Show you what it means
Ask about the storehouse and the host lights up that exact building on the model while it talks. Ask about the crevasse field and the ice glows where the cracks are. It points at the real thing, not at a pop-up box.
How it works
Interactive stories that answer questions
Take a real place, turn it into a 3D model you can walk around in, and add a guide you can ask anything about it. This page explains how we make that work, and why you can trust the answers. No jargon needed.
Step one
Start with a 3D model of the place
A 3D scan is like a photo you can walk into. It shows every wall, stone and crack exactly as they are. But the scan does not know what any of it is.
When you scan a farm or a glacier, you get millions of tiny points and triangles. Together they look just like the real thing, down to the millimetre. To a computer, though, it is only shapes. It cannot tell a barn from a boulder, or a runestone from a rock.
That is fine if all you want to do is look around. It is not enough if you want to ask, “What is that building?” and get a real answer. For that, the model needs to know what its parts are.
Step two
We put name tags on the important parts
Imagine sticking notes onto the model: this building is the storehouse, this dark band across the ice is the crevasse field, these carvings are the runes. The technical name for this is semantic tagging. It just means telling the computer what things are.
Every name tag holds three things: the name, a sentence about it, and where the facts come from. Maybe a museum catalogue, a research report, a drawing from 1747. The tags are written by people who actually know the place: a curator, an archaeologist, an engineer. We never let a computer guess them from the picture.
Tags can be tiny or huge. One rune carved into a stone is a tag. A field of cracks across a whole glacier is a tag. The same idea works for both, which is why one engine can handle a runestone, a farm, a temple and a power plant.
Once the tags are in place, the model is no longer just shapes. It is a place full of things with names and stories. And that is what the guide needs.
Step three
Now the AI host can do things, not just talk
The host is an AI guide that lives inside the tour. Because the parts have names, it can act in the 3D world while it answers you. Four things it can do, all live in the demos.
Take you there
A tour has a route with stops, like chapters. But you are never stuck on it. Say “show me the gate” and the camera flies there, even if no stop goes near it. You lead, the host follows.
See what you see
Ask “what am I looking at?” and the host first checks which tagged things are on your screen right now, and how big they are in view. You can also just tap a building to ask about it.
Only say what it knows
Every answer comes from the name tags and the texts behind them. You can always tap to see the source. If you ask about something that is not there, the host says it does not know. It works in Swedish and English, typed or spoken.
Why you can trust it
We do not let the AI guess
There is a shortcut we could take: point an AI at the model and let it guess what everything is. We do not, on purpose. A guess cannot tell you where it came from. For a museum, a school or a power company, an answer with no source is not really an answer.
Real places are full of details that people disagree about. A storehouse might be “from the 1500s” by tradition, while the timber tests say the trees were cut in 1484. A field.tours story tells you both instead of quietly picking one. The host does the same: it only repeats what the people behind the place wrote down and backed up.
So the host has limits, and that is the point. Asked about something outside its material, it says “that is not in my sources” rather than making something up. That honesty is what makes a museum comfortable putting an AI guide on its own website, under its own name.
Who it is for
Places you cannot easily visit
Museums use it for things that never fit on a wall label: a runestone too worn to read, a farm as it stood in 1850, a temple on the other side of the world, a glacier no one can walk up to. Scanned and tagged, each one becomes a place you can explore and question, on the museum's website, on a touch screen in the gallery, or in a VR headset.
The museum's own staff decide the stops, the texts and the name tags; they are content, not code. An editor for staff to make the changes themselves is on the way; until then we make them for you. The tour speaks Swedish and English. The host can answer a school class one way and an expert another. And it is built for a keyboard, a screen reader such as VoiceOver or motion turned down, and the host can be used by voice alone.
The same engine works for a power plant behind its fence or a factory that never stops for visitors. A turbine hall gets name tags just like a timber storehouse does. The sources are technical manuals instead of museum records, and the visitors are customers and school groups instead of gallery guests.
Compared
How it differs from photo tours and videos
Photo tours, videos, slideshowsA field.tours story
You click through picturesYou talk to a guide, by typing or speaking, in Swedish or English
Pop-up boxes with textThe guide lights up the real building or object on the model
Fixed viewpointsThe camera can go anywhere you ask
Nothing knows what you are looking atThe guide checks what is on your screen; tap a thing to ask about it
A narrator you have to take at their wordEvery answer has a source you can tap
Made for a desktop screenPhone, touch screen and VR headset, from the same tour
Questions
The questions we hear first
What does semantic tagging mean?
Giving the parts of a 3D model names, like sticky notes: this building is the storehouse, this crack in the ice is the crevasse field. Each note also says where the facts come from. That is what lets the AI host point at things, fly to them and light them up.
Does the AI host make things up?
It is built not to. The host only uses what the people behind the place wrote down and backed with sources, and every answer names where it comes from, one tap away. Ask about something that is not there and it says it is not in its sources instead of guessing. It is still a language model and can get something wrong - that is what this beta tests, and the sources let you check any answer yourself.
Do we need a 3D scan before we start?
No. You can use 3D material you already have (scans, point clouds, photogrammetry, BIM or CAD files), we can scan the place with you, or a scanning company captures it and we take it from there.
Which languages does a tour speak?
Swedish and English from the start, more on request. You can ask out loud with your voice, and choose who the answers are for: school pupils, the general public or experts.
Where does the tour run?
In a normal browser on a phone or a computer, full screen on a touch display in a museum or visitor centre, and in a VR headset. One tour works everywhere, and there is nothing to download.
Who writes the texts and the name tags?
The people who know the place: a museum's own staff, a researcher, an engineer. In a Standard tour the stops, texts, tags and sources are theirs: an editor for staff to change them themselves is on the way, and until then we make the changes for them. A Custom story is written together with us, and the texts stay yours to change afterwards.
How is this different from a 360° photo tour?
A 360° tour is a set of photos you click between. Here the whole place is a real 3D model, so the camera can go anywhere, the host can light up real things in it, and you can tap something to ask about it. The tour is a route through a place, not a stack of frozen pictures.
What does it cost?
There is no price list. We plan a first project together with each organisation. Right now we are looking for pilot partners: a handful of organisations that test a tour on a real place and shape the offer with us.
See it live
Reading about it is the slow way to get it.
The demos let you try everything on this page: a farm at Skansen and a runestone, host and all.