DeepSeek and open-source AI for business in Germany

What the DeepSeek decision is about

DeepSeek is an AI company from China, and your company can use its models in four ways. Each way sends your text to a different place, so the choice decides who can read your prompts and documents.

The app and the API send every request to DeepSeek. The open weights of a model such as DeepSeek-V4.1-Flash run where you install them: at a hosting provider under your contract or on your own server. The model family stays the same. The path of your data changes.

Four ways to use DeepSeek, checked in September 2026

Way	Where your text goes	What your tools change	Where it fits
DeepSeek app or chat website	To DeepSeek. Its privacy policy says it processes and stores personal data in the People’s Republic of China.	Nothing. Staff type into the app.	Tests with public text, if your company allows the app at all.
DeepSeek API	To DeepSeek’s servers at api.deepseek.com.	The address and the key. The API accepts the OpenAI and the Anthropic format.	Comparisons with test data that holds no personal data and no trade secrets.
Open weights at a hosting provider in Germany	To the provider’s servers, under a contract your company signs.	The address and the key. vLLM and SGLang serve the weights in the OpenAI format.	Tasks with internal data, once your data protection officer has approved the provider.
Open weights on your own servers	Nowhere outside your company.	The address and the key, as above.	Data that must not leave the company, such as customer files or contracts.

Sources: DeepSeek privacy policy, DeepSeek API documentation, DeepSeek-V4.1-Flash. Licenses and privacy policies change, so I read both again at the start of each project.

Open-source AI as a second supplier

Most companies in Germany buy their AI from a few US vendors. In a Bitkom survey published on September 9, 2026, 76 percent of the companies that use AI named ChatGPT, 35 percent Microsoft Copilot and 28 percent Gemini. Bitkom President Ralf Wintergerst said that German companies rely almost entirely on AI applications from the USA (Bitkom).

An open-weight model gives you a second supplier. Your applications keep one API format, and the model behind the address can change: a cloud model for the hard tasks, DeepSeek or another open-weight model where the test shows it is good enough. If a vendor raises its prices or changes its terms, you change the address and keep the application.

DeepSeek Harness, a developer preview

DeepSeek also publishes DeepSeek Harness, software that runs AI agents. It is under the MIT license and runs on Node.js, and each of its parts is a plugin. InfoQ reported on August 20, 2026, that it can switch between remote API providers and local model servers (GitHub, InfoQ). The GitHub page calls it a developer preview and warns of changes that break compatibility. Until a stable release, the harness belongs in a test environment.

What I set up for your company

Most of the work happens around the model:

• a list of your tasks with the data each one touches, sorted by what may leave the company,
• a test set of your own tasks, run against the cloud model you use today and against DeepSeek,
• one API address per application, so a model can change without new code,
• agents that follow written rules and reach your systems with the rights of a new colleague.

Who I am

I am Michael Wutzke, an IT expert for data platforms in Frankfurt. I have worked more than twenty years from code to CIO in finance, e-commerce and real estate, with cloud platforms and data centers. Today AI agents speed up how I build and transform data platforms.

In client work I rebuild a platform with Claude and Codex under one rule file. The rules and the memory of the agents stay outside the model. An open-weight model such as DeepSeek can take a cloud model’s place in that setup. I have run blockchain nodes in several data centers.