Skip to main content

Getting started

You set up Plainwater AI once. After that, everyone in your team works the same way, with the same rules and the same knowledge. The onboarding in the web app takes you through five steps. Every step has a way out: you can stop and continue later.

Before you start​

You need:

  • an account on the platform (sign-in is with Microsoft Entra ID);
  • the administrator or consultant role for the customer you are setting up;
  • the details of one data source, for example an Azure SQL database or SQL Server: server name, database name and a login. A read-only login is enough to start.
Start small

A read-only login with permission to read metadata and statistics is enough to get to know the platform. You add write permissions later, per source, and only when you really need them.

Step 1 — The customer​

A customer is the space where everything comes together: sources, rules, knowledge, conversations and permissions. That can be an external client, but also a department within your own organisation. What belongs to one customer is invisible to another. Anyone who tries to open another customer without permission sees "not found", not "no access". The platform does not even reveal that the customer exists.

You give the customer a name and choose who has access.

Step 2 — The data source​

You add a source: the kind (for example SQL Server, Azure SQL, PostgreSQL or SAP HANA Cloud), the environment (production, acceptance, test or development) and the connection details.

The password is encrypted immediately, with its own key per source. After that you never see it again: the platform only shows a fingerprint and the moment the secret was set. The agents that later work for you never see the password either. The platform makes the connection itself.

Step 3 — The connection test​

The platform tries the connection and tells you what works and what does not: is the server reachable, is the login correct, which permissions does the login have.

Step 4 — Data access​

Here you decide what an AI model gets to see from this source. There are three modes:

ModeWhat the AI sees
Metrics onlyStructure, counts, statistics. No rows.
Encrypted valuesRows, but every cell is a stable token. The AI can count and join without knowing the value.
OpenLiteral values. Only for sources you explicitly designate here.

The mode of the source is a ceiling. If the customer is stricter, the customer wins. A source can therefore never be more permissive than the customer. More in Data access.

In this step you can also set exceptions per table or column, for example: always hide name columns. The platform makes a suggestion based on the column names. You tick what you accept; nothing is applied without your approval.

Step 5 — Ground rules​

The ground rules are the agreements that apply to all tasks for this customer:

  • which AI models may be used;
  • the budget;
  • whether people may work with their own AI tool (off by default);
  • which sources may be open;
  • guidelines against which work is reviewed afterwards.

Every change creates a new version. Old versions are kept, so you can always see which rules applied when a task started. See also Ground rules.

Your first task​

After onboarding you start a conversation. A few examples to begin with:

  • "Which tables in this database have no primary key?"
  • "Where are the most null values in the orders table?"
  • "Which queries took the most time last week?"

Simple questions are answered right away. For heavier work an agent gets started. It picks the right tools, explains what it is doing and leaves its findings in the conversation.

What you just set up

You have defined the customer, the source, the data access and the rules once. Everyone who now works for this customer, in the web app or with their own AI tool, works under exactly the same rules.

Next steps​