DeepSeek has released Super-agent, an open-source coding agent that monitors developers' group conversations, spots coding problems, and proposes solutions without waiting to be asked. The company's Harness team announced the prototype in a WeChat post on Saturday, October 31, positioning it as a rival to Anthropic's Claude Code and other US programming agents.
The code is published on GitHub under the MIT license as dsh-superagent-prototype, and the repository passed 4,000 stars within a day of going public. DeepSeek labels it a research prototype rather than an official product, and says the public release differs from the version its own team uses internally.
The pitch is a shift from prompt-and-response to participation. Conventional coding assistants wait for an individual instruction; Super-agent is built to sit inside a team's workflow, gather views on a problem from the conversation, and step in. DeepSeek's own stated design goal is to take as little human attention as possible while delivering as much value as possible.
Key facts
- 01DeepSeek's Harness team announced Super-agent, an open-source coding agent prototype, in a WeChat post on October 31.
- 02The code is MIT-licensed on GitHub and passed 4,000 stars within a day of going public.
- 03Super-agent watches one GitHub repository and, optionally, Feishu group chats, and proposes fixes without being summoned.
- 04Its default observe mode only logs what it would accept; auto mode acts only when a team-written rule matches.
- 05DeepSeek has reportedly raised about 100 billion yuan ($15 billion) ahead of a planned IPO.
In practice, Super-agent attaches to one GitHub repository and can optionally listen to Feishu, the workplace chat app also sold as Lark. It runs on DeepSeek Harness, or dsh, an early-preview, plugin-based harness. Source plugins turn incoming events into triggers, and a persistent router session decides whether to accept, attach, stop, or ignore each one.
Accepted work becomes a ticket, and each ticket gets its own worker session and working directory from a pool capped by a maxWorkers setting. A watchdog process detects failed or stalled loops. All state lives in a single SQLite file, replies go out through a durable outbox, and a local web panel shows what the agent is doing.
DeepSeek ships the agent cautious by default. In the default observe mode, Super-agent records the work it would have taken on but accepts nothing. An off mode disables inferred work entirely, and an auto mode lets the agent accept jobs only when a team-written rule matches. Direct requests from people on an allowed list work in every mode, chat messages cannot change those permissions, and auto-merge is off by default.
Setup requires a dedicated GitHub user account, Node.js 22.19 or later, GitHub CLI 2.48 or later, and a configured model provider. The prototype supports github.com only, not GitHub Enterprise Server.
The deployment guide is direct about one tradeoff: worker and router sessions run with full access as the operator's operating-system user. Setup strips environment variables whose names contain KEY, PASSWORD, SECRET, or TOKEN, the panel listens only on loopback, and a blockedCommands list can fence off personal command-line tools. For real isolation, DeepSeek recommends running the agent under a dedicated OS account that holds only the agent's credentials.
DeepSeek formed the Harness team earlier in 2026 to build agents that compete with coding services from US labs, including Anthropic. A harness is the software layer wrapped around a large language model that lets it coordinate tasks, call external tools, and run multi-step workflows, and Anthropic built Claude Code on the same kind of infrastructure.
The release lands as DeepSeek reportedly raised about 100 billion yuan, roughly $15 billion, in one of China's largest funding rounds, ahead of a planned initial public offering.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
