Blog · 27 October 2026 · 6 min read
Local AI and your business management system: export or API?
Orders, customers, stock levels, invoices: the data that matters is in the management system, not in the documents. It is natural to wonder whether the local AI will be able to use it. The answer depends less on the AI and more on how the management system lets its data out.
In short: there are two routes. With export, every so often you download the files you need from the management system (lists, reports, master data) into a connected folder, and the AI reads them by itself: it works with any management system. With APIs, that is, the door some management systems offer to other programs, the AI queries the data directly and always finds it up to date: this is feasible only if the vendor makes them available. In both cases the AI works in read-only mode: it does not enter, change or delete anything in the management system.
Can I query the data in my management system with the local AI?
Yes, if the data can leave the management system in a form that another program can read. Almost all management systems have at least an export function to Excel, CSV or PDF, and that is enough to get started. Some also offer a direct connection.
| Aspect | File export | Connection via API |
|---|---|---|
| When it can be done | Always, if the management system can export | Only if the vendor offers APIs |
| What you need to do | Exporting the data from time to time | Nothing, after configuration |
| How up to date the data is | As of the last export | At the moment of the question |
| Initial work | Minimal | Dedicated development, to be quoted after reading the API documentation |
| Third-party costs | None | Any access costs set by the vendor |
How does it work with exported files?
You always save the export in the same folder and the rest happens by itself. The folder is connected to the AI's archive: Open WebUI's synchronisation tool notices the new file, loads it and removes the version you replaced. For the computer's folders this happens as soon as the file changes.
Three measures make this route much more useful:
- Always export the same views, with the same columns: customers, open orders, stock levels, due dates. The AI learns to find its way around files that always have the same shape.
- Replace the file, don't pile them up. If you keep the stock levels for January, February and March in the folder, the AI may answer with the old ones.
- Give the export a regular schedule: every Monday morning, at month end, after the stocktake. The data is only as up to date as the last time you exported it.
Tip: check whether your management system can schedule exports and save them to a folder by itself. Many can: in that case this route becomes automatic too.
And if the management system has APIs?
Then the connection can be direct: the AI asks the management system for the data at the moment you put the question. Open WebUI lets you connect external services as "tools" that the model uses when needed: once the management system's entry door has been described to the system, a question such as "how many units of this item do we have in stock?" becomes a request to the management system, and the answer arrives with the figure as it stands at that moment.
Before deciding, some information is needed from the vendor of the management system. Here is what to ask:
- Are there documented APIs? You need the documentation, not a promise.
- What data do they expose? The whole history or only part of it? Data from years ago too?
- Can a read-only user be created specifically for the AI?
- Do they have a cost? Some vendors include them, others sell them as a separate module.
- Are there limits on the number of requests?
- Is the management system installed in the company or is it in the cloud? That changes where the connection passes through.
With these answers the work can be estimated. Without them, any quote is a gamble.
Can the AI write to the management system?
No: we connect it as read-only, and we recommend that to everyone. A language model gets things wrong now and then, and a reading error is an answer to correct, whereas a writing error is a wrongly entered order or a damaged master record. That is why the user with which the AI accesses the management system must only be able to look things up, and this limit has to be set in the management system itself, not left to the AI's goodwill.
This applies to exported files too: the AI reads the copy in the folder and does not touch the original data.
What it does not do well
- Totals on files read as text. If an export is merely indexed as a document, the AI is good at finding a single row ("what price did we sell to this customer at?") but is not reliable for calculations over thousands of rows. For sums, averages and comparisons it has to work on the file as data, running the calculations: it is a different feature, which has to be planned for.
- Out-of-date data. With export, the answer is valid as at the date of the file. It is worth asking the AI always to state it.
- Questions that span many tables. "Margin by customer, by product line, over the last three years" requires data already prepared by the management system, not three exports to be cross-checked by eye.
- Changes decided by the vendor. If the management system changes its file format or API version, the connection has to be adapted.
How we go about it
We almost always start with the export: it costs little, works straight away and shows which questions are really worth a direct connection. In the site survey, on site at the client's premises and throughout Italy, we look at the management system together, choose the views to export and set up the connected folder. If the vendor offers APIs, we read the documentation and propose the connection as a separate job, with its own quote.
We develop management systems ourselves too, so we know the problem from both sides. The rest of the project is described on the Local AI for businesses page; to find out which route is possible with your management system, write to us and tell us which one you use.
About the author: Dabryx
Dabryx is an Italian software house that installs local AI in companies, on site across Italy, on servers or Macs. The basic installation starts at €2,900, hardware not included, and consulting at €100 per hour. The office is in Rocca Priora, near Rome.
We also install local AI in Rome, Milan, Parma and Reggio Emilia, and we develop custom management systems and custom software.
Frequently asked questions
Can a local AI be connected to the company's management system?
Yes, in two ways: by reading the files you export from the management system into a connected folder, or by querying it directly through the APIs, if the vendor makes them available. In both cases read-only.
What are the APIs of a management system?
They are the door the management system offers other programs for reading its data in an orderly, controlled way. Not all management systems have them, and some vendors sell them as a separate module.
Can the local AI enter orders or change data in the management system?
No, the connection is set up as read-only with a dedicated user that can only look things up. It is a security choice: a reading error can be corrected, a writing error damages the data.
How much does it cost to connect the AI to the management system via API?
It depends on the APIs: it is quoted after reading the vendor's documentation. On top of this come any API access costs, which are set by the vendor of the management system. The export route has no third-party costs.
Want an estimate for your company?
Tell us how many people will use the AI and what for: we'll tell you which configuration you need.
Response times
Within 2 hours on business days (9:00–18:00) for clients with a support contract.
AI consulting
Basic installation from €2,900, consultancy from €100 per hour, on your company's server or Mac.
Where we work
Throughout Italy: we come on site to the client's premises and also support you remotely.