What is DeepSeek Super-agent?
DeepSeek Super-agent is DeepSeek's open prototype of a coding agent that lives in a repo's conversations. It listens to GitHub and, optionally, Feishu. It sits in the Autonomous Coding Agents category of the AI Coding tools and is open-source, written in TypeScript and licensed MIT. To get started: hand the setup to a coding agent.
Overview
dsh-superagent is an open-source research prototype from DeepSeek of a proactive coding agent for one GitHub repository. It listens to GitHub and, optionally, Feishu. When an allowed person asks for work, it opens a work item, does the job in its own agent session and directory, and replies in the original conversation, for example with a pull request. A web panel shows the items that need a person.
It is packaged as a bundle for DeepSeek Harness (dsh). Source plugins turn incoming payloads into Triggers; a main router session evaluates each one and can accept it as a Ticket with its own Mandate (its authority), Subscriptions and an Assignment. An executor runs one session and one directory per Ticket from a limited set of worker slots, all state sits in a single SQLite database, and an Outbox sends replies back through the source the request came from.
The guiding rule in its design notes is to "take as little human attention as possible, and give as much value as possible." The agent acts only when an allowed person addresses it or when a rule the team wrote allows it, and proactive work starts in a mode that only records what the agent would have done. DeepSeek says the code came from the team's daily work, is not an official product, is not production-ready, and differs from both its internal version and the essay 'Exploring Super-agent Design: Towards a More Autonomous Agent System'.
What it does
- Takes requests from GitHub, with Feishu as an optional second source
- Turns each accepted request into a tracked Ticket worked in its own agent session and directory
- Replies in the original conversation, for example with a pull request
- Proactive work only under team-written rules, starting in a record-only mode
- Supervision panel in the DeepSeek Harness sidebar that lists what needs a person first, with Source, Trigger, Eval, Execution and Settings tabs
- Versioned, editable prompts that the panel compares with the assembled full prompt
- Optional auto-merge through a separate GitHub account allowed to merge
Getting started
Setup is designed to be done together with a coding agent, following SETUP.md and docs/006-deployment.md. You need Node.js 22.19+ or 24+, pnpm, git 2.32+, tmux, Python 3.8+, GitHub CLI 2.48+, DeepSeek Harness with a model provider, and a dedicated GitHub account for the agent invited to the target repository.
Hand the setup to a coding agent
From a checkout of the repository, give your coding agent the prompt the README suggests. It walks through the probe, logins, install and settings with you.
Read SETUP.md in this repository and set up dsh-superagent with me.Check the prerequisites
SETUP.md starts with a one-time probe of the tools and the GitHub CLI login.
node -v; pnpm -v; git --version; tmux -V; python3 --version; gh --version | head -1; dsh --version
gh api user --jq .loginCreate the superagent profile
This creates the 'superagent' DeepSeek Harness profile. Skip it if the profile already exists, because it replaces the profile and deletes its cordis.patch.yml.
dsh superagent --from-default-profileConfirm it is healthy
After saving the github source, policy, executor and prompt settings through the Management API, the status script should exit 0 with no setup problem.
scripts/dsh-superagent statusDevelop on it
One command runs the typecheck, the tests and the build.
pnpm checkCommands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Let teammates ask an agent for fixes in GitHub issues or a Feishu chat and get a pull request back
- Study a working design for a router session that admits work and a pool of worker sessions that do it
- Trial proactive agent behaviour safely by recording what the agent would have done before allowing it to act
- Build a new request source for DeepSeek Harness following the repository's cookbook
How DeepSeek Super-agent compares
DeepSeek Super-agent alongside other open-source autonomous coding agents tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| opencode | ★ 224k | OpenCode is an open source AI coding agent that runs in your terminal, with built-in build and plan agents and an optional desktop app. |
| Claude Code | ★ 154k | Anthropic's agentic coding tool for the terminal, IDE and GitHub: it understands your codebase, executes routine tasks, explains code and handles git workflows from natural-language commands. |
| OpenAI Codex | ★ 133k | OpenAI's cloud and CLI coding agent that writes features, fixes bugs, and proposes code changes across a repo, running tasks in parallel. |
| Pi | ★ 119k | Minimal terminal coding agent harness with four built-in tools, extended through TypeScript extensions, skills and prompt templates instead of forks. |
| Orca | ★ 112k | Open-source desktop workspace that runs Claude Code, Codex, OpenCode and other CLI coding agents in parallel, each in its own git worktree, with a mobile companion app. |
| Gemini CLI | ★ 109k | An open-source command-line AI agent from Google that connects your terminal to Gemini models for reading code, editing files, running shell commands, and searching the web. |
| autoresearch | ★ 99k | Karpathy's minimal harness that hands a coding agent a single-GPU LLM training script and lets it run fixed five-minute experiments, keeping or discarding each change on its own. |
| DeepSeek Super-agent | ★ 3.9k | DeepSeek's open prototype of a coding agent that lives in a repo's conversations |