Frequently asked questions
About data and privacy
Does my data go to the AI model?
You decide, per source. In the default mode metrics only, no rows go to the model, only structure, counts and statistics. In encrypted values, rows go along but every cell is a token. Only in open, on a source you designate for it, do literal values go along. See Data access.
Does the model see the names of my tables?
No. Table and column names go to the model as pseudonyms, such as t_0056. You read the answer back in
your own names. See Pseudonymisation.
Can I not just put the rules in the prompt?
You can, but a model can ignore an instruction. In Plainwater AI rules are enforced in the gateway, where the model cannot get around them. See The AI gateway.
Is everything stored?
Prompts and answers are stored encrypted. Whoever is entitled can read them back. Every read is itself recorded. See Audit.
Can a colleague look into another customer?
No. Anyone without permission for a customer sees "not found". The platform does not even reveal that the customer exists. See Roles and permissions.
About working with the platform
Can an agent break something on my database?
In the work mode look only, no: the agent only reads. To prove something it works in a disposable copy. Writing on the real source is a separate permission per source, and even then every change goes through a proposal a human approves.
Which databases and systems are supported?
SQL Server and Azure SQL, PostgreSQL and SAP HANA Cloud as databases. In addition, connectors for Power BI, Azure Data Factory, Microsoft Fabric, Databricks, dbt Cloud and SharePoint among others, and your own MCP servers. See Sources and connectors.
Do I have to set up every colleague separately?
No. You set up the customer, the sources, the rules and the knowledge once. Everyone who works for that customer works with them. See Getting started.
Does the platform also build dashboards?
Yes. An analysis can end in a dashboard that keeps running on the source and refreshes on a schedule, without needing an AI model for every refresh. See Dashboards.
About models and tools
Which AI models does it work with?
Claude, GPT, Gemini and Mistral are set up, and any model with an OpenAI-compatible API is a matter of configuration. Per customer you choose which models are allowed. See Choosing models.
Can I use my own AI tool?
Yes. Any AI tool that speaks MCP, such as Claude Code, Codex, GitHub Copilot or Cursor, can connect to the platform. The customer has to allow it; by default it is off. See Connecting via MCP.
Does the platform incur model costs when I use my own AI tool?
In the Solo way of working, no: all work then runs on your own subscription and the platform only supplies tools, rules and storage.
About hosting
Where does the platform run?
Today in Azure West Europe, in the Netherlands. We are moving to a European cloud provider. See European processing.
Can I enforce that only European models are used?
Yes. Because you choose per customer which models are allowed, you can limit the list to models with European processing, such as Mistral or Azure OpenAI with EU data residency.
Ask us via Request a demo at the bottom of this page.