Uthereal AG · ETH Zurich spin-off · Technopark Zurich
The Swiss Palantir
Same argument about institutional data. Different answer on who ends up owning it.
Why we accept the comparison
Palantir made one argument better than anyone else, and it was the right one. The value in enterprise AI is not the model. It is the layer that connects an institution's own messy data to the decisions people actually have to make.
We build that layer. Documents, standards, case files, product data, research archives, whatever the institution knows and cannot easily reach. Cortex turns it into structured knowledge and puts it in front of the people who need it, inside the workflow they already use.
So when someone calls us the Swiss Palantir, we do not correct them. We just point out where the two stories separate.
Where the comparison stops
You own the result, not a licence to it
The knowledge layer we build is your asset. Structured, exportable, and yours to keep if you stop working with us. You are not renting access to your own institutional memory.
Swiss jurisdiction, not just a Swiss region
A data centre in Zurich owned by a company answering to another country's law is not sovereignty. We deploy on Swiss infrastructure under Swiss law, with zero data retention, and we can run fully on your own premises when the classification calls for it.
Weeks and tens of thousands, not years and millions
The classic deal starts with a long discovery, a large team on site, and a number you have to defend to a board before you know whether it works. We start with one narrow use case in production, priced so the decision does not need a board.
Model independent by design
Models get cheaper and better every few months. Your knowledge layer should outlive all of them. Cortex swaps the model underneath without rebuilding what sits on top, so you never inherit someone else's bet.
The two deals, side by side
Who owns what gets built
The vendor owns the platform. You licence the use of it.
You own the knowledge layer and the agents built on it.
Where your data is processed
Vendor cloud, region of your choice.
Swiss infrastructure under Swiss law, your tenant, or your own building.
Which model you depend on
The platform's stack, upgraded on the vendor's schedule.
Any model, swapped without rebuilding the layer above it.
Shape of the first engagement
Multi year commitment, forward deployed team, wide scope.
One use case, in production, with a fixed price and a defined end date.
If you walk away
Access ends and the work stays inside the platform.
You keep the structured knowledge, in open formats, and can run it elsewhere.
The left column describes the general shape of large enterprise AI platform contracts, not any single vendor. Terms vary by deal and by client.
What actually gets built
Three layers. Your sources stay where they are. We structure what they contain into Knowledge Objects that carry meaning and provenance, not just text fragments. Agents sit on top and answer inside the tools your people already open every day.
The whole thing runs inside a boundary you define. The model providers sit outside it and are swappable. Nothing you upload trains a shared model, and there is no retention on our side.
Your boundary: Swiss cloud, your tenant, or your building
Your sources
Documents, standards, archives, product data, case files, systems.
Knowledge Objects
Structured, versioned, traceable to source. This is the asset you own.
Agents in the workflow
Answering, drafting, checking, running multi step processes.
Access control, audit trail, and token level egress monitoring apply across all three layers.
Nothing leaves the boundary unless you have allowed it, and every crossing is logged.
Model providers
Outside the boundary. Swappable. Optional.
Built for institutions that cannot be casual about this
Regulated finance
Banks, insurers and advisors where the audit trail matters as much as the answer.
Public agencies
Where classification, residency and procurement rules decide the architecture.
Specialist publishers
STM and professional publishers turning archives into products they still own.
Professional bodies
Associations and standards organisations whose authority rests on being right.
How an engagement runs
Pick the one use case
Half a day with the people doing the work. We leave with a single question your organisation answers badly, slowly, or inconsistently today.
Agree the boundary
Where the data sits, who can reach it, what is allowed to leave, and which of your security controls we plug into. This is settled before any data moves.
Build the knowledge layer
We ingest the relevant sources and turn them into Knowledge Objects. You see the structure, correct it, and sign off on what it says.
Put it in front of real users
A small group uses it for actual work, not a demo. We measure whether it saves time and whether people trust the answers.
Decide with evidence
At the end of the first phase you have a working system and real usage data. Expanding is a decision you make on numbers, not on a roadmap slide.
Hand over the keys
Your team runs it, extends it, and keeps the underlying knowledge layer whatever happens to our relationship.
Questions we get asked
Why does Uthereal accept the Palantir comparison?
Because Palantir made the right argument: the value in enterprise AI is not the model, it is the layer connecting an institution's own data to the decisions people have to make. Uthereal builds that layer. The difference is ownership, jurisdiction, engagement size and model independence.
Who owns the knowledge layer Uthereal builds?
The client. The structured knowledge is exportable in open formats and remains usable if the client stops working with Uthereal. It is an owned asset, not a licence.
What makes this Swiss sovereignty rather than a Swiss region?
Uthereal AG is a Swiss company operating under Swiss law, with zero data retention and the option to run entirely on the client's own premises. A foreign-owned data centre located in Switzerland is not the same legal position.
How large is a first engagement?
One narrow use case taken to production, with a fixed price and a defined end date — weeks and tens of thousands rather than years and millions.
Bring us the question your institution answers badly
We will tell you in one conversation whether it is a good first use case, and what it would take to have it running. If it is not a fit, we will say so.