Connecting Hashavshevet (חשבשבת) to AI and Modern APIs — Honestly

If you run Hashavshevet (חשבשבת), you already know it is solid, deep, and not built for the cloud era. Thousands of Israeli businesses and bookkeepers rely on it, which is exactly why the question keeps coming up: can I connect Hashavshevet to AI and to my other modern systems? The honest answer is yes — but not the way you connect a modern SaaS tool. Let me walk through how it really works.

What Hashavshevet actually exposes

Unlike newer cloud invoicing tools, Hashavshevet does not hand you a public REST API with a token and clean JSON. It is a desktop/on-prem product, and integration happens through a few well-worn channels rather than a single modern endpoint:

• The standardized movement-file interface ("ממשק חשבשבת") — fixed-width / delimited text files that import and export journal entries, customers, items and documents. This is the most reliable, officially supported door in and out.
• Israel Tax Authority files such as the openformat / unified structure (מבנה אחיד, BKMVDATA) that Hashavshevet can produce — useful for read-only analysis and reconciliation.
• Direct database access on the on-prem machine, where permitted, for read-heavy reporting (use with care and only against a copy or with vendor guidance).
• Vendor add-ons and integration modules from Hashavim, which sometimes expose richer programmatic hooks for partners.

In practice, the movement-file interface is the workhorse. You generate or consume a structured file, and AI sits on either side of it: reading what comes out, or preparing clean, validated data going in.

What you can actually build

1. An AI invoice reader: a vendor PDF or photo arrives by email or WhatsApp, AI extracts supplier, amount, VAT and date, and a script emits a ready-to-import Hashavshevet movement file — your bookkeeper approves instead of typing.
2. Automated bank reconciliation: AI matches bank transactions against open entries exported from Hashavshevet and flags only the exceptions for a human.
3. A WhatsApp assistant that answers "what is customer X's balance?" by reading data exported from Hashavshevet — never exposing the raw system.
4. Cash-flow and anomaly alerts: a nightly job pulls fresh figures and an AI model writes a short Hebrew summary plus any red flags to email or Slack.
5. A bridge to your modern stack: push relevant accounting data into your CRM, a BI dashboard, or a customer portal so the on-prem software stops being an island.

On-prem is the real constraint

Because Hashavshevet usually lives on a machine in the office, the hard part is not the AI — it is getting data in and out safely and on a schedule. That means a small agent or scheduled job on-site, a secure path to wherever the AI runs, and disciplined handling of sensitive financial data. Privacy matters here: for many clients I keep the AI processing local or in an Israeli/EU region and never let raw books leave the building unencrypted.

With Hashavshevet, the AI model is trivial. The engineering is moving data off an on-prem box reliably, on time, and without ever corrupting the books.

No-code vs. custom code

Tools like Make or Zapier are wonderful — when both sides have an API. Hashavshevet usually does not, so the no-code route stalls at the first step. You can sometimes bridge it with a watched folder, an email gateway, or a Google Sheet as a middle layer, and for very light, low-volume flows that is genuinely enough. But the moment you need real volume, validated imports, on-prem access, or sensitive data kept private, you need custom code that understands both the file format and the accounting logic behind it.

This is exactly the kind of unglamorous, reliable integration I build for Israeli businesses. If you are looking to hire a developer to connect Hashavshevet to AI or to your other systems — without ripping out software that already works — that is my day job. Tell me what you have running today and what you wish it did automatically, and I will map out a realistic path. The contact form on this page reaches me directly.