From v2.1 to v3.0 and beyond. Everything we are building over the next twelve months to make document AI truly effortless, private, and enterprise-ready.
This is a living document. It reflects our current priorities, informed by real conversations with early adopters and the day-to-day pain points of Italian businesses drowning in paper. Dates are targets, not promises.
Each milestone is grouped by quarter and version number. Features marked with a green check are already shipped. Everything else is planned or in active development.
The feature is live in production and available to every user right now.
Active development. Expected to land in the target quarter.
On the backlog, validated by user research. Timing may shift based on feedback.
March 2026. This is what DataUnchain can do today. Every feature below has been validated on real Italian invoices and deployed on-premise with early testers.
Our proprietary Vision-Language Model reads invoices, delivery notes, and receipts directly from images — no OCR layer required. It handles rotated scans, stamps, handwriting, and complex table layouts.
A dedicated Python layer cross-checks every extracted number: Taxable + VAT = Total, line-item sums, rounding tolerance. Inconsistencies are flagged before anything reaches your ERP.
Fattura Elettronica adapter is production-ready. The architecture is modular: each ERP connector is a standalone Python adapter that maps JSON extraction to the target format (XML, CSV, SQL).
The system learns from every correction: when a user fixes an extraction, the correction is stored and used to improve future prompts. Accuracy climbs with every processed document.
Complete containerised stack: VLM engine, processor, PostgreSQL, hot-folder watcher. Runs on a single GPU-equipped machine. Zero cloud dependency, absolute data privacy.
Drop a PDF into a watched folder or POST it to the API. Extraction starts automatically. Results land in PostgreSQL and are available via JSON or webhook callback.
Switch between document types — invoices, transport documents, contracts — by changing a single prompt string in the configuration. No retraining, no redeployment.
A lightweight vision-guided RPA module that types extracted data directly into legacy interfaces — no selectors, no brittle macros, no cloud APIs. Works with any screen-based ERP.
April – June 2026. The focus shifts from “does it work?” to “can anyone set it up in one afternoon?” We harden the product, onboard the first paying customers, and ship the first third-party ERP connector.
Five selected businesses get a free, fully-managed installation, six months at zero cost, and a direct line to the founder. In return, they help us validate real-world edge cases and shape the product. Apply now →
A guided setup experience that walks new users through GPU detection, Docker health checks, model download, and the first test extraction — all from a single web interface.
A semi-automated pipeline that takes accumulated user corrections, packages them into training data, and fine-tunes the VLM on Italian invoice patterns. The first iteration targets global accuracy improvements across all suppliers.
The first official third-party ERP adapter. Danea Easyfatt is the most widely used invoicing software among Italian micro-businesses. The connector maps extracted JSON directly into Danea's import format, closing the gap between AI extraction and daily bookkeeping.
Full installation guide, API reference, adapter development tutorial, and troubleshooting playbook. Published as a searchable docs site — because great software without great docs is just a puzzle.
A lightweight web dashboard where operators can review extractions side-by-side with the original document image, approve or correct fields, and trigger re-processing. The corrections feed directly back into Progressive Learning.
July – September 2026. DataUnchain learns to handle volume. Batch processing, per-supplier intelligence, multi-language support, and an upgraded API designed for system integrators.
Progressive Learning v2 introduces automatic confidence scoring. When the model's confidence on a field drops below a configurable threshold, the extraction is routed to human review. Approved corrections are incorporated continuously, creating a flywheel that makes the system smarter with every batch.
Instead of a single global model, the system can now maintain lightweight per-supplier LoRA adapters. If a particular supplier uses an unusual layout, the VLM learns that layout specifically — without forgetting everything else. Think of it as muscle memory for each vendor's quirks.
Queue hundreds of documents at once. The engine distributes work across available GPU resources, parallelises extraction, and reports results via a progress webhook. Perfect for month-end invoice dumps or bulk migrations from paper archives.
Before extraction even begins, the VLM classifies the incoming document: invoice, credit note, delivery note, purchase order, receipt. The correct extraction schema is selected automatically, eliminating the need for manual routing.
Extend extraction beyond Italian. English, German, French, and Spanish invoices are supported out of the box, with language auto-detection. Critical for Italian companies with international suppliers or for expanding DataUnchain into new markets.
A redesigned API with versioning, API key management, rate limiting, and proper OpenAPI documentation. Webhook callbacks support multiple endpoints per tenant. Designed so system integrators can embed DataUnchain into larger automation workflows without friction.
Two more heavyweight ERP connectors for the Italian market. TeamSystem and Zucchetti together cover the majority of Italian SMEs and accounting firms. Both connectors follow the same adapter pattern established in v2.1, keeping the architecture clean and testable.
Every extraction, correction, approval, and export is logged with timestamps, user identity, and before/after values. Designed for compliance-sensitive environments where you need to prove exactly what happened to every document.
October – December 2026. The version where DataUnchain becomes enterprise-grade. Multi-tenant architecture, a connector marketplace, hybrid OCR fallback, digital signature workflows, and analytics.
An open marketplace where third-party developers and system integrators can publish, share, and sell ERP adapters. We provide the SDK, the testing sandbox, and the distribution channel. The community builds the long tail of integrations we could never cover alone.
A single DataUnchain instance can now serve multiple organisations with complete data isolation. Each tenant gets its own extraction schemas, learning history, user roles, and API keys. Ideal for accounting firms managing dozens of client companies from one appliance.
For edge cases where our proprietary VLM cannot confidently extract a field — think severely damaged documents or handwritten notes on thermal paper — a secondary OCR engine kicks in as a fallback. The system merges both outputs and surfaces the highest-confidence result.
After extraction and validation, documents can be routed through a multi-step approval workflow and digitally signed. Supports Italian CAdES and PAdES signature formats, integrated with major Italian qualified trust service providers.
Real-time visibility into extraction volume, accuracy rates, processing times, error distributions, and per-supplier performance. Exportable charts help operations teams identify bottlenecks and justify ROI to management. Think of it as the control tower for your document pipeline.
Full documentation and controls alignment with ISO 27001 information security standards. Encryption at rest, encryption in transit, role-based access control, penetration testing, and a formal incident response plan. The goal: make procurement teams at large enterprises comfortable signing off on DataUnchain without months of back-and-forth.
Granular role-based access control with predefined roles (admin, operator, auditor, read-only) plus custom role creation. Single Sign-On via SAML 2.0 and OpenID Connect for seamless integration with corporate identity providers like Azure AD, Okta, and Google Workspace.
Kubernetes-native deployment with horizontal scaling, health checks, and automatic failover. For organisations that process thousands of documents per day and cannot afford downtime. Helm charts and Terraform modules included.
Looking further ahead. These are strategic directions, not committed features. They represent where we believe the market is going and where DataUnchain needs to be.
A managed cloud option for businesses that want DataUnchain without managing hardware. Your data stays encrypted and isolated; the VLM runs on dedicated GPU instances in European data centres. On-premise remains the default for privacy-first customers.
A purpose-built VLM trained from scratch on hundreds of thousands of Italian fiscal documents. Not a generic multilingual model with Italian bolted on, but a model that thinks in Italian accounting terminology, understands Italian tax codes natively, and handles the idiosyncrasies of Italian invoice layouts out of the box.
Direct integration with Italy's PagoPA digital payments platform. Extract payment notices, reconcile them against open invoices, and trigger payments — all within a single automated workflow. A natural extension for public administration use cases.
Pre-configured extraction schemas and adapter bundles for high-value verticals: healthcare (medical reports, lab results), logistics (bills of lading, customs declarations), construction (SAL certificates, material receipts), and legal (court filings, notarial deeds).
A partner programme for Value-Added Resellers, system integrators, and accounting software vendors. Partners get training, co-marketing support, volume licensing, and revenue sharing. The fastest path to scaling distribution across Italy and eventually across Europe.
Move beyond extraction into full document orchestration. The AI agent autonomously decides the next step: extract, validate, route for approval, match against purchase orders, flag anomalies, and push to ERP — all without human intervention except for edge cases that truly require judgment.
DataUnchain is not built in a vacuum. Every feature on this page exists because a real business told us they needed it. The best way to influence what we build next is to become an early adopter.
Early adopters get a direct communication channel with the founder and the engineering team. When you hit a limitation, we hear about it the same day. When you request a feature, it goes straight into our prioritisation matrix with real-world context attached. No feature-request black holes, no six-month silence. This is how good software gets built.
Install DataUnchain, feed it your real invoices, and let us know what breaks. Real usage data is worth more than a hundred surveys.
Every correction you make teaches the system. Every bug you report prevents it from happening to the next user. Every feature request shapes the next quarter.
Join monthly roadmap calls. Vote on feature priorities. Beta-test new capabilities before anyone else. Your name in the changelog.
A bird's-eye view of every milestone from now to 2027.
Proprietary VLM extraction, mathematical validation, Fattura Elettronica adapter, Progressive Learning v1, Docker appliance, hot-folder and REST API, multi-schema config, RPA vision engine.
Early Adopter programme, one-click onboarding wizard, fine-tuning pipeline v1, Danea Easyfatt connector, comprehensive documentation, extraction review dashboard.
Advanced auto-learning, per-supplier fine-tuning, batch processing, automatic document classification, multi-language support, REST API v2, TeamSystem and Zucchetti connectors, audit log.
Connector marketplace, multi-tenant architecture, hybrid OCR fallback, digital signature workflow, analytics dashboard, ISO 27001 alignment, RBAC and SSO, high-availability deployment.
Hybrid SaaS model, Italian-first fine-tuned VLM, PagoPA integration, vertical industry packages, VAR and reseller partnerships, agentic document workflows.
Common questions about the roadmap and what comes next.
Free installation, six months at zero cost, and a direct line to the team building the product. In return, help us validate real-world edge cases and shape what comes next.
Only five spots available. We are looking for Italian businesses that process at least 200 invoices per month and are tired of manual data entry. If that sounds like you, we should talk.