(Yicai) Nov. 6 -- Chinese artificial intelligence giant DeepSeek has released user data for DeepSeek Harness for the first time, revealing that the open-source agent harness framework has more than 1.5 million daily active users.

DeepSeek Harness had 1.54 million DAUs on its official application programming interface over the week ended Nov. 4, up from 429,000 over the week ended Oct. 7 and 271,000 over the one ended Sept. 30, Hangzhou-based DeepSeek said yesterday. Users consume around 92 trillion tokens a day.

The number of third-party plug-ins has surged to about 35,000 since DeepSeek Harness was released, and about 60 percent of users created or installed at least one third-party plugin. It is preparing a support program for plugin authors.

DeepSeek Harness is a software support framework built around large language models, designed to help AI agents manage and execute more complex tasks. It can be seen as a complete set of operating systems an LLM needs to truly become an agent.

As a benchmark for domestic LLMs, DeepSeek is accelerating evolution. On Oct. 31, DeepSeek Harness’ team published a technical article called ‘Exploring Super-Agent Design to a More Autonomous Agent System’ on its official WeChat account.

The article claims that improved model capabilities have given rise to new applications, and personal agents have begun to emerge rapidly. However, mainstream agent design still cannot support this mode at the multi-task and team-collaboration level, and the multi-task limit depends on people’s attention.

To overcome these limitations, it is necessary to break the binding between agents and a one-off manual session, making it a continuous entity that can maintain context across sessions and simultaneously advance multiple complex tasks without interfering with each other. DeepSeek refers to this form of agent and its system architecture as ‘super-agent.’

Editor: Futura Costaglione