DeepSeek has released an open-source coding agent prototype that does not wait to be asked. Its Harness team published Super-agent on GitHub on October 31, 2026, under the MIT licence, and Bloomberg reported it on November 1. The agent is built to watch developer group chats, spot coding problems, and propose fixes before anyone summons it. The target is plain: Anthropic's Claude Code.
On paper, the design is simple. A Router takes permitted requests from GitHub, and optionally from the Feishu workplace chat app, and hands each one to an isolated Worker session. Each Worker tackles one issue in its own directory, then returns the result to the original conversation, for example as a pull request. Because Workers run separately, several issues can move in parallel.
The correction that matters: this is research code
DeepSeek's own framing is narrower than the headlines. Super-agent is research and reference code, not an official company product, and not production-ready. Some ideas in the accompanying design write-up are not even implemented in the published code. So the release reads better as an architecture demonstration than as a tool you hand your repository to and walk away from.
The framework underneath, DeepSeek Harness, came first as open source on August 13, 2026, also MIT-licensed and self-hostable. It installs through Node.js (npx @deepseek-ai/dsh web) with a local interface at 127.0.0.1:3080. Reports put the prototype repository at roughly 4,117 stars by the morning of November 1 UTC. We are writing that as a reported figure, not a live count of our own.
Before you try it: the pilot conditions
There are warnings that cannot be skipped. Worker sessions run commands without approval, as the operating-system user running Harness. Harness is still experimental preview software with no security audit, and one critical sandbox-escape flaw, CVE-2026-82533, was only fixed in version 0.1.2-alpha.1. A sane pilot means: a disposable repository with no secrets, the latest Harness release only, minimal privileges, the control port kept on localhost, and human review on every pull request.
Why is this release getting attention? The most capable coding agents today are usually paid services with closed internals. Super-agent inverts that bargain: code you can read, change, and run yourself, with no licence fee, but the safety engineering moves to the team using it. For a team with a sandboxing plan, the prototype is cheap study material. For a team that needs a day-one tool with vendor support, a paid agent still makes more sense until open prototypes like this mature.
The timing explains why DeepSeek chose now. Reports say the company raised about 100 billion yuan, roughly $15 billion, in its latest funding round ahead of an IPO. On the other side, Anthropic, the owner of Claude Code, is expected to start its IPO roadshow later this month. The coding-agent race is heating up, and DeepSeek just put its most open offer on the table. We use the term Super Intelligence on this site, explained on our About page. The other Super Intelligence story this week that is likewise gated to a limited circle, Gemini 4 Carbon, is in our October 31 update.
Sources: Bloomberg via DigitalToday (Nov 1, 2026), Coheser (Nov 1, 2026), TradingView/Stocktwits (Nov 1, 2026), DeepSeek's user manual on GitHub. Figures are snapshots at writing time, evening November 1, 2026 WIB, and can change.
Writing about crypto and Super Intelligence: market news, new projects, Super Intelligence models, and early opportunities like airdrops and testnets. Every piece starts from the official announcement, with the costs worked out up front.