Self-hosted open-source models
Open models such as Llama or Mistral run on servers in Germany or in your own infrastructure. Your data stays within your control, and you're not tied to any single vendor's pricing or terms of use.
Private AI
We set up AI so your data stays protected: EU hosting, self-hosted open-source models and clear rules for access and use.
Private AI means you know where your data is processed, who can access it and under which contract. Codefellas runs open-source language models such as Llama or Mistral in German data centers or on your own premises, and connects commercial models through EU regions on AWS Bedrock, Azure or Google Vertex AI. Add anonymization, roles, logging and an AI policy for your team, and you have AI you can actually sign off on.
What we do for you
Open models such as Llama or Mistral run on servers in Germany or in your own infrastructure. Your data stays within your control, and you're not tied to any single vendor's pricing or terms of use.
When a commercial model is the best fit, we connect it through cloud platforms like AWS Bedrock, Microsoft Azure or Google Vertex AI and choose EU regions where available. Codefellas is independent and picks the model based on the use case.
We document which data goes where, which data processing agreements are needed and which providers are involved. That gives you a solid basis for discussions with your data protection officer.
Personal data such as names, addresses or customer numbers can be detected and replaced before processing. That way you can use AI on sensitive text without that data ever reaching the model.
Roles and permissions define who may use which AI features and data; logs make usage traceable. An AI policy and training give your team clear guidance and support AI literacy under Article 4 of the EU AI Act.
In practice
A law firm or consultancy wants to use AI for drafting and research but can't send client data to arbitrary services. A self-hosted open-source model in a German data center processes the documents without them going to external AI providers. The team works with AI, and confidentiality is preserved.
Employees use personal AI accounts because there's no official option. We provide an internal AI chat with company sign-in, vetted models and clear rules, backed by an AI policy. Usage becomes visible and manageable instead of banned but happening anyway.
A company wants to summarize and categorize tickets with AI, but the tickets include names and contact details. An anonymization step replaces personal data before the model sees the text and restores it in the result. Support benefits from AI without exposing customer data unnecessarily.
A manufacturer can't move engineering and process data into a public cloud. A language model runs on its own hardware at the plant and answers questions about documentation and machine manuals. The knowledge becomes usable while the data stays on the internal network.
Control over your data
Many companies hesitate with AI because it's unclear what happens to their data. That's a fair question, but not a reason to skip AI: today there are solid ways to run models in the EU or entirely on your own infrastructure.
We work with you to choose the right architecture for your protection needs, from EU cloud models to a server in your own building, and document everything so you can review it with your data protection officer.
Contact usHow we work
We get to know you, your business and your goals – free of charge and with no obligation.
We review the status quo, define goals and KPIs and give you a concrete proposal.
Design, development, campaigns: we work in agile iterations and keep you updated at short intervals.
After go-live we keep measuring, optimizing and supporting you – for as long as you need us.
FAQ
Yes, AI can be used in line with the GDPR once the legal basis, data flows, data processing agreements and technical safeguards are in place. What matters most is the choice of provider, the processing location and which data reaches a model in the first place. Codefellas plans and documents the technical setup accordingly; the legal assessment is done with your data protection officer.
A self-hosted LLM is a language model that runs on your own servers or in a rented German data center rather than with a large cloud provider. It makes sense for highly confidential data, strict industry rules or when you want to avoid depending on a single vendor. Codefellas runs open-source models such as Llama or Mistral and sets up access and updates.
For many business tasks, such as summarizing, classifying, extracting data or answering questions about your own documents, open-source models deliver very good results. For especially demanding tasks, leading commercial models often still have an edge. Codefellas tests both options on your real tasks and recommends the model that best balances quality, data protection and operating effort.
Since 2 February 2025, Article 4 of the EU AI Act requires companies to ensure a sufficient level of AI literacy among staff who use or operate AI systems. Codefellas supports this with training and an AI policy that sets out which tools may be used for which purposes. Whether further obligations apply in your case should be checked by a legal expert.
Bans rarely work as long as there's no good alternative. Codefellas sets up an official AI solution with company sign-in, vetted models and suitable hosting, and adds a clear AI policy plus a short training session. Employees get a tool they actually want to use, and you keep control over data and access.
Goes well with
With our AI consulting in Hamburg, we find out where AI genuinely saves your company time and bring the first applications safely into everyday work.
Learn more →We train your team to use AI safely and effectively, hands-on, with your own tasks and in line with the AI Act's literacy requirement.
Learn more →We make your internal knowledge findable with AI: a company assistant answers questions from your documents, with sources and the right access permissions.
Learn more →Book a free, no-obligation intro call – together we'll figure out what we can do for you.