Vitruvius

This policy explains how the AI agent in Vitruvius works with your data and what it will and will not do.

Who provides the AI

The agent uses AI by Anthropic models, accessed through Anthropic's commercial API. Requests are processed by Anthropic, likely outside the UK and EU, under its commercial terms.

Your data is not used for training

Customer data is never used to train models: not your drawings, files, briefs nor conversations, by us or by Anthropic under its commercial API terms.

What is sent

Only the minimum context needed for each request: your message, the relevant part of the model (levels, walls, openings, rooms), the brief, and short summaries or excerpts of files you attach. Text from uploaded files is clearly marked as untrusted so that instructions hidden inside documents are not followed.

What the agent can and cannot do

It can read your project and propose edits through the same drawing tools you use.
It cannot send email, browse the web or delete anything outside the project.
In Plan and Ask modes nothing changes until you approve. In Auto mode every step leaves a checkpoint you can rewind to.
Deleting more than 10 elements always needs your explicit acceptance.

Accuracy and responsibility

AI output can be wrong. Areas, compliance hints and design suggestions are advisory. Every change records who proposed it and who accepted it. You remain responsible for your drawings.

Limits and safeguards