AI Usage Policy

How AI is used in Pulsor Suite, how customer data is treated and how outputs should be used.

Draft document, awaiting legal review. It describes how Pulsor Suite actually works but has no contractual value until the reviewed version is published. Items in square brackets are still to be completed.

1. Models and providers

Pulsor Suite is multi-model: each Customer's deployment uses the AI provider chosen by Pulsor or the Customer among those supported (OpenAI, Anthropic, Google, Mistral, DeepSeek, Groq) or a model hosted in the Customer's own infrastructure. The provider in use is visible to administrators. Search embeddings can run on a local model.

2. What is sent to a model

only the excerpts the requesting User is allowed to read, selected after the rights check, plus the question;
never documents outside the User's rights, never the whole document base;
the instructions given to the model forbid following orders hidden in documents.

3. Use of customer data

Customer Data is not used to train shared AI models of Pulsor;
with third-party providers, Pulsor uses their business offers whose terms exclude training on submitted data [to be checked for each provider used];
questions and answers are kept in the Customer's history, which its administrators control.

4. Human oversight

every answer cites its sources so the User can check them;
when the documents do not contain the answer, the assistant says so instead of inventing;
AI agents that act can be placed behind an approval, and a requester cannot approve their own run unless the Customer allows it.

5. Limitations of AI outputs

Answers can be incomplete, outdated if the documents are, or wrong. They do not replace professional advice (legal, medical, financial). Users remain responsible for what they do with them.

6. Prohibited uses

See the AI section of the Acceptable Use Policy: no hidden instructions, no automated decisions with legal effect on a person without human review.