Ebiose
Open-source framework from Bordeaux that automatically builds, tests and gradually improves AI agents for a task you define.
What it does
You describe a task and define how success is measured. Ebiose then generates AI agents, scores them with your evaluation functions and evolves the best ones further – similar to an evolutionary process. The publicly available beta (tag beta-0.1) is MIT-licensed and built on Python and LangGraph. A rewritten version with a so-called Genomic Engine, which is meant to improve how agents themselves are created, has been announced and is to be released under the MIT license.
Who it suits
Development teams and AI researchers who want to optimise agents systematically against measurable criteria instead of building them by hand. You need programming skills and your own test cases against which the agents can be measured.
Data protection for businesses
Ebiose is developed by Ebiose AI SAS, a spin-off of the French research institute Inria based in Bordeaux. According to the vendor, your data, evaluators and deployed agents stay private; proven building blocks join a shared pool only if you explicitly opt in. If you run the open-source framework yourself, the language models you connect determine where your data goes. The vendor does not state a server location, a data processing agreement or business plans.
Limits
The project is in transition: the beta is no longer developed, issues and pull requests are closed, and the code of the new version is not yet public. Access to the new version is currently (as of November 2026) by private invitation only; the public release is announced for November. This makes the tool not yet mature enough for production use.