DeepSeek Harness reaches 1.5 million daily active users

Rapid growth in user adoption

DeepSeek recently shared its first set of user data for DeepSeek Harness, a framework designed to help large language models function as autonomous agents. The data shows a sharp increase in engagement over a very short period. The platform reached 1.54 million daily active users (DAUs) during the week ending November 4.

This figure represents a significant jump from previous weeks. The platform recorded 429,000 DAUs during the week ending October 7, and it had only reached 271,000 DAUs during the week ending September 30. These numbers indicate that the framework is gaining momentum quickly among developers and researchers.

The scale of activity is further evidenced by the sheer volume of data being processed. According to the report from 一财全球Yicai Global, users on the platform are consuming around 92 trillion tokens every day. This high volume suggests that the framework is being used for substantial, high-intensity computing tasks.

The role of third-party plugins

A major driver of this growth appears to be the ecosystem of third-party plugins. Since the release of DeepSeek Harness, the number of available plugins has surged to approximately 35,000. These plugins allow the AI to interact with different tools and data sources, expanding what the model can actually do in a real-world setting.

The data shows that the community is highly engaged with these tools. About 60 percent of the users on the platform have either created or installed at least one third-party plugin. This level of participation suggests that the framework is being customized by users to meet specific, diverse needs rather than being used as a generic tool.

To support this growing ecosystem, DeepSeek is currently preparing a dedicated support program for plugin authors. By providing resources to those who build these extensions, the company aims to ensure the framework remains flexible and capable of handling new types of tasks as they emerge.

From models to autonomous agents

DeepSeek Harness is described as a software support framework built around large language models. Its primary purpose is to help AI agents manage and execute complex tasks that go beyond simple text generation. The company views the framework as a complete set of operating systems that a model needs to truly become an agent.

An AI agent is different from a standard chatbot because it can perform actions independently. While a chatbot responds to a prompt, an agent can plan a sequence of steps, use tools, and complete a goal. DeepSeek Harness provides the underlying structure that allows these models to interact with the digital world in a more structured way.

By providing this infrastructure, DeepSeek is positioning itself as a leader in the evolution of domestic models. The goal is to move past simple interactions toward systems that can handle multi-tasking and team collaboration, which are currently difficult for mainstream agent designs to manage effectively.

The vision for super-agents

The company recently published a technical article on its official WeChat account exploring the design of what it calls 'super-agents.' This concept addresses a current limitation where agents are often tied to a single manual session. DeepSeek argues that for agents to be truly useful, they must be able to maintain context across different sessions.

The goal of a super-agent is to function as a continuous entity. This means the AI should be able to remember previous interactions and work on multiple complex tasks simultaneously without the tasks interfering with one another. This approach aims to overcome the human limitation of only being able to pay attention to one thing at a time.

DeepSeek believes that as model capabilities improve, personal agents will begin to emerge rapidly. However, the transition to these more autonomous systems requires breaking the bond between the agent and a one-off manual prompt. The super-agent architecture is designed to facilitate this shift toward more independent, persistent AI behavior.