Document Intelligence
on your terms
Your documents. Your infrastructure. Your control.
Extract structured data from invoices, waybills, POs and more — privately.
1. Upload
PDF, images, CSV, TXT, Office docs. Multi-page, multi-file.
2. Configure
Choose up to 20 fields. Invoice, VAT, supplier, waybill and custom.
3. Extract & Export
Structured results. Excel, CSV, JSON, PDF or direct to your systems.
Local Server Connection
In production this interface connects to the processing engine running on your Windows or Ubuntu server.
Document Intake
Drop files here or click to browse. Supports PDF, PNG, JPG, TIFF, CSV, TXT, DOCX and more.
Drag & drop documents
or click to select files
Extraction Fields
Select up to 20 fields. Presets available. Configurations are saved locally in your browser.
0 / 20 fields
Processing
In the production system this runs on your local server. Demo simulates extraction.
Extracted Results
Review and edit values. Confidence scores shown in demo.
| Field | Value | Confidence | Source Page |
|---|
Output & Storage
Choose how results are delivered. Nothing is stored by Nexa — data stays with you.
Download
Send to Destination
Webhook, local folder, or cloud connector.
Integrations
Connect once. Results flow to the systems you already use.
Google Sheets
Append rows to a spreadsheet.
Microsoft SharePoint / OneDrive
Save files or lists.
SFTP / Local Folder
Drop files on your network share.
Webhook / API
POST JSON to any endpoint.
Help & Instruction Manual
For clients and administrators. Print this page for a paper manual.
1. Deployment (Administrator)
Install the Nexa Extract processing engine on a Windows or Ubuntu server inside the client network. The engine runs OCR and AI models locally. Staff access this interface via browser at the server’s IP address (e.g. http://192.168.1.50:8080). No internet required for core processing.
2. Connect
Open the interface in any modern browser (Chrome, Edge, Firefox, Safari). Enter the server IP on the Dashboard if prompted. The connection is local — documents never leave the client’s infrastructure.
3. Upload Documents
Go to Intake. Drag-and-drop or select files. Supported formats include PDF (multi-page), images, CSV, TXT, and common Office formats. Multiple files can be processed in one batch.
4. Select Fields
In Fields, choose a document type or build a custom set (max 20). Common fields: Invoice Number, Supplier VAT, Supplier Address, Receiver Details, Date, Subtotal, VAT Amount, Total (Rand), Product Lines, PO Number, Proforma Number, Waybill Number, POD Reference, etc. Save the configuration — it persists for future batches.
5. Process & Review
Run extraction. Review the results table. Correct any values. Confidence scores help prioritise checks. Original documents remain on the client server; only extracted structured data is produced.
6. Output
Export to Excel-compatible CSV, JSON or PDF. Or send via webhook, Google Sheets, SharePoint, OneDrive, SFTP or local folder. OAuth logins are performed by the client — Nexa never stores credentials or document content.
7. Retention & Deletion
Retention periods are set by the client according to their legal obligations (tax, Companies Act, etc.). The system supports configurable retention and verified deletion. Nexa, as operator, processes only on instruction and retains no client data.
South African Compliance
Designed for POPIA, tax record rules and related legislation. This is guidance, not legal advice.
POPIA (Act 4 of 2013)
Nexa acts as operator. Client is responsible party. Written operator agreement required covering security (s19), processing only on instruction, and immediate breach notification (s21). No personal information is retained by the provider.
Retention
Tax Administration Act & VAT Act generally require 5 years for specified records. Companies Act s24 provides 7 years for certain company records. Client configures retention per record type. System supports deletion verification.
Security & Incidents
Encryption in transit and at rest on client infrastructure. Audit logs. Incident response process: operator notifies responsible party immediately. Responsible party has s22 obligations.
Other Laws
PAIA access procedures, ECTA electronic records, Cybercrimes Act, Consumer Protection Act disclosures. Cross-border processing (if any) requires s72 assessment. Default recommendation: keep processing inside South Africa or on client servers.
Hugging Face & Model Usage
Platform is independently branded. Compatible open-source models (including Hugging Face ecosystem where licences allow) may be used. Hosted Inference Endpoints currently run in US/EU regions and retain logs for 30 days — not recommended as default for confidential South African documents. Prefer local model deployment.
Service Plans
Software, deployment, model management, support and updates. Not a pure Hugging Face resale.
Starter
- Up to 500 pages / month
- Standard field templates
- CSV / JSON export
- Email support
- Client-hosted option
Business Private AI
- Unlimited pages (fair use)
- Custom fields & rules
- All exports + integrations
- Local model deployment
- Priority support + updates
Enterprise
- Dedicated infrastructure
- Custom integrations
- SLA & on-site training
- Advanced governance
- Multi-department configs