Skip to Content

What we are doing with AI, and where the data stays

Odoo apps built with agents, and the rule that comes before them: your data stays where it is.

Table of contents

No titles in the blog post

Sooner or later somebody in the company pastes a bill of materials into a chatbot to get a hand with it. That is the moment the bill of materials leaves the company, and usually nobody notices.

We are not going to tell you to stop. We are telling you where the line is, and what we are doing about it.

What we are actually doing

Since April 2026 Andrea has been doing part-time research at the Rizzo AI Academy: Odoo apps built with agents, on real company projects rather than demos.

One rule comes before the rest: the data stays where it is. Either the model runs on your own machines, or it runs somewhere else but the text is stripped of anything that identifies you before it goes out. Everything else is a consequence of that.

Two pieces you can go and check

rizzo-pii is a local-first privacy guard: it recognises names, addresses, tax codes and the rest, and replaces them before the text is handed to a language model. MIT licence, over a thousand stars and seventy-one forks on GitHub as of September 2026. Andrea's contribution, accepted on 8 August 2026, is the docker-compose file for the web application container, which is what lets it run on a machine of yours rather than on someone else's.

mcp-odoo-panda is a bridge that lets an assistant read and write Odoo records directly, instead of you copying them out by hand. MIT licence as well. Andrea's contribution, accepted on 30 April 2025, is the documentation for the JSON-RPC mode and for Odoo 18.

Both of them are somebody else's project. That is how open source works, and it is also why the work is checkable: the commits are there with a date on them.

If somebody in the company has already started pasting things into a chatbot, it is worth talking about it before it becomes a habit. Write to us.