# AI integration: ChatGPT, Claude and Gemini in your company systems

Mesterséges Intelligencia Központ Kft. is a software development company based in Budapest, Hungary. We connect the language models of OpenAI (ChatGPT), Anthropic (Claude) and Google (Gemini) to company systems, linking everything that exists with artificial intelligence.

> **In short:** through an API, the language model becomes part of the existing systems, and the result lands where people work with it: in the CRM, the ERP, the webshop or the mailbox. We put the model in a separate layer, so the provider can be replaced, and only the data the task needs reaches the model.

## What does AI integration mean?

It is not one more window that staff copy text into and out of. Through a programming interface (API), the language model becomes part of the existing systems: it receives the data the task needs, does what it is good at (extracting, summarising, classifying, translating, drafting), and the result lands where people work with it: in the CRM, the ERP, the webshop or the mailbox. Our article [How do we connect business systems with artificial intelligence?](https://www.mestersegesintelligenciakozpont.hu/en/knowledge-base/connecting-business-systems-with-ai) shows how this is built, step by step.

If the model must not only answer but also take steps, such as recording, sending or booking, it is an AI agent; we cover that on the [AI chatbots and agents](https://www.mestersegesintelligenciakozpont.hu/en/services/ai-chatbots-and-agents) page.

## What can it be used for?

| Area | Example |
|---|---|
| Documents | extracting and cross-checking data from invoices, contracts and delivery notes |
| Email | sorting and forwarding incoming mail, drafting replies for staff |
| CRM and ERP | summarising customer history, writing quote texts, asking questions about the data in plain language |
| Webshop | product descriptions, translations, more accurate search, customer service replies |
| Internal knowledge | a knowledge base built from policies, manuals and past projects that staff can query |

You can read about AI built into business systems on the [Business software](https://www.mestersegesintelligenciakozpont.hu/en/services/business-software) page, and about automated processes that span several systems on the [Process automation](https://www.mestersegesintelligenciakozpont.hu/en/services/process-automation) page.

## The model is replaceable

Language models evolve quickly: what is the right choice today may not stay so. That is why we put the model in a separate layer. The prompts, the data access and the quality measurement live in the company's system, and the provider (OpenAI, Anthropic or Google) can be replaced without rewriting the system. We measure every switch on the company's own examples before it goes live.

## Data protection: what goes where?

AI integration is also data processing, so the first design question is which data may leave the company's systems. What we pay attention to:

- **Data minimisation:** only the data the task needs reaches the model; personal data is masked or pseudonymised wherever possible.
- **The provider's terms:** we check where and for how long the model provider stores data sent through its business API, and whether it uses it for model training.
- **Contract:** the model provider is typically a data processor, so a data processing agreement is needed (GDPR Article 28); for high-risk processing, a data protection impact assessment as well (GDPR Article 35).
- **Access rights and logs:** the model sees only what the user asking the question may see, and it can be traced who asked what.

This is information, not legal advice: for the final assessment of compliance, it is worth consulting a data protection officer or a lawyer as well. The text of the regulation: [the General Data Protection Regulation (GDPR) on EUR-Lex](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32016R0679). If the AI talks to customers as a chatbot, our article [AI Act and GDPR: what to watch for in a business chatbot](https://www.mestersegesintelligenciakozpont.hu/en/knowledge-base/ai-act-and-gdpr-for-business-chatbots) sums up the transparency and data protection rules, with a checklist.

## How does an AI integration project run?

1. **Free first consultation:** which work the machine should take over, and which systems hold the data.
2. **Requirements analysis on real samples:** on the company's own documents and emails, anonymised where needed, we see what the model can do and where it makes mistakes.
3. **Size measurement and offer:** we measure the size of the development in COSMIC function points, and the offer and the schedule are based on it.
4. **Development in stages:** integration, access rights, logging and quality measurement, with regular demos.
5. **Handover and operation:** the contract governs source code and usage rights; after handover we also provide operation and maintenance, including monitoring how the model performs.

## Our own examples

We work this way ourselves. On our [homepage](https://www.mestersegesintelligenciakozpont.hu/en/), our AI assistant, which runs on a Claude language model, watches the game: it reacts to the big moments in roughly 1–2 seconds and talks with visitors in the game's 14 languages. The game's scenery and obstacles are made by our own production line: a language model plans, a generative image model draws, the software checks the result, and a person decides what is accepted.

## Frequently asked questions

### Can an AI integration be GDPR-compliant?

Yes, if it is designed that way: with data minimisation, a data processing agreement, suitable provider terms, access control and logging. Whether a specific processing activity complies depends on its purpose and the data involved; this is information, not legal advice.

### Which model do you choose?

It depends on the task. Our criteria: the quality of text in Hungarian, the type of task (text, documents, images), speed, usage fees, data handling terms and availability. We try the candidates on the company's own examples and recommend one based on the measured results. Since the model is replaceable, the decision can be revisited later.

### Can the AI answer from our own documents?

Yes. The method is called RAG (retrieval-augmented generation): the system first finds the passages in the company's documents that relate to the question, and the model answers from them, citing the source. It works with Hungarian documents and Hungarian questions too. More on this: [AI chatbots and agents](https://www.mestersegesintelligenciakozpont.hu/en/services/ai-chatbots-and-agents).

### What do you need from us?

- A clearly defined goal: which work should become faster or more accurate.
- A few real samples (documents, emails, questions), anonymised where needed.
- Access to a test environment or the interface of the systems to be connected.
- A contact person who knows the process.
- A decision on which data may be sent to the model.

### What does the price depend on?

It has two parts. The development fee follows from the functional size, which we measure in COSMIC function points (ISO/IEC 19761). The running cost is the model provider's usage-based fee, which is proportional to the amount of text processed (the number of tokens); we estimate it during design and weigh it when choosing the model. Details are on the [Pricing](https://www.mestersegesintelligenciakozpont.hu/en/pricing) page; if you have a question, write to us using the details on the [Contact](https://www.mestersegesintelligenciakozpont.hu/en/contact) page.