AI Charter
Document published by Vitruhome, a simplified joint-stock company (RCS Salon-de-Provence 994 224 715), operating the “Vitru’home” brand. See the legal notice.
This English translation is provided for convenience only. The French version is the legally binding one and prevails in case of discrepancy.
This Charter documents the artificial intelligence systems of which Vitru'home is the provider within the meaning of Article 3, point 3, of Regulation (EU) 2024/1689 (the “AI Act”), and which it makes available in its suite of applications under its own brand.
AI Act classification. The AI systems described below are classified as limited risk within the meaning of the Regulation and do not constitute high-risk AI systems (Annex III). They are subject to the transparency obligations laid down in Article 50 of the Regulation, applicable from 2 August 2026.
High-risk systems. The obligations applicable to the high-risk AI systems of Annex III have been postponed to 2 December 2027 by Regulation (EU) 2026/1744, and those relating to systems integrated into the products of Annex I to 2 August 2028. This postponement does not alter Article 50. Our analysis concludes that no Vitru'home system falls within these categories; it is reasoned in writing and is being validated by our legal counsel.
“No training” commitment. Your data and documents are not used to train the models of third-party providers, nor by Vitru'home. This commitment relies on the Zero Data Retention requested from the Vercel AI Gateway on every call. For the processing of personal data sent to our AI systems, see the Privacy Policy.
AI chat
Description. The Vitru'home AI chat is an AI system intended for conversational assistance to users of the platform. It is built on general-purpose AI models supplied by third parties (Gemini, Google; Claude, Anthropic). Vitru'home is the provider of this system; it does not itself supply any general-purpose AI model.
Models used. A single inference route is used: Gemini (Google) or Claude (Anthropic), called through the Vercel AI Gateway with inference pinned to the European Union region and zero retention requested on every call (text vectorisation, Amazon Titan, goes through the same gateway with zero retention, without a selectable region). Provider documentation: Anthropic model card.
Intended use. Conversational assistance, document summaries, drafting support, answers to questions about the platform and about the data you provide to it.
Known limitations. Answers may contain errors, inaccuracies or hallucinations. They are a decision-support aid and do not replace the judgement of a qualified professional (legal, technical, financial).
Automated analyses
Description. The Vitru'home automated analyses are specialised AI systems intended for the aggregation and interpretation of data relating to a site or a construction project. Two services are covered: site analysis (aggregation of public data, natural and technological hazards, demographics, infrastructure, from 39 external APIs) and compliance analysis (automated assessment of criteria drawn from the HQE and BREEAM standards).
Models used. A single inference route is used: Gemini (Google) or Claude (Anthropic), called through the Vercel AI Gateway with inference pinned to the European Union region and zero retention requested on every call (text vectorisation, Amazon Titan, goes through the same gateway with zero retention, without a selectable region). Provider documentation: Anthropic model card.
Intended use. Decision support for AMOs (owner’s project consultants) and construction or urban planning professionals: preliminary assessment of a site, compliance checks, generation of summary reports.
Known limitations. The analyses rely on the quality and freshness of the aggregated public data. They do not replace a site survey, legal advice, a technical inspection or consultation with a qualified engineering consultancy. Investment or construction decisions must not be taken solely on the basis of these analyses.
Vision and OCR
Description. The Vitru'home Vision and OCR system is a specialised AI system intended for the extraction of textual and structural data from PDF documents and architectural plans provided by the user.
Models used. A single inference route is used: Gemini (Google) or Claude (Anthropic), called through the Vercel AI Gateway with inference pinned to the European Union region and zero retention requested on every call (text vectorisation, Amazon Titan, goes through the same gateway with zero retention, without a selectable region). Provider documentation: Anthropic model card. These language models extract text and structured data from PDF documents. Proprietary computer vision models trained by Vitru'home detect architectural elements on plans. These models are trained on internal datasets made up of building plans, supplemented by annotations produced by our teams. The documents and plans you upload to the platform are never used to train or retrain our models: they remain in your storage space, which the training pipeline does not access.
Intended use. Automated extraction of text and structural elements (walls, doors, windows, furniture) from plans and documents provided by the user, in order to feed the analyses.
Known limitations. The quality of the extraction depends on the quality of the source document (resolution, contrast, standardisation). Elements may be omitted, misclassified or mispositioned. The result does not replace human verification before operational use.
Human oversight and redress
The outputs of Vitru'home AI systems are a decision-support aid and do not produce any legal or similarly significant decision in a fully automated manner (Article 22 of the GDPR). Final responsibility for the use of the results lies with the professional user.
You may report a problematic output, ask a question about how an AI system works, or exercise a right relating to this processing by writing to team@vitruhome.com.
Last updated: 26 August 2026