Skip to content
AI & agents / Context
Grounded, not guessed
AI & agents
Language models with context · Agents with rules
Every answerwith sources · with rules · with approval

AI needs meaning.

Language models and AI agents write brilliantly, but they do not know your organisation. MAKOR gives them what they lack: your data, your terms and clear rules. That makes answers verifiable and agents predictable. The decision stays with people.

Language models
With context
Agents
With rules
Answers
With sources
Gaps
Named openly
Decision
With people
Operation
In Germany
In simple terms

Language model, agent, MAKOR: what is what?

Three terms are enough to understand how AI works in MAKOR. Explained without prior knowledge, each with a comparison from everyday work.

Language model

LLM

A program that understands and writes language. It has learned from a vast number of texts how language works, but not how your organisation works.

AI agent

Agent

A language model that does not just answer but carries out tasks in several steps: querying data, checking results, preparing reports, notifying people.

MAKOR

Foundation

The foundation both work on: verified data, clear terms, defined rules and a source for every statement.

More terms in the glossary
Why context

A language model on its own guesses.

The same question, asked twice: once to a language model without access to your data, once in MAKOR. The difference is not in the wording but in the foundation.

The question

“Which suppliers put our production in March at risk?”

Without MAKOR · language model alone
“Suppliers with long lead times and high dependency typically pose a risk. Check your most important suppliers in particular and keep alternatives ready.”
  • No source, no date
  • No reference to your suppliers
  • Sounds certain, is guessed

The language model knows neither your suppliers nor your orders. It writes what sounds likely.

With MAKOR · language model in context

“Two suppliers are critical. Data is missing for a third.”

  • L-204 has been 12 days behind schedule.ERP · as of 8 Nov
  • L-311 is the only supplier of component B-7.Master data
  • There has been no delivery data for L-118 since October.Gap flagged
  • Source and date for every statement
  • Your suppliers, your terms
  • Gaps named openly, not glossed over
Context instead of assumptions

What MAKOR gives the AI.

A language model is only as good as what it knows about your situation. Before every answer, MAKOR compiles the relevant extract for it, from five kinds of knowledge. Specialists call this grounding: the answer is grounded, not guessed.

01 · Facts

What is proven

Data from all connected sources, verified and with provenance: which source, which version, from when.

From Makor Logos
02 · Terms

What is meant

What “customer”, “outage” or “order” means in your organisation, defined in a shared ontology.

From Makor Logos
03 · Processes

Where it stands

Which process step a question is at and what depends on it.

From Makor Äther
04 · Place and time

Where and when

Where something happens and how it has developed, now and over time.

From Makor Atlas
05 · Rules

What is allowed

What an AI may see and do, and when it has to ask a person.

From Makor Äther
Agents with rules

Agents that know what they are allowed to do.

AI agents take routine work off your hands: they reconcile data, detect deviations and prepare decisions. In MAKOR they work within a fixed framework, like a specialist with a clear job description.

Clear assignment

Each agent has one task and a defined framework, such as “check the southern supply chain daily”.

Limited access

An agent only sees the data its role allows. Permissions apply to it just as they do to people.

Every step logged

What was queried, inferred and proposed is recorded in the audit trail, traceable back to the source.

Approval by people

Anything that carries weight, the agent submits for decision. It only acts after approval.

Specialists call this framework a harness: the gear that steers the power of a model in one direction. In MAKOR, Makor Äther takes on this task.

From asking to acting

You decide how much the AI does on its own.

Not every task needs an agent. MAKOR distinguishes four levels, from a single question to a prepared action. With each level, the AI takes on more work. The responsibility stays with you.

Level 01

Ask

How many orders are open?

An answer in conversation, with a source for every statement.

Workspace
Level 02

Analyse

Show me how the suppliers are connected.

A network, time series or map is created from the question.

WorkspaceAtlas
Level 03

Monitor

Let me know if a supplier falls more than five days behind.

An agent keeps an eye on data streams and reports deviations.

ÄtherAtlas
Level 04

Prepare

Prepare a request to a replacement supplier.

An agent carries out several steps and submits the result for approval.

ÄtherWorkspace
The AI takes on more work · the responsibility stays with people
Decisions that carry weight are made by a person.
At every level: the technology recommends, the decision stays with you.
Sovereign AI

The AI stays in your hands, too.

Models of your choice

MAKOR integrates language models that suit your requirements, operated in Germany or in your data centre.

No third countries

No routing through third countries, no externally controlled services in the data path, not even for the AI.

Your knowledge stays with you

You harmonise your data yourself and use it to train MAKOR. What MAKOR learns stays within your organisation.

Traceable for auditors

Sources, logs and approvals support requirements for transparency and human oversight, such as those of the European AI Act.

Questions about the AI

Answered in brief.

What the AI in MAKOR may do, what it relies on and who decides in the end.

All questions

How can I check an answer?

Every piece of information carries its source, version, time and reliability. An audit trail records every processing step. That keeps it traceable how an answer came about, even if a language model worded it.

Who trains MAKOR?

You do. You harmonise your data yourself and use it to train MAKOR. What MAKOR learns stays within your organisation.

Does the AI in MAKOR make up answers?

MAKOR is designed to prevent exactly that. Answers rely on your data and terms, not on what a language model considers likely. If the data is not sufficient, MAKOR says so openly and shows where it is missing instead of filling the gap with a guess.

Who decides in the end?

Always people. Your experts assess, weigh and decide. MAKOR provides the foundation, the responsibility stays with you.

What is an AI agent, and what may it do in MAKOR?

An agent carries out tasks in several steps: checking data, reporting deviations, preparing a report or a request. In MAKOR, every agent works with a clear assignment, only sees the data its role allows and logs every step. Anything that carries weight, it submits to a person for approval.

Which language models does MAKOR use?

MAKOR integrates language models that suit your requirements for performance and sovereignty, operated in Germany or in your own data centre. There are no externally controlled services from third countries in the data path.

Does MAKOR help with the requirements of the European AI Act?

MAKOR supports key requirements such as transparency, logging and human oversight: every answer names its source, an audit trail records every processing step, and decisions that carry weight are made by a person. Whether and how the regulation applies to your deployment depends on the specific use case.

What would you ask your data first?

Bring a question that today fails because of scattered sources. We will show you how language models and agents answer it in MAKOR, with sources and approval.

Contact info@bonn-systems.com