What is OpenAI, and what does the ChatGPT API do?
OpenAI is the US-based AI company that develops ChatGPT. The language models behind ChatGPT are also offered to developers through the OpenAI API, which people often simply call the "ChatGPT API". With the API, these models can be connected to a website, a customer portal, a CRM or an internal application: understanding text, summarizing, classifying, translating, extracting information from documents and answering questions in natural language all become part of your software.
An OpenAI integration means connecting these capabilities to your data and workflows in a controlled way. What matters is not only the model itself but also which data is sent to it, how the output is checked and how records are kept.
What do we connect with OpenAI?
| Data | Direction | Frequency |
|---|---|---|
| Incoming email or form text | System → model → reply draft | On arrival |
| Document (quote, contract, invoice) | System → model → structured fields | When uploaded |
| Product information | ERP / IMFLEXI → model → description draft | As needed |
| Visitor question and approved knowledge source | Website → model → answer | Real time |
| Activity log (request, response summary, cost) | Middleware → log | On every call |
Typical scenarios
- Email reply drafts: An email arriving in the support inbox is classified, the related order or customer details are added, and a reply draft is prepared; the decision to send stays with the employee. See answering email with AI.
- Extracting data from documents: PDF quotes from suppliers are read, line items and amounts are put into a table, and the employee only checks them.
- Website assistant: An assistant is set up that uses the product and service information on your site as its source and, instead of making things up, hands off to your team when it doesn't know. The Globya assistant on our own site is a live example of this approach.
Globya's approach
We build AI integrations as part of our AI-powered software service. Our principles are simple:
- Task first, model second: We measure up front how many minutes a task takes and whether AI really helps.
- Human approval: For every output that goes to a customer or affects money or a contract, the final decision stays with a person.
- Model independence: We design the middleware so it can be switched to Anthropic Claude or Google Gemini when needed.
How do we work?
- Analysis: We define the task to be automated, the data to be used and the expected level of accuracy.
- Mapping: The fields sent to the model are chosen, personal data is masked if it isn't needed, and instructions and output format are written.
- Test environment: It is tried on real examples (anonymized if necessary) and the results are scored with your team.
- Go-live: It is switched on first in draft mode, with human approval.
- Monitoring: Every call is logged; on errors or timeouts the request is retried, and cost and accuracy are reported regularly.
What to prepare before we start
- A description of the task you want to automate and how long it takes today
- Sample documents, emails or records used in that task (anonymized if necessary)
- Where the output should go: admin panel, CRM, mailbox
- Who will check and approve the output
- Your company's AI usage rules, if you have any
We run a small trial on your examples, measure the result together and then decide.
Security, cross-border transfer and KVKK
OpenAI is a service headquartered abroad; data sent to the model is processed outside Türkiye. That is why, before sending personal data, we assess whether it is truly needed and, where possible, mask or anonymize it. If personal data will be sent, the cross-border transfer rules and the notice obligation under KVKK (Türkiye's Personal Data Protection Law) are addressed. We also review the provider's current terms on data retention and use for model training with you. The API key is kept in a restricted configuration, not in the code, and is used only on the server side. For details, see our article on AI and company data.
To talk about your AI project, call +90 850 432 55 13 or use the contact page. Pricing follows a written proposal after the discovery call. Describe your needs now in 3 minutes
Frequently asked questions
We have a ChatGPT subscription. Is the API separate?
Yes. A ChatGPT subscription is for use in the browser and the app; to connect to your systems, a separate OpenAI API account is opened and billed based on usage.
What happens if the AI gives a wrong answer?
That is why we recommend human approval for outputs that go to customers. For assistants, we set them up to answer only from an approved knowledge source and to hand off to the team when they are not sure.
Will our data be used to train the model?
That depends on the provider's current terms and your account settings. We review the terms together before setup and configure the necessary settings.
Can we switch to a different model later?
Yes. We design the middleware so the model can be changed.