Ebiose: when open source AI development meets natural evolution

The story of Ebiose began with a fundamental question: how can we create artificial intelligences capable of self-improvement?

Behind this thinking is a team of two people: Vincent Barbot, a serial entrepreneur, and Xabier Jaureguiberry, a doctor in machine learning. Their observation: current AI models are often limited by centralised, costly and inflexible infrastructures. What's more, AI learning is generally based on supervised approaches, requiring huge amounts of data and constant human intervention.

Inspired by biological ecosystems, where organisms interact, evolve, reproduce and adapt to their environment, the founders of Ebiose have imagined a distributed AI system where each agent learns and improves autonomously. "In addition to this evolutionary vision, we wanted to pave the way for AI designed by everyone and for everyone," explain the startup's founders. By making their technology open through an open source approach, they hope to foster collaborative innovation and enable the community to build more robust and ethical artificial intelligences. It is thanks to this vision that Ebiose has been selected and supported by Inria Startup Studio for a year.

The system developed by Ebiose is based on a set of intelligent architectural agents. Designed to create other intelligent agents, they are capable of improving themselves by interacting with their environment. Users will be able to test these agents and modify them "by hand" if necessary. "We're positioning ourselves between low-code approaches, which are easy to set up but not very customisable, and lower-level approaches such as LangChain, which enable custom agents to be created but require advanced skills," stress the co-founders. 

Where traditional models require prior training on massive data sets, Ebiose's agents use evolutionary algorithms. Inspired by the mechanisms of natural selection, these allow the best-performing agents to pass on their characteristics to subsequent generations, creating a process of continuous improvement.

By making its code accessible to everyone, Ebiose is encouraging a collaborative and transparent approach to AI development. Scientists, developers and companies can not only use the technology, but also contribute to improving it, by proposing new algorithms or optimising interactions between agents.

"Open-source is a fundamental choice," explains the team. We want our ecosystem to be a testing ground where users can implement new building blocks according to their needs: a space where AI is constantly evolving thanks to the contributions of each and every one of us".

The support provided by the Inria Startup Studio programme was essential for Ebiose: "Inria startup studio gave us the resources we needed to challenge an innovative idea and turn it into a deeptech startup," stress the founders. "The programme gave us the opportunity to structure our project, define a precise roadmap and identify commercial opportunities.

The story of Ebiose shows that an ambitious vision coupled with strategic support is the key to the emergence of disruptive innovations. Inria Startup Studio continues to support researcher-entrepreneurs ready to transform their discoveries into concrete solutions, like this startup that is redefining artificial intelligence models.

And the story doesn't end there for Ebiose! In fact, the first beta version has begun to be shared with a few developers (you can request access to it on the Ebiose Discord channel), and soon the open source version will be available to everyone! This is a major milestone for this start-up, and the excellent news that it is now part of the Unitec incubator.

With Ebiose, artificial intelligence is entering a new era, where autonomy, collaboration and continuous evolution are replacing certain rigid and centralised models. With the support of the Inria Startup Studio, the startup is well on the way to demonstrating the potential of its approach and fundamentally transforming the way we design and think about the AI agents of the future.