Default: our own servers
Open models answer every request in our data center in Germany.
Open models on our own servers in Germany, GDPR-compliant. You add frontier models only when you need them.
Open models · Our own servers · Germany
By default, open models answer every request on our own servers in Germany. Frontier models, meaning the models of the big AI providers, only come in when you explicitly switch them on.
Open models answer every request in our data center in Germany.
Off until you switch frontier models on. Then processed in the EU.
As long as frontier models are off, only our own models answer.
You don't have to pick a model.
From multi-LLM chat and tool-enabled agents to the workflow builder — everything your team needs.
Data sovereignty
By default, our own models answer every request in our data center in Germany. If you switch on a frontier model, processing stays in the EU. Contractually secured, technically enforced.
Trust is architecture
Data processing agreement, data minimisation, right to deletion: contractually anchored, not bolted on.
Our own data center in Germany.
Our own open models. No vendor lock-in.
Data processing
DPA included
ISO 27001
Certification in progress
Microsoft Partner
Official status
Your agents read and write in the systems your team already uses.
Latest

AI in Business
Our users do around 97 percent of their work on Kasimir with open models. And stronger ones arrive every month. What has happened since the summer, how we test new models and where the licenses hide traps.
14 October 2026 · 7 min read

Basics & Guides
A search finds candidates, a reranker selects. What the second stage in RAG systems really delivers, what it does not, and how Kasimir runs it on its own infrastructure in Germany.
13 September 2026 · 5 min read

AI in Business
Company knowledge lives scattered — across cloud folders, document stores and notes. Kasimir now unifies it in one searchable index, answering questions with evidence and page numbers. Inside our biggest knowledge update since launch.
23 August 2026 · 3 min read
Get started with Kasimir. Sign up — we'll activate your access and reach out shortly.




We get back to you personally.
No call center: your contact person knows Kasimir.
Kasimir is a GDPR-compliant AI platform for German mid-market companies. Your team chats with the leading AI models, analyses documents, transcribes meetings, runs multi-step workflows and collaborates in shared projects — all through a single secure interface hosted in Germany.
By default, every request runs on leading open-source models that we self-host in a German data center — your data never leaves the infrastructure. When a task needs a more powerful frontier model, it is processed GDPR-compliant in the EU — same contractual safeguards, no training reuse.
Data does not leave the EU. We have data-processing agreements with every model provider, including training opt-outs. Your inputs are not used to train models.
Microsoft 365 (Outlook mail & calendar, OneDrive, SharePoint), Google Workspace (Drive, Calendar, Sheets, Docs, Slides, Tasks), Jira, Confluence, Microsoft Teams and SQL databases are already connected — your agents work right inside these systems. You can also sync folders from Microsoft 365 or Google Drive as a knowledge base.
Pilot contracts are currently agreed individually. Detailed pricing will be published with the public launch.
Public AI services store data in the US and use inputs for training by default. For German companies with GDPR requirements that's not enough. Kasimir runs on self-hosted open-source models in our German data center by default — and when you genuinely need a frontier model, it is processed GDPR-compliant in the EU, contractually secured without training reuse.
If your company enables memory, Kasimir remembers durable facts from your own chat messages – such as your role, how you like to work or ongoing projects – and uses them in later conversations. Everything stored is visible in your notebook, where you can edit or delete it, and you can switch memory off at any time. Content from documents, attachments or the web is never stored.