Connect your own API to AI: your own system joins the conversation too
A system that appears in no catalogue at all can be connected to AI as well. That might be an ERP built in house, a planning system made for you years ago, or a portal run by an industry association. Together we decide which data the assistant may request, and we always begin with reading. Letting it write is a separate choice: it happens only on an explicit request and the assistant shows first what it intends to send.
What the assistant can do inside your own system
Request information
The assistant retrieves what is held in the system and turns it into an answer in plain language. For the user it feels like asking a colleague who happens to have the file in front of them.
Only what was agreed in advance
Before anything works, we record together which part of the system the assistant may consult. Whatever falls outside that never ends up in an answer.
Asking instead of guessing
If it is unclear which file, which period or which branch is meant, the assistant simply asks. With an in-house system that matters more than elsewhere, because there is no standard setup to fall back on.
Writing as a separate choice
If you want the assistant to be able to create or update something, we enable that separately for a bounded part. It shows first what it intends to send and acts only on an explicit request.
Industry systems and portals too
Home-grown software is not the only candidate. A package that exists only within your sector, or a portal run by an industry association, can be made reachable for the assistant in the same way.
Systems with no ready-made connector
Industry and manufacturing
- Situation
- Production runs on an ERP built to measure years ago that nobody outside the company knows.
- What the agent does
- The assistant retrieves the status of an order and explains in plain language where it stands in the process.
- Result
- The office team answers without interrupting the one colleague who really understands the system.
Healthcare
- Situation
- A planning system made specifically for this organisation decides who works where and when.
- What the agent does
- The assistant answers questions about the coming days' schedule based on what that system holds.
- Result
- Fewer calls to the planner and fewer misunderstandings about who is expected.
Government
- Situation
- Part of the work runs through an umbrella organisation's portal that only a few colleagues can access.
- What the agent does
- The assistant retrieves the details that belong to a case currently in progress.
- Result
- More colleagues can answer a question, without everyone needing their own account in that portal.
Construction and installation
- Situation
- Every project sits in a work preparation system that appears on no standard list.
- What the agent does
- The assistant finds the project in it and summarises the open points.
- Result
- Preparation moves faster, including for colleagues who rarely touch that system.
Real estate
- Situation
- An in-house management application tracks maintenance reports, but nobody looks at it in between.
- What the agent does
- On request the assistant summarises what is outstanding per property and, if that is enabled, can create a report after approval.
- Result
- Reports sit around for less time, while creating one remains a deliberate act.
The path we walk together
-
1
First establish what may be opened up
We go through with your administrator which data from the system may become available and which definitely may not. That agreement is the foundation; only then is the connector switched on.
-
2
Access you keep in hand
The access comes from your own system and stays under your management. Withdrawing it is possible at any moment and takes effect immediately.
-
3
Who may use it and what it sees
The roles in the platform decide who gets this assistant. It sees only the part that was agreed, never the system as a whole.
-
4
What gets recorded
Every question and every change goes into the log. With an in-house system that is especially welcome, because there is no vendor keeping track for you.
An in-house system calls for firm agreements
In a system built specifically for you there is no standard division between what is sensitive and what is not. We therefore make that boundary explicit before anything is switched on.
Where the data lives: Nederland of de EU
The key stays yours
You issue the access and can withdraw it yourself. There is no detour around it and we keep no second way in.
Reading is the starting point
By default the assistant only retrieves information. Writing stays off until you deliberately choose it, and then remains limited to the part we agreed on.
What was asked stays visible
The log shows which questions were put and which changes were carried through, with the moment and the reason attached.
On a dedicated server
The model runs on our own server in the Netherlands. The data from your own system is not sent to an American model provider and is never used for training.
Questions about opening up an in-house system
Can I connect our own system to ChatGPT?
A public chatbot has no route into software that exists only inside your company, and it is not the place to copy business data to either. IntraGPT is used as a chat, but with a connection you open and close yourself, and with agreements about which part of the system comes into view.
Does IntraGPT work with our own system?
In practice almost always, provided the system can hand data to another application. This connector exists precisely for software that appears on no standard list, and it is in use at clients for exactly that kind of system.
Our software is bespoke and the documentation is thin, is that a problem?
No, and with in-house systems that is the rule rather than the exception. We start small, with a handful of questions that genuinely matter, and expand once it is clear what comes out well. Whatever is unclear the assistant asks about rather than turning it into an assumption.
Can the assistant break something in our system?
Not as long as it only reads, because then nothing changes. If you enable writing, a change can be just as real as a change made by a colleague, and sometimes just as irreversible. That is why the assistant shows first what it intends to send, waits for approval, and why the option stays limited to the part you have pointed out.
Our system is not reachable from the internet, what then?
That is a common situation and we look at it case by case with your administrator. It matters whether the system runs only inside the network or whether a secured way in is possible. If it cannot be done safely, we say so and look at whether another route delivers anything.
What happens if we change our system later?
Small changes go unnoticed. If the way data is made available changes fundamentally, a small adjustment is needed and you will notice because a particular question stops producing an answer. Tell us when a bigger overhaul is coming and we stay ahead of it rather than behind.
What does opening up an in-house system cost?
This is the one connector where the work genuinely differs per client, because every system is different. What decides it is how many kinds of questions the assistant has to handle, how well the system can hand over data and whether writing back is allowed. After a first conversation with your administrator we know fairly precisely.
How long does setup take?
A first version that can answer a few frequently asked questions usually stands within days. The rest grows with the usage, because only practice reveals which questions colleagues really ask.