Service · AI Studio
Data security and hosting.
AI in companies rarely fails because of the technology, but because of legitimate questions about confidentiality and control. That is why data protection is part of our architecture, not an add-on.
01
Three hosting options.
You decide where the agent team runs. All three variants work with the same principles.
EU cloud, operated by us
Cloud servers in Germany, a dedicated instance for you. We take care of operation and maintenance. The fastest route into the pilot.
EU cloud, self-operated
Cloud servers in Germany, dedicated infrastructure. You operate, your IT team monitors, we deliver team and rule set.
Hybrid on premises
AI models in the cloud, sensitive data locally on your premises. Your own infrastructure, your own IT team, connected via controlled interfaces.
Model provider of your choice
We are not tied to one provider and choose the model per task. EU regions wherever the provider offers them.
02
Six principles.
Read only
Agents read data. Writing happens only after explicit approval, for example tickets or drafts.
Separate permissions
Each agent gets exactly the rights its task requires, set per system and revocable at any time.
Humans in the loop
Critical steps pass through approval by your staff. Changes to the knowledge base and rules only after review.
Credentials in a vault
Credentials are kept separately per team in a password vault, encrypted, never in plain text and not with the agents.
Complete logs
Every agent action is logged with a timestamp. Every analysis is stored with its query.
No training with your data
Your data does not flow into model training. Excluded by contract with the providers we use.
03
Data processing and contracts.
- GDPR data processing agreement before the first data access, with a list of the sub-processors used.
- Service contract for pilot and support with clearly described scope, access levels and notice periods.
- Your data stays your data: rule set and documentation belong to you, exports possible at any time.
- Reference naming only with your explicit consent. That is why you will not find customer names on this website.
04
How the large model providers handle data.
For business customers, the three large providers contractually exclude training with customer data. Differences exist in retention and EU hosting. The details come from each provider’s documentation, last checked on 8 October 2026; the provider’s contractual terms at the time of your pilot are decisive.
| Provider | Training with your data | Retention according to provider | EU hosting |
|---|---|---|---|
| OpenAI (API, Enterprise) | no, API default since March 2023 | Abuse-monitoring logs up to 30 days, varying by endpoint; Zero Data Retention and modified abuse monitoring only after approval by OpenAI | yes, EU region with activation; data residency is not the same as zero retention |
| Anthropic (API, Enterprise) | no, excluded by contract | Inputs and outputs are deleted within 30 days; Zero Data Retention by agreement; exceptions for policy violations and individual features | via AWS Bedrock EU or Google Vertex AI EU |
| Google (Gemini via Vertex AI) | no, not without explicit permission | In-memory cache 24 hours by default, abuse-monitoring logs possible, request logging off by default; zero retention requires configuration (cache off, no storing, no logging) | yes, via Vertex AI EU regions |
Sources: OpenAI, Your data · Anthropic, data retention (as of 1 July 2026) · Google Cloud, Vertex AI zero data retention. Which settings apply to your pilot is defined in the data processing agreement before the first data access. There is no blanket promise of zero data retention; it depends on the provider, the contract and the features used. Responsible for these statements: Venista AI Studio, reviewed quarterly.
Frequently asked
Data security and hosting: frequently asked.
Does our data leave the EU?
Not if you don’t want it to. Hosting in Germany, model providers in EU regions, and with the hybrid variant sensitive data stays on your premises.
Do we need personal data in the pilot?
In most processes, no. Metrics, aggregates and internal identifiers are enough. Where individual cases are necessary, we clarify the legal basis beforehand and regulate it in the data processing agreement.
Who has access to our systems?
The agents with the permissions set, and the named people at Venista who support the team. Credentials are kept in the vault; every access is logged.
More services
Fixed-price pilot · Operation and support · How we work · All use cases
Does this sound familiar?
In a first conversation we look at your processes together and sketch out where an agent team would make the biggest difference. Free of charge and without obligation.