Inria spin-off

Self-evolving
agent factory.

You define the task and what success looks like. Ebiose builds, evaluates and evolves AI agents to meet it.

Try Ebiose Watch the video 3 min
  • MIT core
  • Your models
  • Your code
  • Your stack
forge-saas-support.pyIllustrative run
forge = Forge("saas-support")
forge.brief = "Help customers with their SaaS account and subscription."
forge.tools = [crm, documentation]

forge.cases = load("support-cases.jsonl")
forge.evaluator = Judge.from_brief(forge.brief, forge.cases)

$ uv run python forge-saas-support.py --budget 10

Reference agent score · 55%

Generation 10 · 25 agents created · best score 89%

Cycle complete · $9.80 spent

Best agent exported ./saas-support

From your mission
to your agent.

From your mission to your agent. You provide a mission and a budget. Inside Ebiose, agents are built and evaluated; the best are kept and evolved into the next generation, and the loop runs again. When the budget is reached, you receive your agent, ready to deploy. From your mission to your agent. You provide a mission and a budget. Inside Ebiose, agents are built and evaluated; the best are kept and evolved into the next generation, and the loop runs again. When the budget is reached, you receive your agent, ready to deploy.
sec 02  the film

Inside the factory, in three minutes.

3 min EN

The simulator

Support agentSupport the customers of a SaaS product.
Illustrative scenario

I’ll build an agent for your customer support.

sec 04  recursion

Every run improves the factory.
Including yours.

Each forge builds your agent and improves what comes next.

forge yours

shared

  1. 01

    The best architects stay.

    Architects behind the best agents build everyone’s next forge.

  2. 02

    Proven genes travel.

    Successful genes join the shared pool only if you opt in.

  3. 03

    Better foundations for everyone.

    Part of every cycle runs foundational forges, creating better architects and reusable genes, open to all.

everyone’s next forge

never shared

  • your data
  • your evaluators
  • your deployed agents

We are at the very beginning. Today, Ebiose runs in private tests. The public release is planned for November, and that is when the shared pool really starts.

Take part early
sec 05 values

AI for all.
Built together.

we believe

  • 01 Open AI should be built in the open.
  • 02 Ownership AI you depend on should be AI you own.
  • 03 Common AI is a shared effort. Its benefits should be shared too.
  • 04 Sustainable AI should earn its place: useful and worth what it costs.

The manifesto goes further: shared knowledge, shared compute, communities gathered around the challenges they care about. That is where we want to take this.

Read the manifesto
CONTRIBUTING.md

Build with us.

Now

  • Genomize nodes & agents

    Bring models beyond LLMs (tabular, decision) and agent frameworks like CrewAI into the genome.

  • Foundational forges

    Create reusable genes for other agents.

  • Evaluation forges

    Build agents that judge and help evaluate other agents.

  • Challenges

    Bring real problems to explore.

  • Community

    Write guides, share examples, help others.

Coming later

  • Domain expertise

    Guide agents with what you know.

  • Shared resources

    Contribute compute, data or models.

Start contributing (opens in a new tab)Join the Discord (opens in a new tab)
sec 06  start

Where do you want
to start?

for builders

Try Ebiose

Request an invitation
for companies

Discuss a case

Book a call
for contributors

Contribute

Join the Discord (opens in a new tab)

backed by

  • Inria
  • Inria Startup Studio
  • Bpifrance
  • Unitec