AI for your data, without your data walking out the door.
Plainwater AI is the workspace where your team works on data with any AI model: answering questions, writing and proving SQL, building dashboards. Meanwhile the platform decides what the model gets to see.
Built by Plainwater, the makers of Yres. Runs in Europe.
What does the model actually see?
The same question, two sides. On the left what you see, on the right what goes to the model. Pick a mode and watch what happens to the result.
Data access for this source
The model gets metadata, counts and statistics. Not a single row. Counts below ten become "<10".
What you see
SELECT Region, COUNT(*) AS Customers, SUM(Revenue) AS Revenue FROM Sales JOIN Customers ON … GROUP BY Region
Result
| Region | Customers | Revenue |
|---|---|---|
| North | 412 | €1,204,000 |
| East | 288 | €873,500 |
| South | 7 | €41,200 |
| West | 356 | €990,100 |
What the model sees
SELECT c_0201, COUNT(*) AS c_0207, SUM(c_0204) AS c_0204 FROM t_0031 JOIN t_0042 ON … GROUP BY c_0201
Result
Rows held back at the gateway
- 4 rows
- c_0201: 4 distinct values, 0 empty
- c_0207: sum 1,063, smallest group <10
- c_0204: decimal(12,2), 0 empty
The strictest rule wins: a source can never be more open than the customer.
Nine reasons, one platform
Plainwater AI is not a chatbot on your database. It is a workspace with a gateway in front, built for teams who work with data.
Control
You decide what the AI sees, and you can prove it afterwards.
Work
From question to proven change to a dashboard that keeps running.
Choice
Set it up once, and everyone works the way they like.
From question to proven answer
An agent reads metadata first: objects, dependencies, quality and performance. It does not need rows for that.
A proposal is not promised but measured, on a disposable copy of the database. Before and after, with a verdict. Only then does a person approve.
Work the way you already work
The rules live in the gateway, not in the tool. That is why it does not matter where the model runs.
Fully online
Conversations, tasks, dashboards and monitors in the web app. The agents run in the platform.
With your coding AI
Connect Claude Code, Codex or another MCP client. Your AI directs, the platform can take over heavy work.
MCP only
Everything in your own AI tool, on your own subscription. The platform provides tools, rules and storage, and no model costs.
Any model, chosen per customer
A fixed route per model, a list of what is allowed per customer. A care provider picks EU processing only, another customer the newest model.
- Claude
- GPT
- Gemini
- Mistral
- OpenAI-compatible
And from your own tool
Claude Code, Codex, GitHub Copilot, Cursor and other MCP clients connect to the same platform with OAuth.
European, and honest about what that means
Today
- Platform and data in Azure West Europe (the Netherlands).
- EU models available: Mistral, and Azure OpenAI with EU data residency.
- Enforceable per customer: only models with EU processing.
Where we are heading
- Moving to a European cloud with a European owner, such as StackIT or Hetzner.
- As many models as possible that also run in Europe.
Working with Yres?
Plainwater AI works together with Yres through the Yres MCP server. An agent that knows the rules of Yres: load types, history and changes. Nothing bypasses the change process.
See it on your own data
In 30 minutes we show the platform on a sample environment and discuss what it would see in yours. After that you know whether it fits.