On-premise AI for geospatial data
Ask in plain Dutch or English. Orion finds the right datasets, checks its plan against the real data and answers in minutes, with a map and a download. Or an honest “I can’t”, with the reason.
Runs on your own servers. No question or dataset leaves the building.
Orion knows what data you have, where each dataset lives and how to combine them. Anyone in your organisation can ask it, from their first day.
Today a simple question becomes a ticket, waits until the GIS team has capacity, then waits for an analyst. With Orion, the person who asks gets the answer.
Orion runs open-weight models on your own hardware. There is no cloud AI subscription and no charge per question, however many people ask.
Nothing leaves your servers. Every answer is traceable to the datasets, plan and query behind it. And what your GIS expert knows stays in the organisation when they move on.
Designed with the GDPR, the EU AI Act and government data sovereignty in mind.
A wrong number looks exactly like a right one, and it travels: into a memo, a council question, a budget. Orion treats a refusal with a reason as a good outcome, and is tested that way.
“412 buildings in the district still have label F or G. Counted from the issued-label register.”
“Do you mean buildings built before 1945, or listed as monuments?”
“The bench register doesn’t record material, so I can’t tell wooden benches apart. I can count all benches instead.”
Illustrative answers
Each star of Orion is one stage. On the live demo you watch every question travel through them.
Chat, a question about your catalogue, or a spatial question.
MeissaPlaces, groupings, dates and conditions, matched to official codes.
BetelgeuseSearches your catalogue by meaning, in Dutch and English.
BellatrixWhat is counted, what restricts it, and what it is joined to.
MintakaSimilar registers compared on measured overlap; each condition bound to a real column.
AlnilamA structured plan, never SQL, tested by more than thirty checks.
AlnitakCode turns the plan into a safe query on your own database.
SaiphThe answer, a map, a download and a note on what was counted.
RigelHomes more than 500 m from the nearest waste container.
Which neighbourhoods have no playground?
Most common energy label per neighbourhood.
Buildings that still have no energy label.
Average floor area per district of buildings without a label.
Benches with no other bench within 200 m.
The perimeter of the five largest ponds.
Square metres of green space per resident, per district.
Everything Orion connects to, today and next. Solid cards work now; dashed ones are on our roadmap.
Discover how Orion can be tailored to your own use cases.
Type a question and watch it travel through every stage: concepts lighting up, datasets falling into place, the plan drawn between them, and the result landing on a real map.
The demo needs a personal account. Request access and we’ll email you a login. We improve the reliability of its answers every day.
We run a limited number of pilots. We install Orion on your systems and you see it answer your own questions.
Followed by an optional twelve-month licence.
No implementation work on your side.
Models and data stay on your own hardware.
No commitment. Stop at any time.
Read access to a geodatabase and one workstation GPU.
The articles are in English.
Tell us about your data and the questions you want answered.
