The starting point: yes to AI, no to losing data
Municipalities, care providers and mid-sized companies face a dilemma: on the one hand, pressure is growing to introduce AI-supported processes. On the other, the GDPR rules out the uncritical use of cloud-based AI services that process data on US servers.
“A public authority processing citizens' data in a US data centre is not merely negligent – it is acting unlawfully.”
Data protection commissioner, Schleswig-Holstein, 2024
This is exactly where KOSMO comes in: as an AI assistant that can be run entirely locally, KOSMO combines the capability of modern language models with full control over infrastructure and data.
What on-premise really means
The AI runs on your hardware. No API calls to OpenAI. No training data passed to third parties. No vendor lock-in.
- Data never leaves your own infrastructure
- No training on your content
- Full control over model updates and versions
- Operation in German data centres possible
- Open-source basis – no black-box model
RAG instead of hallucination
KOSMO uses retrieval-augmented generation (RAG) – the language model does not answer from memory but looks things up in your documents. The result: precise answers you can evidence.
With every answer, KOSMO shows which passage of which document it came from. No guessing, no inventing.
