1.32 / $0.66 $0.044 / $0.022 $3.96 / $1.98

Peak hours are 01:00–04:00 and 06:00–10:00 UTC on weekdays. DeepSeek also says that from November 1, 2026 (Beijing time) the Search and Fetch calls DeepSeek Harness makes through its API are billed per request ($0.75 per 1,000). Current Moyi API prices are on the pricing page.

Run Super-agent on a Moyi API API key

New accounts start with a $0 balance — top up before your first request. Top up from $5. If a request returns 402 (400 on the OpenAI format, which DeepSeek Harness uses here) with "wallet balance is insufficient", your balance is empty — your base URL and key are fine. Top up and retry.

Moyi API gives you one key for DeepSeek, GPT, Gemini, Kimi, GLM and Claude, billed per token from a prepaid balance with no subscription. Top up from $5 by card.

  1. Create an API key in the Moyi API console.
  2. Before the first scripts/dsh-superagent start, save the provider block from the short answer as ~/.dsh/settings.yaml.
  3. Put the key in ~/.dsh/.env:
bash
echo 'MOYIAPI_API_KEY=sk-your-key' >> ~/.dsh/.env
chmod 600 ~/.dsh/.env
  1. Start Super-agent. On its first start DeepSeek Harness copies settings.yaml into the superagent profile as llm-pi-ai and agent-default-model entries and renames the file to settings.yaml.imported.

If you have run DeepSeek Harness before, settings.yaml may already have been imported into another profile, and a new one is only imported by the next profile you start. In that case add the same two entries to ~/.dsh/profiles/superagent/cordis.patch.yml yourself, after the dsh-superagent entry:

yaml
- id: llm-pi-ai
  config:
    providers:
      moyiapi:
        displayName: Moyi API
        apiKeyEnv: MOYIAPI_API_KEY
        api: openai-completions
        baseURL: https://api.moyi-api.com/v1
        models:
          - id: deepseek-v4-pro
            name: DeepSeek V4 Pro
          - id: deepseek-flash
            name: DeepSeek V4.1 Flash
- id: agent-default-model
  config:
    provider: moyiapi
    model: deepseek-v4-pro

We tested this on November 2, 2026 with DeepSeek Harness 0.2.1-rc.3 and the prototype's first commit (6337fc8): the superagent profile picked up the Moyi API provider, Super-agent started and waited for its GitHub setup, and a request with these entries went to api.moyi-api.com. To check your own key without setting up GitHub, run one task through the same entries:

bash
dsh headless --patch ~/.dsh/profiles/superagent/cordis.patch.yml "Reply with one word: pong"

A "wallet balance is insufficient" error means the request reached Moyi API and your balance is empty; a 401 means the key in ~/.dsh/.env is wrong.

FAQ

Is Super-agent an official DeepSeek product? No. The README calls it a research prototype for reference only, with code that may still contain bugs and that differs from DeepSeek's internal version.

Is it a Claude Code alternative? Not directly. Claude Code is an interactive assistant in your terminal; Super-agent is a service that takes requests from your team on GitHub and answers with pull requests. The closer DeepSeek counterpart to Claude Code is DeepSeek Harness itself: see DeepSeek Harness vs Claude Code, Codex and OpenCode.

Do I need a DeepSeek account? Only for the default deepseek-official provider. With an OpenAI-compatible provider such as Moyi API, the key and base URL come from that provider.

Can it work on several repositories? The prototype works on one github.com repository per installation.

Next steps

Ready to connect?Log in · top up · create an API key — three steps to start.