EU-sovereign AI
Every layer that holds what we know is ours, and it runs in the EU: the graphs, the ledger, the agents and the infrastructure they run on. Language models are parts we can replace.
If a model provider changes its price, its access or its behaviour tomorrow, what does your company still know?
Knowledge kept inside a model account, a prompt history or a vendor’s database can change hands with a change of terms. The risk is structural. It does not depend on anyone’s intentions.
We built the other way round. What our systems learn is kept in our own graphs and our own ledger, in formats we define, on infrastructure we control in the EU. Models do narrow, checked tasks, and a model can be replaced while the knowledge stays.
European by design
EU-sovereign, and GDPR compliant.
Sovereignty also means knowing whose rules apply. For us, that is EU law.
- Based in the EUA European company.
Our home is Stockholm, Sweden, and we work under EU law.
- Run in the EUThe core runs in the EU.
The graphs, the ledger and the agents run on infrastructure we control in the EU.
- GDPRGDPR compliant.
Personal data is handled as the GDPR requires: for a clear purpose, limited to what that purpose needs, and with every person’s rights respected.
Sovereignty is a loop we keep running.
Own the core, record the evidence, choose the models, run it ourselves in the EU and learn from every outcome.
Own the core.
The graphs, the ledger and the attribute language are our code and our data, in formats we define.
- No data model is rented from a vendor.
- The whole graph can be copied, audited and moved.
- Nothing we have learned is silently lost.
Record the evidence.
Every fact enters with its source, its date and a grade for how strongly it is supported.
- What was seen is kept apart from what was inferred.
- A change is reported only once it is confirmed.
- Every source is used only as its terms allow.
Choose the models.
Language models do narrow, checked tasks, and no single provider is essential to us.
- A model is proven on a task before it works at scale.
- What a model infers is labelled as inferred, never as seen.
- A model can be swapped without losing what was learned.
Run it ourselves.
The graph, the ledger and the agents run on infrastructure we control, in the EU.
- No vendor holds the graph.
- Every change can be traced and reversed.
- Every run leaves an audit trail.
Learn from the outcome.
Outcomes are written back, so each decision starts from more evidence than the last.
- Our own calls are scored against what happened.
- Gaps stay visible as unknown, never counted as zero.
- We know where our coverage is strong, and where it is thin.
Proprietary by design.
Five layers we wrote ourselves. Each one is documented, can be inspected, and is ours.
- 01/Any source, one graph.
Players, relationships and signals, joined in one graph we own.
Intelligence Graph - 02/The industry, modelled.
Actors, workflows, constraints and decisions, in a format we define.
Product Graph - 03/One language.
Attributes let every vertical share the same intelligence core.
Attributes - 04/Agents you can audit.
Every action is logged and every output is graded.
Agents - 05/Proof with every fact.
Source, date and grade travel with every fact.
Evidence Ledger
What we own, and what we only use.
Sovereignty claims are cheap. This is the list.
| Layer | Status | What that means |
|---|---|---|
| Knowledge graphs and the attribute language | Owned | Our code, our data, our formats. |
| Evidence ledger | Owned | Ours, on infrastructure we control in the EU. |
| Agents and the checks they run under | Owned | Written and maintained by us. |
| Infrastructure that runs the core | Owned | Operated by us in the EU, not a rented data platform. |
| Language models | Used | Called for narrow tasks, and replaceable. What they infer is labelled as inferred. |
| External sources | Used | Used only as each source’s terms allow. |
Bring a market. Keep the knowledge.
We build vertical products with partners on an EU-sovereign core that no model vendor holds.