Skip to main content
Evidence infrastructure for finance teams

Reconcile money across banks, exchanges, and invoices.

Bring PDF statements, scanned pages, spreadsheet exports, and invoice photos into one controlled ledger. Every extracted row keeps its source, every AI call runs under an approved budget, and a person signs off before anything is published.

No bank passwords or wallet custodyEvery number keeps its sourceHuman approval before close

Evidence queue

Current workspace

Sources reconciled
Evidence chainDEMO-AE-2026-07

Bank PDF (scanned) · July 2026

Source evidence preserved

OCR · 6 pages · 12 rows need review

Exchange CSV · July 2026

Source evidence preserved

Reconciled

Invoice photo · ZS-DEMO-0726

Source evidence preserved

3 fields need confirmation
Real problems, fixed

Month-end still runs on files nobody can trust.

Bookkeepers and finance operators receive statements as PDFs, scans, and phone photos, then rebuild them by hand. zStatement replaces that chain with governed extraction, review, and evidence packs.

Statements arrive as scans, not data

01

Bank PDFs without a text layer, WhatsApp photos of invoices, and Arabic receipts stall the close until someone retypes them.

Managed document OCR from Mistral, Azure, and Google reads images and scanned PDFs, including Arabic tables, and records page count and confidence on every row.

Close depends on one spreadsheet

02

Gaps, overlapping periods, and opening balances that do not match the previous statement surface weeks later.

Source accounts carry expected periods and balance continuity; exceptions get an owner, a due date, and a revalidation receipt.

AI answers you cannot audit

03

Generic chat tools invent totals, leak documents, and leave no record of which model produced a category.

OpenAI and Anthropic run only through allow-listed models with per-request and monthly budgets, redacted telemetry, and provenance on each suggestion.

Bank connections you cannot grant

04

Clients will not hand over online-banking credentials, and aggregators do not cover regional banks.

zStatement never asks for bank passwords or wallet custody. Evidence enters by upload, email, WhatsApp, or Telegram and stays in your private storage.

Every number keeps its evidence

Bank activity, exchange exports, receipts, and invoices move through one reviewable chain instead of disappearing into disconnected tools.

01

Verified ingest

PDF, scan, spreadsheet, and image sources enter with identity, fingerprint, malware scan, and provenance.

02

Traceable extraction

Every extracted row retains page, provider, confidence, and processing version.

03

Human review

Proposals, edits, and approvals stay explicit, assigned, and auditable.

04

Controlled publishing

Evidence packs, reports, and accounting handoffs reuse the same reviewed data.

What you can rely on

Built for operators who answer to auditors.

Scanned and Arabic statements

Images and scanned PDFs go through managed document OCR with per-page budgets; Arabic and English tables are reconstructed, not guessed.

Segregation of duties

A preparer records the revision, a different owner or administrator approves it, and the approval is a signed, dated receipt.

Evidence packs on demand

Deterministic CSV, XLSX, history, and manifest exports with hashes for every approved statement scope.

Fail-closed by design

Missing scanner, missing credential, or exhausted budget stops the job with a recorded reason instead of a silent fallback.

Managed assistance stays tied to evidence

Text tasks use the approved OpenAI or Anthropic route; document reading uses Mistral OCR, Azure Document Intelligence, or Google Document AI. Every call is reserved against a budget first, and no hidden fallback exists.

Grounded retrieval

Answers use approved managed models with source-retrieved statement context and citations to workspace evidence.

Fail-closed policy

An unavailable provider returns a verified error instead of a synthetic or browser-generated result.

Auditable operations

Provider, model, pages, confidence, cost, and policy decisions are recorded for operator review.

Who it is for

Designed for the teams that actually touch the statements.

Boutique bookkeeping firms

Receive client statements in any format, keep preparer and approver roles separate, and hand auditors an evidence pack instead of a folder of PDFs.

UAE and GCC operators

Arabic-first interface, Arabic-Indic numerals, RTL documents, and OCR that handles bilingual bank layouts and VAT invoices.

Finance teams without bank feeds

Multi-bank, multi-currency evidence without credential sharing; exports shaped for QuickBooks and Xero when you are ready.

Operate the stack from one console

Platform health, runtime settings, encrypted secrets, provider routing policy, OCR and AI budgets, and audit controls are governed from a single administration surface.

Applications and dependencies
Runtime configuration
Encrypted credentials
Provider budgets and audit

Platform control

System overview

API and workerReady
Database schemaMigrated
Document OCR providersMistral · Azure · Google
Object storageConnected
Runtime changes are explicit, audited, and restart-aware.
Document studio

Compose reports from reviewed data

Structured document tools preserve workspace context, revisions, RTL support, typography, and print-safe output without turning the editor into a separate product.

Open document studio

Management report · July 2026

Operating position

Revenue
124,820
Outflow
98,144
Net position
+26,676

Operating cash remained positive through the review period. Supplier outflow is the primary variance, with evidence references attached to each material classification.

First controlled run

Start with a synthetic statement, then bring your own.

Download a known fixture, watch extraction, review, and approval happen end to end, and only then connect real evidence. No card, no bank login, no custody.

zStatement | Financial clarity, explained