Set it up once, then everyone works in it
Noor ran the first AI analysis at the wholesaler. Now purchasing wants in, and the controller, and the consultant who runs the warehouse. Everyone with their own AI tool and their own rules is exactly what the board does not want.
In Plainwater AI you set up a customer once: sources, data access, ground rules, knowledge and permissions. After that everyone works in the same environment, and every agent knows the same naming, definitions and limits.
Updated .
What do you set up once?
Per customer or department you record sources, data access, ground rules, knowledge and permissions. Everything that happens afterwards reads those settings, so nobody has to explain them again.
The platform is multi-tenant: one installation, a separate environment per customer. Inside an organisation a customer can be a department, so purchasing and finance each get their own sources and rules.
Access to another customer returns a 404. The platform does not even reveal that the customer exists, so a permissions mistake stays a closed door rather than a hint.
What does onboarding look like?
Onboarding has five steps, and every step has a way out if you would rather sort it later:
- Customer. Name and basic settings.
- Data source. For example the wholesaler's Azure SQL warehouse.
- Connection test. The platform checks the source is reachable before anyone works on it.
- Data access. Metrics only, encrypted or open, plus exceptions per column.
- Ground rules. Which models are allowed, the budget and the guidelines for the audit.
Noor finishes it in one session. Purchasing can then start a conversation straight away on the same source, under the same rules.
Who can do what?
Everyone signs in with Microsoft Entra ID, and permissions come from the platform itself, never from the user's request. There are four fixed roles: administrator, consultant, reviewer and viewer.
You can add your own roles as a subset of the existing permissions. A 'purchasing analyst' role might hold conversations and view dashboards without managing sources.
Nobody raises their own permissions, administrators included. An attempt to do so becomes an alert at the highest severity.
How does every agent know how you work?
Skills and knowledge documents stack in layers: platform, customer profile, customer and assignment. The most specific layer wins, so an assignment can sharpen a general agreement without replacing it.
Bram records once what 'margin' means at the wholesaler, and Noor records the warehouse naming. Every agent of every colleague receives those agreements when it starts.
You import knowledge documents from Word, Excel or PDF. An agent can propose a skill itself when it learns something that keeps coming back; a person approves or rejects it.
| Layer | Example | Managed by |
|---|---|---|
| Platform | How to review SQL | Platform admin |
| Customer profile | Agreements for all wholesalers | Platform admin |
| Customer | Definition of margin, warehouse naming | The customer's consultant |
| Assignment | Only the Fasteners product group | Whoever creates the assignment |
How does the platform handle test and production?
Per customer you record environments such as development, test, acceptance and production, each with its own rules. Production can be stricter than test, and then it is.
Per environment an agent and a person draw a landscape map together: sources, ETL, warehouse and reports. A person publishes it, and every agent in that environment receives it as a starting point.
A first responder answers simple questions directly and hands heavier work on by itself. Notifications appear in the app and optionally by email; an email never contains content.
Where it stops
Setting up once does not mean nothing changes. Sources get added and rules get stricter; the platform records every change as a new version.
A landscape map is a starting point, not the truth. The agent checks what it reads on it, and a person confirms what is certain.
Knowledge only works if someone keeps it current. The platform makes sharing easy, but it does not decide what your definition of margin is.
Frequently asked questions
Can I use Plainwater AI for several customers or departments?
Yes. The platform is multi-tenant: every customer or department has its own sources, rules, knowledge and permissions. Without access you cannot even see that another customer exists.
How do users sign in?
With Microsoft Entra ID. Permissions are read from the platform again on every action, not from what the user sends along.
Can I create my own roles?
Yes, as a subset of the existing permissions. Next to the four fixed roles you can, for example, set up an analyst role that cannot manage anything.
How do agents learn our agreements?
Through layered skills and knowledge documents: platform, customer profile, customer and assignment. At every start the platform works out which versions apply and passes them on.
Do I get an email when something is waiting for me?
If you switch it on. Notifications always show in the app; the email only says something is there and never contains conversation content or source names.