/
DeepSeek releases open-source 'Super-Agent' prototype to rival Claude Code; it watches developer chats
DeepSeek's Harness team released a prototype of Super-Agent, an open-source coding agent aimed at Anthropic's Claude Code, which can monitor developer group chats and suggest fixes unprompted. It was announced on WeChat; DeepSeek warns sessions run with full access and advises a separate OS account.
A
2 min
The Chinese artificial intelligence company DeepSeek has released a prototype of Super-Agent, an open-source coding agent that it hopes will rival Anthropic's Claude Code. Super-Agent can monitor developers' group chats for feedback on coding issues and suggest solutions without being prompted. The company's Harness team announced it in a WeChat post on Saturday, 31 October, and a user manual is published on GitHub.
What the agent does
DeepSeek's stated goal for the agent is to take as little human attention as possible and give as much value as possible. Unlike a tool that waits for a developer to ask it for help, Super-Agent watches the chats in which a team discusses problems and offers a fix on its own. A report describes it as a more autonomous coding agent than the company's earlier offering.
A harness is the software layer around a large language model that helps AI agents handle complex tasks. DeepSeek formed the Harness team earlier in the year to build agents that address coding services from leading US labs, including Anthropic.
Open source and security
The code is open source, and one report says that it was published under the MIT license. DeepSeek says that sessions run with full access as the operator's operating system user, and it recommends using a separate operating system account for isolation. One article describes broader capabilities, including code scanning and API and web access, and says that DeepSeek has not disclosed whether the agent includes the kind of runtime isolation that protects a machine when generated code is run. Those claims were not corroborated by the other reports found.
An earlier report, from August, covered DeepSeek Harness, an open-source, model-agnostic alternative to the agent infrastructure underlying Claude Code and similar tools, which launched alongside the V4-Pro model on its API at higher prices. The new Super-Agent is a separate release from that one.
DeepSeek raised about 100 billion yuan, roughly $15 billion, in its latest funding round ahead of an initial public offering, according to one report.
The release adds to a crowded field of coding agents that developers can run on their own machines. The reports found describe it as a prototype and do not give any measure of how well it performs compared with Claude Code. Its main difference is that it is open source and that it acts on its own after reading team chats, which raises the question of what access it should be given.
The same week, a tampered extension of Anthropic's Claude Code was found on Cursor's marketplace, which shows the security risks that come with tools of this kind.
The reports found do not give benchmark results, the model that powers the agent, or the cost of running it. They do not say which chat platforms it can monitor, or how it decides when to speak. They do not carry a comment from Anthropic.
A
2 min
What did you think?
