For bookkeeping firms · one file per client, one place to manage them
Close every client's books — in a day.
Sfarim learns each client's accounts, classifies the transactions, and produces export-ready journal entries. Your team approves instead of typing.
Automatic classification
Every bank, card, and processor line maps to the right account, learned from the client's own history.
VAT, computed from documents
Input and output VAT computed from the documents themselves, with alerts on gaps before you file.
Export to your ledger
Journal entries in the format your accounting software ingests — no re-keying.
Multi-client
A file per client, one consistent state across the firm, as many closes in parallel as you need.
The hours that went into typing — gone.
Most of the time on a client file isn't judgment — it's typing, scanning, and manual reconciliation. Sfarim takes that part off your plate, and frees your team for the work that's worth more.
Every invoice matches its charge on its own — cross-checked with the Tax Authority.
The invoice arrives from the inbox, the vendor, or straight from the Tax Authority's own feed — and matches automatically to the card or bank charge. No manual invoice entry, no duplicates.
Every client file, in one place — each in its own state.
Every client gets a separate file with its own live status: ready to export, needs a look, or already sent to the ledger. You see at a glance which files are waiting on you, without opening each one.
Foreign-currency journal entries — including הפרשי שער, automatically.
Every movement is booked in its original currency and translated to shekels at the day's rate. When a customer or supplier pays at a rate different from the one the charge was booked at, the exchange-rate difference is computed and closed to the right account automatically — no manual rate math, no correction entries.
The machine proposes. You sign.
Automation in bookkeeping is worth exactly as much as you can trust it. So nothing is posted without human approval — the classification is deterministic, learned from the books themselves, and every action stays transparent and auditable.
- Deterministic, not guesswork — learned from the client's own history
- A full audit trail: what was proposed, who approved, and when
- Anything ambiguous is flagged for review, never posted silently
The hesitations we hear, answered straight.
These systems make mistakes — and I'm the one who signs the return.
Exactly why no number reaches the books without your sign-off. Classification is learned from the client's own books, not a model's guess, and anything ambiguous is flagged for review instead of posted quietly. The honest comparison isn't machine versus perfect, it's machine versus manual keying, which has its own errors. Every entry stays transparent: what was proposed, who approved, and when.
I don't have time to learn a new system.
There's no new system to learn. The entries land straight in the accounting software you already run, and you approve instead of keying. No chart of accounts to map, no rules to maintain — the system learns from the historical books. You start with one client, not a firm-wide migration.
I already scan documents with an OCR tool.
Scanning isn't bookkeeping. A capture tool hands you a document and a few fields; someone still classifies it, posts the journal entry, reconciles the bank, and computes VAT against the Tax Authority. That's the part that takes the time, and it's exactly what we do. Keep the scanning — the question is what happens to the data after it.