Azin Etemadi and Vivendur Are Rethinking Who Controls AI Memory

An AI tool drafting a customer email might need an order history. It should not automatically gain access to unrelated personnel records. For a business introducing AI into its daily work, deciding where that access begins and ends can be as important as deciding which tool to use.

That problem is central to Vivendur, an American technology startup founded by Azin Etemadi. The company is developing privacy-first AI infrastructure intended to give organizations greater authority over what their systems remember, access, and share. Its longer-term ambition is to extend that control across the different models and applications people choose to use.

“I don't believe privacy should be the price of intelligence,” Etemadi says. “The next generation of AI should be powerful enough to understand us and responsible enough to respect our boundaries.”

A perspective shaped by medicine

Etemadi trained as a physician in Iran before moving to the United States and pursuing medical research at Johns Hopkins University. Her experience in those fields shaped how she thinks about sensitive knowledge and the responsibilities of the people entrusted with it.

In medicine, personal information has a purpose. A detail shared to help someone receive care does not become appropriate for every other use simply because it is available. Etemadi brings that concern to technology, where the convenience of collecting and connecting information can obscure the choices being made on a user's behalf.

“I didn't leave medicine because I stopped caring about people,” she says. “I entered technology because I realized that some of the most important decisions affecting people's lives will increasingly be made through technology.”

Moving between countries and professions meant building a new professional identity. As an immigrant woman entering the technology industry from medicine, Etemadi brought a perspective shaped outside the conventional startup career path. The connection between her earlier work and Vivendur is a concern for individual autonomy: whether people can understand and influence what happens to information about them.

Making AI memory accountable to its users

Personalized assistance depends on context. Remembering a preference or an earlier decision can save a user from explaining the same thing repeatedly. For a business, access to relevant internal knowledge can make an AI tool more useful than one working without that background.

But remembering also creates decisions that extend beyond the original task. Should information provided for one purpose be available for another? Which application should be able to retrieve it? When several tools are involved, who determines what each one receives? These questions are central to the infrastructure Vivendur  builds.

Vivendur describes its broader vision as “sovereign AI memory.” In practical terms, this means giving individuals and organizations authority over the context their AI systems retain and the conditions under which it can be used again. The goal is for that authority to remain with users as their needs and choice of tools change.

The distinction matters because useful memory and unrestricted access are different design choices. An organization might want an assistant to remember how a project is organized while limiting access to confidential material within it. Vivendur's proposed direction is to make those distinctions part of how AI works.

Giving each tool the information it needs

The company's development approach emphasizes local-first architecture, which favors keeping sensitive information in environments controlled by the user or organization. It also prioritizes compartmentalized access: separating information according to the permissions and needs of a particular tool or task.

Returning to the customer-email example, the intended approach would allow a tool to use relevant account information without treating the entire company's records as available context. If another tool subsequently helped analyze the exchange, its access would need to be considered separately. Collaboration between applications should not automatically erase the boundaries around the underlying data.

Vivendur aims to develop the infrastructure for managing those relationships. Its focus is on the layer between sensitive knowledge and the AI tools that use it, with the objective of reducing unnecessary exposure while preserving the context needed to complete a task.

Working across models and applications

Vivendur is pursuing a model-agnostic approach, intended to support different AI models rather than tie users to a single provider. For an organization, the potential benefit is flexibility: the ability to choose tools suited to different jobs while maintaining a consistent approach to information access.

Its longer-term vision includes infrastructure through which multiple AI agents, software systems that carry out tasks, could work together. In that setting, permissions would need to remain meaningful as work moves from one agent to another. Etemadi's ambition is to support that collaboration without making every participant privy to everything the others know.

The company is at an early stage of development and plans to begin with software-based privacy and security solutions for businesses. That starting point gives the broader vision a defined focus: helping organizations use their internal knowledge while retaining authority over how it is disclosed and reused.

A founder building toward a broader ambition

In 2025, Vivendur received a Bronze Stevie Award in the Startup of the Year - Consumer Products Industries category at the Stevie Awards for Women in Business. The recognition marked an early milestone for the venture and for Etemadi's transition into technology entrepreneurship.

Her longer-term ambition extends to individuals as well as organizations. Across both settings, the principle is the same: a system becoming more familiar with its users should not leave them with less say over their own information.

For Etemadi, success would mean that people could benefit from AI that remembers them while continuing to decide where that knowledge belongs. Vivendur's work begins with the business decisions that make such control concrete: what a tool needs to know, when it may use that knowledge, and how far its access should extend.

For business inquiries or to learn more about Vivendur, contact Dr. Azin Etemadi at azinetemadi@vivendur.com.