Skip to why this matters
EuroAgent.Coming soon

EuroAgent, the open agent stack for European growth.

Adopting AI agents today means a single vendor's bundle (your data in, your IP out) or a pile of open source you assemble yourself. EuroAgent is the third way: one enterprise-ready stack where agents come to your data. Start now, own the program your company runs on, and keep every vendor replaceable. Including us.

A connected European horizonA minimalist line drawing of connected European landmarks.

Why this matters

“We spent 50 years making computers programmable. Now we're making companies programmable.”

Hiten Shah— Hiten Shah

He's right. And when your company becomes programmable, the program becomes your most valuable IP. These four things make it compound for you, not for a vendor.

Start now

Agents go to your data. No data-transformation programme before the first result.

Compounds as yours

What your agents learn becomes your IP, not a vendor's training data.

Real choice

Every layer stays exchangeable, so every vendor stays replaceable, and prices stay honest.

Collaborative scale

Start in-house. When it pays, collaborate with partners and peers, without pooling your data.

Five layers. Two ways to run every workload.

The fitting-together is already done: one enterprise-ready stack, every layer tested with every other, so choosing vendors never means integrating them. And because agents go to the data, the data stays where it is.

Layers
Fully sovereignCustomer-controlled end to end
Pragmatic fitSovereign where required. Global where useful.

Agents and workspaces

Fully sovereignSelf-hosted agents, in federations you control
Pragmatic fitYour own and approved third-party agents, managed together

Apps and data

Fully sovereignSelf-hosted apps and connectors. Data stays inside your boundary.
Pragmatic fitExisting apps stay in place. Only approved context travels.

Models

Fully sovereignOpen-weight models, self-hosted, including your own fine-tunes
Pragmatic fitOpen and frontier models, routed by task, risk, quality and cost

Compute

Fully sovereignOn-prem, private or sovereign EU cloud
Pragmatic fitHybrid and multicloud. Workloads move; data stays.

Hardware

Fully sovereignServers and trusted accelerators, under your control
Pragmatic fitCPUs, GPUs and edge devices, chosen per workload
Flower runtime · SuperGridOne common contract for agents, connectors, federations, model access, state and execution, on both paths.
Agents
Connectors
Federation
Model
State
Execution
Your program
Agent configurations · workflows · skills · evaluations · policiesCustomer-controlled. Reusable. Compounding over time.

Choose per workload. Change without rebuilding.

What this replaces: the single-vendor bundle. Convenient, until the levers appear (pricing, packaging and egress set by the vendor once your data and workflows live inside), and your program compounds as theirs.

And the do-it-yourself pile: all the right pieces, but you integrate them, maintain them, and re-integrate them every time the field moves.

The reference architecture keeps both promises: pre-integrated like a bundle, exchangeable like open source, and the program stays yours. Partner seats in every layer are announced at launch; every vendor stays replaceable. Including us.

4 min read

Growth runs on choice.

Every market we admire got better the same way: competition and real alternatives pushed prices down, quality up, and growth on. Nobody argues the opposite. Yet most European organisations are settling into AI with two or three vertically integrated providers. Not from conviction, from convenience.

That convenience has a price schedule. First everything is generous: subsidised, integrated, easy. Then your data and workflows live inside, and the levers appear (pricing, packaging, egress), with a roadmap written for the vendor's biggest customers, not for you. Until now the only alternative was a pile of open source you assemble yourself, behind a data-transformation programme measured in years. Enterprises have time for neither.

So we built the third way: the convenience of a single-vendor stack, with the flexibility and choice of open source. One enterprise-ready agent stack, fitted together and tested end to end. It just works. And because agents go to the data (on the mainframe, in the factory, in the cloud you already use), you start now, and any migration happens in the background, never as a prerequisite.

Read that as an operator. Speed: value in weeks, because modernisation stops being a gate. Ownership: a programmable company runs on its own playbook (workflows, agent skills, everything your agents learn), and on this stack that program compounds as your IP, your margin, your moat. On a rented stack, it compounds as someone else's. Leverage: when workloads can move, prices stay honest; you keep negotiating power over every vendor. Including us.

None of this requires joining a network: most companies start with agents entirely in-house, on their own data, for their own workflows. But the same runtime that connects your layers can, when it pays, connect you outward: to vendors, customers and peers, collaborating without pooling data. That is how smaller companies get big-company scale without the merger, and it answers the constraint the Draghi report priced: Europe's fragmentation costs scale, but it is also differentiation. Agents return the scale. You keep the difference. Scale without sameness.

Europe has the parts: frontier models, the world's open-model shelf, independent clouds, hardware people love, and enterprises whose know-how agents multiply. That is why we have come together as partners (model makers, cloud operators, hardware builders, agent companies), not to help Europe regulate, but to help Europe grow. We cooperate on one open framework; everyone competes on what they build with it.

EuroAgent launches this autumn with the reference stack, partner showcases, and hackathons across Europe. The test for joining is one question: does more choice make your business bigger? Then you're one of us.

— The EuroAgent alliance

Take part in what comes next.