Three classes of content
Classification is by the nature of the content, never by the customer — today the real projects have names, in six months they will have different ones. What the class determines is concrete: which source you need, which anchoring strategy applies, and where the cost is going to sit.
Animated guide
A character present in the place, who tells you where you are and answers what you ask.
The visitor raises the phone and someone is there: a guide who speaks, gestures, points at what is being described and follows along as you walk. The conversation is not improvised — it runs over a curated knowledge base of points of interest, the same engine already in production on Lares, so the guide answers about this room rather than about rooms in general. The character walks with you through steering behaviour computed locally, so nothing in the movement waits on the network.
- What it needs
- A mesh of the venue or of the reconstructed subject, a rigged character with an animation library, and a curated set of points of interest with positions and text. Here the content is the expensive part, not the software — and no infrastructure reduces the cost of a historically grounded model.
- What it produces
- An app that places the guide in the real venue and keeps it there while the visitor moves around, answering out loud and pointing at real positions.
- Who it is for
- Museums, archaeological sites, historic houses, visitor centres — guided tours of places that no longer look the way they once did.
- Anchoring
- Geospatial / VPS outdoors, assisted manual alignment on existing physical references indoors. The tap belongs to another class: the visitor does not know where the wall was.
Showroom configurator
The product just configured, at full scale, in the room where it will actually go.
The AR app is a view onto the configurator’s state: change the finish, the size or the variant and the object in the room changes with it. This is the class the field measurements confirmed with the widest margin — a couple of centimetres of drift are irrelevant when the question being asked is whether it fits in the living room and whether that oak works next to that floor.
- What it needs
- A catalogue of meshes authored to a declared naming convention, which is what makes variants addressable, plus a connection to the configurator that owns the state. It does not need a CAD pipeline — believing otherwise is how you end up paying several thousand a year to solve a naming problem.
- What it produces
- Live AR rendering of the configured product, placed by the customer where they want it, with material and variant swaps applied in real time.
- Who it is for
- Furniture and kitchen manufacturers, sanitaryware, doors and windows, machinery sold from a showroom — anything configured on a screen and installed in a place.
- Anchoring
- Plane detection plus tap. The customer chooses the spot, and for this class that is the correct behaviour rather than a limitation.
Universal repair assistance
A step-by-step procedure anchored to the physical machine, with each component highlighted where it really is.
Every step highlights the parts it concerns on the real object — isolate, ghost, explode — with AI assistance to interpret the manual and answer questions that fall outside the sequence. The measurement campaign is what shaped this class: at ~2.7 cm of visual-only drift a 30 cm panel highlights correctly while a 5 cm connector lands on its neighbour, and highlighting the wrong part is worse than highlighting nothing. So Class C anchors to a physical reference, and its accuracy budget is stated as a working radius around it: at one metre of arm, one degree of angular error is 1.75 cm of lateral error — which is what decides how many markers a machine needs.
- What it needs
- A CAD assembly — the semantics has to be derived rather than declared, because “M8 screw, position 3” must be addressable as an entity — plus the procedure as a document and a physical reference on the machine.
- What it produces
- A guided sequence on the real object: the right component highlighted at the right step, with the assistant available for questions off the happy path.
- Who it is for
- Machinery manufacturers with a field service network, technical support operations, training departments, consumer electronics repair.
- Anchoring
- Model target on the object, or a marker with a known offset. Manual alignment stays available as a fallback, because on smooth or symmetric machinery the other two degrade.
Side by side
| Comparison of the three content classes across source, semantics, anchoring, interaction and quality metric. | Class A Animated guide | Class B Showroom configurator | Class C Universal repair assistance |
|---|---|---|---|
| Content | A place, and a character narrating it | A configurable product from a catalogue | An assembly, and a step-by-step procedure |
| Typical source | Mesh — scan, photogrammetry, artistic reconstruction | Mesh authored to convention, or catalogue CAD | CAD assembly, plus a text document |
| Semantics required | None | Conventional | Derived — mandatory |
| Anchoring | Absolute, on the real place | Relative, chosen by the user | Absolute, on the physical object |
| Interaction | Conversation, gestures, following the visitor | Variant, finish and size changes at runtime | Step navigation, component highlighting |
| Quality metric | Credibility of the presence | Material fidelity — it is a sales tool | Legibility — which part, in which order |
Not sure which one yours is?
Most projects announce themselves in one sentence: what the content is, and whether anything inside it needs to be addressed by name. Write that sentence and I will tell you which class it falls into and what it needs.
Get in touch →