AI Solutions

AI in the ERP: What Works, What Is Being Built

We put AI around the ERP, not in place of it. This page separates what actually works today from what is under development and what is still on the roadmap.

Let’s try it on your own data

There are two ways to bring AI into ERP projects. The first is to promise a new system that rewrites everything. The second is to leave the working system in place and add what is missing on top. We do the second.

In practice this means one thing: AI makes the data a user is already allowed to see in the ERP visible faster. It does not widen permissions, it does not copy data out, and it does not make the decision.

Where we stand today

Working

Turning a plain-language question into validated SQL (Kai)
Running that query read-only and parameterised
Running in-house on a local model (Ollama / TensorRT)
Optional cloud model support
Questions and answers written to an audit log
Reports, dashboards and threshold alerts with ErpwareBI

Being built

Document-grounded question answering that cites its source (RAG)
Description and class suggestions on part and stock data (KaiDraw pilot)
Automatic summaries of deviation and anomaly alerts

Roadmap

An agent layer able to call tools
Standard connections to outside systems through MCP
Authorised and auditable ERP transactions (only in an approved flow)

The roadmap items are not products today, and are not described as such here. Ask where we are with any of them and you will get a straight answer.

How a question becomes an answer

Question
Language model
Validated SQL
ERP database
Answer and chart

The model never reaches the database directly; the query it produces is validated first, then run read-only and parameterised.

The chain of permission and audit

User
Role
Data permission
Read-only query
Audit log
Transaction rights roadmap

Data a user cannot see in the ERP is data the AI cannot see either. Operations that change data are out of scope today; when the agent layer arrives, they will be opened only in an approved and logged flow.

Where does the model run?

Inside the company (default)

The model runs on your own hardware; the question, the schema and the data never leave the building.

Local execution with Ollama or TensorRT
No internet connection required
Hardware, maintenance and updates are yours

Cloud (optional)

Where a more capable model is needed, a cloud model can be used within an explicitly agreed scope.

What data leaves the building is defined in writing
Personal data and commercially sensitive fields are kept out of scope
Local is the default; the cloud is an option, not a requirement

Read on

The ERPware Architecture: Which Product, and When? Which product comes in when
/kullanim-senaryolari What people actually ask, in six areas
Enterprise AI, Under Your Control The permission chain, audit logging and where data stays
Kai — Ask in Plain Language The working product: questions in plain language, local model
ErpwareBI — Business Intelligence The data layer: reports, dashboards, alerts
AI Glossary RAG, agents, MCP, inference cost and other terms

Every item on this page is written to the extent that it exists in the field. If a capability is marked “roadmap” here, it is not being sold today.