How to Connect Arnet (ארנט) POS to AI and External Systems

If you run a store or a restaurant on Arnet (ארנט), you already know it does the day-to-day job well: ringing up sales, printing receipts, tracking inventory, producing your end-of-day reports. What it was never designed for is talking to anything outside itself. There's no public, documented REST API you can just plug an AI agent into the way you would with a modern cloud SaaS. That doesn't mean you're stuck — it means the integration is an engineering job rather than a checkbox.

How Arnet actually exposes data

When a system has no advertised API, you work with the surfaces it does have. With Arnet, in practice, there are three realistic ones, in roughly descending order of cleanliness:

• The Israeli Tax Authority unified-structure export (מבנה אחיד / Mevne Achid). Every certified Israeli POS must be able to produce this standardized file of sales, documents and items. It's a documented government format — which makes it a stable, legal, well-defined way to read your transactions.
• Built-in report exports — sales reports, item movement, Z-reports — that Arnet can output to CSV/Excel or print files. These can be picked up, parsed and pushed onward automatically.
• Direct access to the underlying data store or installation files on the local/server machine. This is the most powerful and the most delicate — it must be read-only and handled carefully so you never corrupt the live register data.

The pattern is always the same: get a clean, reliable copy of the data out of Arnet on a schedule (polling, since there are no webhooks), normalize it into a sane shape, and only then hand it to AI or another system. The unglamorous middle step — scheduling, deduping, mapping fields, surviving format quirks — is the whole job.

What you can actually build

1. Daily AI sales briefing: every evening, AI reads the day's Arnet sales and sends the owner a plain-language WhatsApp summary — top items, slow movers, an unusual-day flag.
2. Automatic accounting handoff: the unified-structure export is parsed and pushed into your accounting software (Rivhit, iCount, Hashavshevet), so documents reconcile without manual re-keying.
3. Inventory and reorder alerts: AI watches item movement and pings your supplier or a Slack/WhatsApp channel before a fast-seller runs out.
4. Anomaly and shrinkage detection: flag voided sales, odd discount patterns or suspicious end-of-shift gaps for a human to review.
5. A chatbot over your own numbers: ask 'how did the Herzliya branch do last Shabbat vs. the week before?' in plain Hebrew and get an answer grounded in real Arnet data.

With a closed system like Arnet, the AI is never the hard part. Getting clean, trustworthy data out — reliably, every single day, without touching the live register — is the real work.

No-code vs. custom code — the honest call

If Arnet could already drop a CSV into a folder or email a report, a no-code tool like Make or Zapier can watch that file and do something useful — and for a single small business, that's often enough. But the moment you need to read the unified-structure file directly, run on volume, keep sales data private and on-prem, handle multiple branches, or apply real logic that no off-the-shelf connector covers, you've outgrown drag-and-drop. That's the line where you want a developer who'll work against the data directly and own the reliability end to end.

This is exactly the kind of work I do. If you're running on Arnet and looking to hire a developer to connect your existing systems to AI — without ripping out the register your business depends on — I build that quiet, reliable layer in between. Reach out through the contact section below and tell me what you'd want it to do; I'll tell you honestly what's possible and what isn't.