DeepSeek has released a free, open-source coding agent called Super-Agent, pushing the Chinese artificial intelligence lab into the market for AI software-development tools less than two weeks after Bloomberg reported a funding round of about $15 billion ahead of a planned initial public offering.

The Hangzhou-based company published the prototype, formally named dsh-superagent, on GitHub under the permissive MIT licence over the weekend, according to the South China Morning Post. The GitHub page links the code to a DeepSeek article titled "Exploring Super-agent Design: Towards a More Autonomous Agent System," while noting that some designs in that article are not implemented in the prototype. The tool competes for the same developers as Anthropic's Claude Code and OpenAI's Codex, in what Digital Trends called a rapidly growing market for AI software development. For investors in Asia, the release shows where DeepSeek may look for revenue once it is a listed company: not only cheap models, but the agent software that sits on top of them.

What DeepSeek's Super-Agent does

Most coding assistants wait for a developer to type a request. DeepSeek's prototype is built to sit inside a team's workflow instead. According to the project's documentation on GitHub, it listens to a single GitHub repository and, if configured, to the Feishu workplace chat app. When a permitted user asks for work, the agent opens a work item, carries it out in its own session and replies in the original conversation, for example with a pull request. A web panel shows the items that need a person.

The design article proposes ways to cut the human effort needed to coordinate several AI agents and track their progress, the Post reported. DeepSeek's GitHub page sums up the aim as using as little human attention as possible while delivering as much value as possible.

The company has built in limits. Help Net Security, which reviewed the code, reported that the agent acts only when an approved user asks or when a rule written by the team covers the task. In the default mode, indirect requests, such as a message not addressed to the agent or a comment from a bot, are assessed and logged as ones the agent would have taken on, but none is accepted. The agent also does not switch on automatic merging under its own account by default, so people still merge code unless a team chooses otherwise.

DeepSeek is careful about what the release is. The GitHub page calls it a research prototype "for reference only, not an official DeepSeek product," says much of its engineering was done by coding agents, and warns that it may contain bugs. It is packaged for DeepSeek Harness, the company's agent platform. Digital Trends reported that DeepSeek set up a Harness team earlier this year to build agents that can match the coding services of the leading U.S. labs.

DeepSeek funding, IPO plans and U.S. restrictions

The launch comes in a busy month for the company. Bloomberg reported on Oct. 22 on a DeepSeek funding round of about 100 billion yuan, or roughly $15 billion, one of the biggest in China, according to TechCrunch's summary of that report. DeepSeek has not announced that the round has closed. Battery maker CATL and Tencent were named as potential investors. Digital Trends, also citing Bloomberg, said the round comes ahead of a planned IPO.

That report came two days after Washington tightened its stance. On Oct. 20 the U.S. Department of Homeland Security published an order in the Federal Register that excludes products and services developed by Hangzhou DeepSeek Artificial Intelligence Co. from federal civilian procurement. It also requires agencies to remove DeepSeek products, and third-party products built using DeepSeek technology, from their systems.

The order narrows one market for DeepSeek's tools but leaves the much larger private developer market in the U.S. and elsewhere open. An MIT licence lets any company download, modify and run the code on its own servers, which gives the prototype a route to adoption that does not depend on DeepSeek's own cloud service.

Why investors are watching Chinese AI coding tools

DeepSeek made its name with low-cost open models. Its news page shows the company released a preview of its DeepSeek-V4 model in April with a context window of one million tokens, and followed with a smaller, faster DeepSeek-V4.1-Flash in September. Open models, though, are hard to charge for. Coding agents offer a clearer path to revenue, because companies pay for tools that do work across whole engineering teams.

The release also adds to a crowded field of Chinese labs seeking capital and public listings. Rival Moonshot AI, the maker of the Kimi chatbot, is weighing a Hong Kong IPO of up to $5 billion, and AI already takes the largest share of startup funding across Asia. A credible paid product in coding would give DeepSeek a revenue story that public-market investors can measure, rather than a reputation built on benchmark scores.

For now, Super-Agent is free and unsupported, and DeepSeek has not said whether a commercial version will follow. The company has given no date for its IPO or the exchange it would use. The next concrete signal will be whether DeepSeek turns the prototype into a supported Harness product with published pricing.