Soil Analysis Solutions Pricing Contact Open platform →
AI Policy

Machines that read soil, rules that keep them honest.

How Aspace Company srl builds, validates and discloses the AI systems behind its soil intelligence, under the EU Artificial Intelligence Act (Reg. EU 2024/1689) and related transparency rules in force since 2 August 2026.

1. Who is responsible

As the provider of the AI systems described here, Aspace Company srl, c/o DTA Headquarters, Brindisi (BR), Italy · VAT and tax code 02759770742 · REA BR-252874, is responsible for how they are designed, validated and labelled. Questions about our AI go to info@aspace.company; questions about personal data go to privacy@aspace.company.

2. What this notice covers

This notice applies to every AI system ASPACE operates, on the public website, on the ASPACE platform and API, and inside partner products we power (including Village Chief). Personal data handling is covered separately by the Privacy Policy; contractual use by the Terms of Service and each customer's Data Processing Agreement.

3. Where AI runs in ASPACE

Soil intelligence models
Estimate soil properties (pH, organic carbon, nutrients, texture, moisture indices) and land evaluation scores from satellite imagery and field data. Every property model is validated against a compiled reference frame of 345,000+ laboratory-measured soil points across 186 countries before it ships.
Yield engine
Predicts crop yield per field, calibrated against measured harvests reported by our customers. Minimum three ground-truth labels before any calibration applies.
Crop disease identification
Classifies likely disease from a photo of a leaf or plant, against a published crop taxonomy. A probability and the model's confidence accompany every result.
Ask aspace assistant
The chat surface in the dashboard ("Ask aspace") answers questions from the customer's own field, task, alert and market data. Version 5D-A1.0 is a deterministic rules engine: it matches intent and queries live records — no generative model writes the answer, and every reply carries its data sources.
Narrative generation
Some reports and regional bulletins include commentary drafted by a language model (self-hosted open models, or a commercial provider under a processing contract). Numeric claims in those sections always come from the underlying analysis, never from the model.

4. Classification under the EU AI Act

We have assessed each system against Regulation (EU) 2024/1689. None of them falls within the prohibited practices of Article 5, and none falls within the high-risk categories of Annex III — our systems advise on soil and crops; they are not safety components of regulated products, do not determine access to services, and make no decisions about people.

They are limited-risk systems subject to the transparency duties of Article 50, in force since 2 August 2026. That is why this notice exists, and why the assistant and AI-drafted content are labelled wherever they appear.

5. Transparency in practice

  • You are told when you talk to AI. The Ask aspace composer identifies itself as an AI assistant and shows the model version in every conversation.
  • AI-drafted text is marked. Report and bulletin sections written by a language model are identified in the document. We are phasing in machine-readable marking of AI-generated content as the technical standards under Article 50(2) settle.
  • Provenance travels with the answer. Assistant replies list the records they used; analytical outputs carry the date of the imagery and the validation basis.
  • Frozen versions are stamped. Every output carries a 5D- frozen version (for example 5D-2.4 for the current soil pipeline, 5D-A1.0 for the assistant). A published version never changes — improvements ship under a new number, so any result can be traced to the exact pipeline that produced it.
  • Demo keys are time-bound and disclosed. Spectral demo access is metered and expires; nothing in a demo is presented as a customer result.

6. Limits and human oversight

AI outputs are estimates that support agronomic judgement — they do not replace a qualified agronomist or laboratory analysis where a law, certification scheme or public programme requires one. Customers stay in the loop: field boundary edits, planting records and harvest reports are human-entered, and every recommendation can be reviewed against its sources before acting.

If an AI output looks wrong, contest it: write to info@aspace.company with the field and date, and we will re-check the analysis against the source data.

7. Data our AI uses

Soil models train on aggregated, laboratory-measured soil datasets and satellite imagery — not on your personal data. Assistant conversations belong to the account that created them, are not used to train models for other customers, and inherit the retention and access rules of the platform's Data Processing Agreement. Harvest yields you report calibrate your own workspace; aggregate statistics may improve the shared calibration set in anonymised form.

8. AI literacy

In line with Article 4 of the AI Act, staff who build, deploy or support these systems receive training on their capabilities, their limits and the claims we may and may not make about them.

9. Changes and contact

We update this notice when our systems, their classification or the law change. Material changes are announced on this page with a new version number. For anything about our AI — including requests to review or contest an output — write to info@aspace.company.