DeepSeek has introduced Creator Mode in its DeepSeek Harness platform, allowing users to build custom AI plugins by describing what they want in ordinary language. Instead of manually writing an extension, users can ask the AI agent to create tools, interactive workspaces and visual interfaces that become part of the application.
The November 2, 2026 announcement included examples such as a 3D model visualiser, a video-editing workbench, a design workspace and a sandbox game. The launch marks another step in DeepSeek’s effort to compete in the AI agent and coding-tool market, where companies are increasingly moving beyond chatbots towards agents that can build and operate software. However, Harness remains in public preview, and the company has not disclosed independent performance benchmarks or production-level reliability figures for Creator Mode.
Key Highlights
- DeepSeek introduced Creator Mode for its DeepSeek Harness agent platform on November 2, 2026.
- Users can describe a tool or interface in natural language and ask the AI agent to build it as a plugin.
- Demonstrations include a 3D language-model visualiser, a design workspace and a video-editing workbench.
- DeepSeek published seven example prompts to help users reproduce and customise demonstration plugins.
- DeepSeek Harness is an open-source project available under the MIT licence.
- The platform is in public preview, with its core plugins and APIs still evolving.
- Users should exercise caution because the experimental software can execute generated code and interact with local files and network resources.
What Is DeepSeek Creator Mode?
Creator Mode is a plugin-building capability inside DeepSeek Harness, an AI agent environment developed by the Chinese AI company.
Traditionally, creating a plugin requires programming knowledge. Developers need to understand the application’s architecture, write code, connect the required tools and test whether the extension works correctly.
Creator Mode attempts to simplify this process by allowing users to describe the functionality they need through a conversation.
For example, a user could request a workspace for editing videos, an interactive 3D visualisation or a custom design tool. The AI agent can then work on the plugin’s implementation, create the necessary files and integrate the extension into Harness.
The approach is designed to make software customisation more accessible without removing the underlying code-generation and development process.
DeepSeek describes Harness as an open-source, reusable and composable agent environment. Its plugin architecture allows users to extend the agent’s tools, skills and interface.
What Can Users Build With Creator Mode?
DeepSeek showcased several examples to demonstrate how Creator Mode can turn natural-language instructions into functional applications.
Some of the demonstrations go beyond simple chatbot extensions and involve interactive software interfaces.
| Example plugin | What it can do |
|---|---|
| 3D model visualiser | Generate and display interactive 3D models |
| Video-editing workbench | Provide a video preview, timeline and chat-based editing interface |
| Design workspace | Create an environment for custom visual and design tasks |
| Sandbox game | Build an interactive gaming experience |
| Transformer visualiser | Explain how a Transformer AI model processes information |
| Scientific paper reader | Display research papers with highlighting and note-taking features |
| Conversation statistics dashboard | Show usage information such as token counts and estimated costs |
The examples demonstrate how an AI agent could generate more than text or isolated code snippets. It can attempt to create an interface that users can interact with inside the application.
DeepSeek also released a collection of seven reference prompts in English and Chinese. Users can adapt these prompts to create similar tools or use them as starting points for their own projects.
The company has cautioned that some demonstrations were refined through multiple conversations. Therefore, entering a single prompt may not reproduce the exact final result shown in its demonstrations.
How Creator Mode Works
Creator Mode combines natural-language instructions with the plugin-development capabilities of DeepSeek Harness.
The general workflow involves several steps:
- Describe the idea: The user explains the tool or feature they want to create.
- Understand the requirements: The agent examines the available plugin architecture and identifies the components needed.
- Generate the implementation: The agent creates the relevant plugin files and code.
- Install the plugin: The generated extension is added through the plugin management system.
- Test and refine: The user can continue interacting with the agent to fix problems or modify the interface.
A demonstration on DeepSeek’s Harness website shows the agent building a Pomodoro timer. The agent examines the available interface providers, creates the plugin files, installs the extension and checks whether the timer appears in the installed plugins list.
This illustrates an important distinction: Creator Mode is not simply an AI tool that suggests code. It attempts to handle several stages of software development, including implementation, installation and verification.
However, the final result still needs to be tested. A generated plugin may require additional instructions, debugging or manual changes before it becomes reliable.
DeepSeek Harness Is Designed for AI Agents
Creator Mode is part of a broader effort to develop DeepSeek Harness as an environment for AI agents.
An AI agent differs from a conventional chatbot because it can use tools and perform multiple steps towards completing a task.
For example, a chatbot can explain how to create a dashboard. An agent environment can attempt to generate the files, connect data sources, build the interface and test the result.
DeepSeek Harness provides the software infrastructure that allows an AI model to interact with tools, files and an application workspace.
The platform also supports coding, document work, data analysis and research. Its plugin-based architecture allows additional capabilities to be installed or developed.
DeepSeek has been working to strengthen its agent and coding capabilities as competition increases from companies such as OpenAI and Anthropic.
The broader industry is moving towards systems that can complete longer workflows rather than responding only to individual prompts.
How Users Can Access DeepSeek Harness
DeepSeek describes Harness as an open-source project in worldwide public preview.
The project is available through its official GitHub repository under the MIT open-source licence.
Users with Node.js installed can start the web interface through the following command:
npx @deepseek-ai/dsh webThe application runs locally, with port 3080 used by default. Users can also clone the source code and build the project themselves.
The setup process involves configuring a supported DeepSeek model and API key in the application settings before starting an agent session.
Creator Mode is intended for users who want to customise the Harness environment. It is therefore different from simply opening the standard DeepSeek chatbot and asking it to generate a response.
The platform is currently aimed more towards developers and technically inclined users than casual consumers.
Security and Reliability Remain Important Concerns
Although Creator Mode lowers the barrier to building plugins, it also introduces security considerations.
DeepSeek Harness is experimental developer-preview software. Its ability to generate code and install third-party plugins means that users must understand what permissions they are granting.
The application can execute model-generated code and commands, access files and interact with network resources made available to it. A poorly written plugin or malicious extension could therefore create security problems.
Sandboxing and approval controls can reduce risks, but they do not guarantee complete isolation or prevent every possible form of damage.
Users should avoid testing untrusted plugins in directories containing sensitive files, credentials or important business data. They should also review generated code and permissions before allowing an extension to run.
There are other limitations as well. DeepSeek has not published independent success rates, generation costs, average build times or reliability benchmarks for Creator Mode.
Its demonstrations show what the system can attempt, but they do not establish how consistently it will build complex applications in everyday production environments.
The Bigger Picture
DeepSeek’s Creator Mode reflects a broader change in the AI software industry. AI systems are increasingly being developed to build tools and interfaces, rather than simply generate answers or write isolated pieces of code.
If these systems become reliable, businesses could use them to create internal dashboards, research workspaces, document tools and specialised applications without building every feature manually.
For developers, the opportunity lies in spending less time on repetitive implementation and more time on system design, testing and problem-solving. However, generated software still requires engineering oversight, particularly when it handles sensitive information or business-critical workflows.
Looking Ahead
The next stage for DeepSeek will be improving Harness’s plugin architecture, expanding compatibility and making Creator Mode more reliable. The open-source release and reference prompts give developers a way to experiment with the technology, but wider adoption will depend on how well the system performs beyond curated demonstrations.
The feature also adds to competition among AI companies building agent-based development platforms. As OpenAI, Anthropic and other developers expand their coding and agent products, the ability to create complete, usable software through natural-language instructions could become an increasingly important part of the AI market.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
