DeepSeek Releases Open-Source Super-Agent to Take On Claude Code

Chinese AI lab DeepSeek has released an open-source prototype called Super-agent, a coding agent built to spot problems in a software team’s work and propose fixes without being summoned. Bloomberg News reported late Saturday, Oct. 31, 2026, that the DeepSeek Super-agent is aimed at Anthropic’s Claude Code, one of the most widely used AI coding tools. The code went up on GitHub at 5:32 a.m. EDT on Saturday under the permissive MIT license.

Key figures: released Oct. 31 under MIT license · 6,801 GitHub stars and 676 forks in about 34 hours · DeepSeek’s latest round about 100B yuan (~$15B)

What the DeepSeek Super-agent does

The release comes from DeepSeek’s Harness team, a group formed earlier this year, according to a WeChat post cited by Bloomberg via Business Standard. In AI engineering, a “harness” is the scaffolding of tools, memory and rules wrapped around a large language model. The agent can watch developer group chats and suggest fixes on its own, the report said.

The project’s GitHub manual sums up its goal in one line: “take as little human attention as possible, and give as much value as possible.” In practice, the prototype connects to a single GitHub repository and, optionally, the Feishu workplace chat app. When it picks up a task, it opens a work item, does the job in its own session and working directory, and answers with a pull request. A web panel lists the items that need a person’s decision.

Guardrails, and what it is not

DeepSeek labels the code a “research prototype for reference only, not an official DeepSeek product.” The README adds that it is not identical to the version the company uses internally or to the system described in its paper, “Exploring Super-agent Design: Towards a More Autonomous Agent System.”

The design documents set limits on that autonomy. The agent acts only when asked or when a rule written by the team allows it, and work nobody requested begins in a record-only mode. It asks one question at a time, and the guidance to teams is “Start narrow.” Messages in chat cannot change its permissions. Users should also note a trade-off: worker sessions run commands without approval under the operator’s own operating-system account. The agent’s account never turns on auto-merge itself; the default merge rule covers only pull requests under 200 changed lines that already have approvals.

Proactive agents are becoming a crowded field. Anthropic has previewed its own proactive agent, as TBR reported in our story on Claude Duet.

Data: early traction and a $15 billion war chest

Developers moved quickly. By about 2:35 p.m. EST on Sunday, the repository had 6,801 stars and 676 forks, roughly 200 stars an hour since launch, by TBR’s calculation. Forks, or copies made for independent work, equaled about one-tenth of the star count.

The launch follows a large fundraising. DeepSeek raised about 100 billion yuan, roughly $15 billion, with CATL and Tencent among the biggest investors, Mint reported, citing Bloomberg. People familiar with the deal said the target rose from 50 billion yuan to 80 billion yuan before the final tally, meaning the round doubled from its starting goal. Even so, OpenAI and Anthropic have each raised more than $100 billion, so the new round is less than a sixth of either rival’s total, by TBR’s calculation. Anthropic’s finances are laid out in its IPO filing.

The Washington backdrop

DeepSeek’s tools face limits in the United States. The Department of Homeland Security ordered federal agencies to remove DeepSeek AI from their systems last month, as covered in TBR’s report on the DHS order. Because Super-agent is open source, companies can inspect the code and run it on their own machines, though the license gives them no support or warranty. DeepSeek’s latest model is V4.1 Flash, and the company has tapped CITIC Securities to prepare for a possible 2027 listing on Shanghai’s STAR Market, Mint has reported.

Sources

• Bloomberg via Business Standard: DeepSeek drops enhanced coding agent with aim to rival Anthropic (Nov. 1, 2026)
• GitHub: deepseek-ai/dsh-superagent-prototype, README and design documents (accessed Nov. 1, 2026)
• Mint, citing Bloomberg: DeepSeek raises $15 billion to challenge Anthropic, OpenAI and other US AI labs (Oct. 22, 2026)
• Mint: DeepSeek doubles funding target to $15 billion ahead of 2027 IPO (Oct. 8, 2026)