Privacy & EU Hosting

How Kasimir delivers GDPR compliance, EU hosting and data sovereignty — from self-hosted models to the non-EU gate and cookie consent.

Kasimir was built from the ground up for the German mid-market: GDPR-compliant and with clear data sovereignty – by default, all AI models run on Kasimir’s own servers in Germany or in the EU. This page explains where your data is processed, how to tell which AI model writes data where, how administrators control access, and which legal building blocks (DPA, cookie consent, privacy policy) are wired into the product.

The sovereignty principle in brief

Kasimir assigns every AI model to exactly one of three "sovereignty tiers". This classification governs whether data leaves the EU — and it is surfaced everywhere in the product.

Tier	Meaning	Where processed	Data leaves the EU?
Self-hosted	Runs in Kasimir’s own operation – the most sovereign option, default	Germany	No
EU cloud	EU region, but via a processor	EU (data centres of external providers, e.g. Microsoft Azure)	No
Non-EU	Processing outside the EU – only enabled by opt-in	Outside the EU	Yes

The default is maximum sovereignty. The pre-selected chat model (Gemma) is self-hosted and runs exclusively on the EU infrastructure in Germany. Non-EU models are technically disabled until an administrator explicitly enables them.

Where your data lives

The core of Kasimir is self-operated in a data centre in Germany. It holds and runs:

your data – accounts, chats, files, chat attachments and audio recordings
the default language model
the preparation of your documents for search and citations
the transcription of speech and meetings
web search via Kasimir’s own search instance – the search terms are passed on to external search engines (including Bing and Google), but without your IP address or account. Your company can turn web search off, for example with the source policy “Company knowledge only”.

In addition, Kasimir uses external processors, some of them based in the US (covered by standard contractual clauses): Cloudflare (delivery and protection of the website and app – all connections to Kasimir pass through it), SendGrid/Twilio (platform emails) and – only with consent – Google Analytics. The full list with legal bases is in the privacy policy.

All connections to Kasimir are encrypted (recognisable by the padlock and https:// in the address bar).

Spotting the sovereignty tier in chat

You don't need any provider knowledge — Kasimir shows the data-processing location right on the model.

In the model picker

When you open the model picker in the chat composer, every model carries a badge:

A green Own servers badge for models running on Kasimir’s own servers in Germany
A grey EU cloud badge for models running in the EU with an external provider
An amber Non-EU badge for models that process data outside the EU – they only appear if your company has enabled them

Hovering a model opens a detail panel with provider, context window, vision support and a Data processing section. It states in plain language where the data ends up:

Processed on Kasimir's own servers in the EU (shield icon, green)
Processed in the EU · Azure AI Foundry (Microsoft) (EU globe icon, blue)
Processed outside the EU – data leaves the EU (warning triangle, amber)

The model picker with sovereignty badges and the data-processing detail.

In the composer hint

If you've selected a model that isn't self-hosted, the composer also shows a hint:

For EU-cloud models: Processed in the EU via Azure Foundry (Microsoft)
For non-EU models (warning tone): Note: processed outside the EU – data leaves the EU

For the default (self-hosted) model no hint appears by design — everything is sovereign here.

Non-EU models: per-company opt-in

Models that process data outside the EU are hidden entirely by default in Kasimir and must be enabled per company. Currently, every model Kasimir offers runs on Kasimir’s own servers or in the EU cloud.

How to enable them (administrators only)

Open Administration → Chat settings.
Tick the Allow non-EU models checkbox.
Save.

The hint text below the switch makes the consequence unambiguous: “Chat data then leaves the EU — not GDPR-sovereign. Off by default.”

The non-EU models opt-in toggle in Chat settings.

While this option is off, non-EU models neither appear in the picker nor can they be used via the API.

What enabling changes

With Allow non-EU models on, the enabled models appear in the picker for all employees — each carrying the Non-EU badge and the composer warning hint. The individual user then makes a conscious choice to use such a model for a given conversation. Self-hosted and EU-cloud models remain available at all times and stay the default.

Audit log and medical documents

Under Administration → Security, owners and administrators control two more things:

Audit log: shows who triggered which administrative action or deletion and when – never content, so no messages, files or inputs. Retention can be set between 30 and 3650 days (default 730); you can export the log as CSV or JSON.
Medical documents: Kasimir only processes health data (HL7 FHIR/CDA) once you turn on the Process medical documents (HL7 FHIR/CDA) switch. Only do so if your DPA includes the health-data annex.

Data Processing Agreement (DPA)

For production use with real personal data you enter into a Data Processing Agreement (DPA) with Kasimir under Art. 28 GDPR. It is built right into the product.

Open Administration → DPA.
The agreement is pre-filled with your company details and fully readable.
Use Download as HTML to keep a copy for your records.
To execute it, tick the confirmation statement and click Accept DPA now.

After acceptance the page shows status Signed with the name, email, timestamp and version number of the signing person.

Only the owner of the company account can accept the DPA — all other administrators can read the agreement but, instead of the button, see a note that only the owner may sign. Until acceptance, the page states that signing is required before production use with real personal data.

Kasimir sets only the bare minimum without consent and loads everything else only after your active opt-in — in line with the TDDDG (formerly TTDSG) and GDPR.

Category	Contents	Consent required?
`Necessary`	Login session, language preference, dark mode	No (always on)
`Statistics`	Google Analytics — pseudonymous reach measurement	Yes
`Marketing`	no tools active at this time	Yes

On your first visit the banner We respect your privacy appears with three equally prominent options: Accept all, Reject and Settings. Settings opens the per-category toggles — and the non-necessary ones are not pre-ticked.

The cookie banner with equally prominent accept and reject.

Google Analytics is only ever loaded after you consent — before that no data is sent to Google and no _ga cookie is set. If you later withdraw consent via Cookie settings, the analytics cookies are deleted immediately.

Your decision is stored in a first-party cookie (kasimir_consent) and lasts about six months. You can reopen and change it at any time via the Cookie settings link (e.g. in the footer).

Your rights and further information

The full privacy policy at /privacy lists all processing activities, legal bases, processors, retention periods and your data-subject rights (access, rectification, erasure, portability, objection, complaint) under the GDPR. You can also delete your chat content and generated texts yourself at any time. For privacy requests contact support@kasimir.ai; the controller is Viedev GmbH.

BackAccount & Profile NextRoles & permissions