ANSWERS TO
THE REAL QUESTIONS

It reads your trade documents and returns structured output. Feed it a carrier invoice, bill of lading, customs entry package, or rate quote — and it extracts the relevant data, cross-references documents against each other and against your contracts, and flags issues. The output is a structured report your team can act on. It handles the document intelligence layer your team is currently doing manually.

Custom-built for your operation. Every system starts from a discovery session where we map your specific document types — your carrier invoice formats, your trade lanes, your commodity codes. The model is fine-tuned on your actual documents, not trained on generic freight industry data. A freight broker and a 3PL get different builds. An importer moving consumer goods and an industrial materials exporter get different builds. There's no template being reskinned.

PDFs (including scanned documents), Excel and CSV files, Word documents, and plain text. For logistics: freight invoices, bills of lading, commercial invoices, packing lists, certificates of origin, customs entries (commercial invoice, packing list, CMR, SAD), rate confirmations, carrier statements, booking confirmations. Multi-language support is included for operations handling supplier documents in languages other than English.

Integration approach depends on your TMS. For most platforms, we build a simple file-drop workflow — documents exported from your TMS are placed in a watched folder, the AI processes them, and the report goes back into the same folder or an output location your team uses. Direct API integration is available for systems with accessible APIs. We establish the integration method during the discovery session.

You receive the full source code, all model weights, the complete system documentation, and a handover session with your team. You are not buying a licence to use software we retain. You own the build outright. If Marapone shut down tomorrow, your system keeps running exactly as it does today. You can modify it, extend it, or have another developer work on it. There are no usage restrictions after purchase.

Four tiers: Starter ($1,500 CAD), Pilot ($4,900 CAD), Full Build ($9,500 CAD), and Plus ($15,000 CAD). Starter covers one workflow module. Pilot covers two modules. Full Build is the complete suite. Plus adds server/air-gapped deployment, extended team onboarding, and priority support. Exact scope and price confirmed after the free assessment. See the full pricing page for a detailed breakdown.

No mandatory ongoing costs. The system runs on your hardware with no external dependencies — no API keys, no cloud subscription, no usage fees. Optional support retainer is available at $499 CAD/month or $3,500 CAD/year for updates when your document formats or trade lanes change.

Documents you send for fine-tuning are processed in our secure build environment and deleted after the build is verified and handed over. They are never stored long-term, never used to train other clients' systems, and never transmitted to third parties. Once deployed, the system runs on your hardware — documents you feed into it after deployment never leave your network.

For most logistics builds, a modern business workstation or server is sufficient. Specific requirements depend on your document volume and the number of modules. We confirm hardware requirements during the scoping session. Most operations don't need dedicated GPU servers — logistics document processing runs efficiently on standard server hardware. Air-gapped deployment is available for high-security environments.

Yes, by design. Because the system runs on your infrastructure with no third-party data calls, GDPR compliance becomes a question of your hardware and your data-handling policies — not a vendor's. We can deploy with EU-resident processing only (our Rome office handles EU builds end to end), provide a Data Processing Agreement (DPA), and the system supports right-to-be-forgotten via standard file-system controls. We've shipped builds for clients with strict GDPR posture; details vary, so we cover specifics during scoping.

You choose. Default is Canadian processing via our Toronto build environment. EU clients can require Rome-only processing — documents you send for fine-tuning never leave EU jurisdiction. For sensitive operations (defence, regulated commodities, classified shipments), we offer on-prem build sessions where Marapone engineers work inside your network and no documents are exported. After handover, processing location is whatever you choose — usually your own server.

The system writes structured logs locally: timestamp, document hash, module invoked, output summary, user who ran it. Logs stay on your machine — they never leave your network. Retention is configurable (default: 12 months, then rolling deletion). For SOC 2 or ISO-aligned operations, we ship the build with logging hooks compatible with Splunk, Datadog, or your existing SIEM. The audit trail is yours to query, forward, or purge.

Customs documents (entry summaries, B3s, EU NCTS declarations, broker filings) are treated as your most sensitive data class. The model is fine-tuned on your customs operation's actual filings; the model weights live on your hardware; no part of the customs corpus is ever transmitted off your network after deployment. For brokers regulated under CBSA's CARM, CBP's ACE, or EU NCTS, the system is configurable to mirror your retention and access-control requirements (typically 6 years for CA/US, varies in EU).

Documents you send for fine-tuning during the build are deleted from our environment within 30 days of system handover — verified by signed checksum. You can request immediate deletion (within 24 hours of handover) by email and we'll send a deletion confirmation. The fine-tuned model itself doesn't memorize specific documents — it learns the patterns of your document language. If you ever discover a document was sent in error, we re-run the deletion process and confirm in writing.

Locally. The model runs entirely on your hardware with no external network calls after deployment. There are no API keys, no cloud endpoints, and no usage-based billing. The system is fully self-contained — it processes documents on your machine or server and returns output to your local environment. Internet access is not required for operation.

Yes. Scanned PDFs and image files (JPG, PNG, TIFF) are processed using an embedded OCR layer — Tesseract and AWS Textract (on-prem or private cloud) depending on your infrastructure preference. Output accuracy on clean scans is comparable to native PDF processing. Very low-quality or heavily degraded scans may require a pre-processing step, which we flag during scoping.

Full source code (Python/FastAPI backend, any frontend we build), fine-tuned model weights in standard format (GGUF or safetensors), a complete technical README covering setup, configuration, and operation, and a live onboarding session with your team. If we build custom integrations (file-watch scripts, TMS bridges, email parsers), those are included. Everything needed to run, modify, and maintain the system without us.

Yes. If your trade lanes, document formats, or compliance rules change, the model can be re-fine-tuned. You can do this yourself using the training scripts we include, or engage us for an update under the optional support retainer ($499 CAD/month or $3,500 CAD/year). Most operations go 12–18 months before needing a material update; minor format changes are handled by prompt adjustments without re-training.