Almost every AI product this publication has covered this autumn attaches itself to a building that is being designed or built. Qonic’s agents run inside the BIM model. Mach9 drafts from survey data. The common assumption is that somewhere there is an authored model, and the AI’s job is to work on it.
A great deal of the existing building stock has no such model. What it has is a facilities manager, a drawer of PDFs, and a spreadsheet.
qapture, a company in Linz, Austria, has spent six years attacking that problem from the other end — laser scanning existing buildings and turning the captures into digital twins. It has now put an AI agent on top of the result. The product is called qentin, it is in early access, and the launch was reported on 12 November by the German trade title Business Geomatics and on 13 November by AEC Magazine. qapture’s own press page lists coverage rather than releases, and this publication found no dated company release for the launch — but the record is not silent before November: that same page carries a MeinBezirk piece of 30 October 2026 headlined, in translation, “‘qentin’ is coming”, trailing the product ahead of its arrival.
What It Does
The sequence qapture describes runs: scan, detect, ask.
A building is scanned and georeferenced. The software then automatically detects, names and locates objects in that scan — fire extinguishers, escape-route signage, gates, plant and equipment — and text recognition reads the labels and asset tags physically fixed to them. A facility team then asks questions through a chat interface, in German or English, and the answer appears in the 3D viewer, pinned to the place in the building it refers to, optionally with a route to it.
The company’s qentin page lists its own example questions, which give a better sense of the scope than any description would:
“How many fire extinguishers are in this hall?”
“Where is the nearest emergency exit?”
“Show me all damage on the ceiling.”
“Which exits and material-handling routes became obstructed since the previous inspection?”
“Compare all facilities against their latest baselines and show the highest-risk exceptions.”
Those last two are a different class from the first three. Finding the nearest fire extinguisher is retrieval. Comparing a facility against its own prior baseline, or a portfolio against each of its baselines, is a question about change over time — and it is only askable because the underlying asset is a dated capture rather than a drawing. qapture says that when something changes, only the affected area is rescanned, which it argues is faster and cheaper than redrawing plans.
The requirements are modest and the company states them: a qapture scan or the customer’s own data, a digital twin with objects already detected, and a browser. No installation, no new hardware.
The Limit Is Printed Next to the Capability
What is less usual about this launch is where the constraint sits: on the marketing page, in the vendor’s own words, at the same size as the feature list.
qentin draws only on the customer’s scan and digital twin, not on general internet knowledge. Every answer is rendered in the 3D viewer so it can be checked there, and a user who finds an object misclassified can correct it in the model. Business Geomatics summarises the design intent in three German words — Jede Antwort ist prüfbar, every answer is checkable — and the product page states the refusal case directly: “What is not in the data, qentin does not answer.”
The page also labels qentin an AI assistant and links that label to EU AI Act transparency obligations; says building data sits on the qapture platform, hosted to ISO 27001 in the EU, with the language model run by a specialised AI provider and the data processed there agreed with the customer in advance; and says qentin is not a replacement for a CAFM system.
Analysis. This publication has started to treat a published refusal as a signal rather than a courtesy. Mach9 shipped an aerial product with a list of the jobs it will not do. Planera’s Procore agent annotates and nothing else. A vendor willing to write down where its model stops is making a claim that can be tested against the product, which is more than a vendor claiming only capability has given you.
The specific refusal here is also the right one for the domain. A facilities question has a correct answer that exists in physical space, and a model that fills a gap with a plausible fire-extinguisher location is worse than useless — it is a safety document that is wrong. Grounding the agent in the scan, pinning the answer to coordinates, and declining the unanswerable question is the shape the problem demands. Whether qentin actually holds that line across a real portfolio is not something a product page can establish, and neither this publication nor the trade coverage has tested it.
The Customer Is the Evidence
What the launch does have is an operating customer with numbers.
Energie AG Oberösterreich, the Upper Austrian energy utility, is described by Business Geomatics as qentin’s first customer, and both it and AEC Magazine report the utility using the product in live operations. On qapture’s own site the engagement is quantified: 137 sites, 730,000 square metres of existing buildings, 43 hydropower plants, and 6,300 labels automatically detected. AEC Magazine reports that the most-used feature there is asset-tag search — a user types a tag and the 3D model shows where that asset is.
That is a mundane-sounding capability and it is almost certainly the honest one. Finding a numbered asset somewhere in an estate of that size is the kind of task that burns a working day and produces nothing anyone would call innovation.
qapture’s wider reference list, published on its own site, names IKEA — a partner since 2020, with 38 sites and more than two million square metres captured, per AEC Magazine and Business Geomatics — along with voestalpine, TGW Logistics Group, the Austrian Parliament, Drees & Sommer, ArcelorMittal, HABAU, DPR Construction, Wien Energie, ASFINAG and BIG Bundesimmobiliengesellschaft. A counter on the site puts the total captured at more than 20 million square metres across roughly 200 projects. These are the company’s own figures and are reported as such.
The Company
Daniel Höller and Dominic Koll founded qapture in Linz in 2020. Business Geomatics reports that they built it without outside investors — ohne Investoren — and that about 25 people now work between Linz and an office in Vienna; Höller is chief executive and a co-founder. Nothing on the coverage list qapture publishes announces a funding round, though it does record the company joining the portfolio of tech2b, an Upper Austrian startup incubator, in February 2024. So the fair statement is that qapture has announced no funding round and describes itself as founder-built — not that it has never taken outside money of any kind.
Höller’s framing of the problem, as quoted by Business Geomatics, is that a building’s knowledge today sits in Excel lists, old plans, PDF reports and in individual people’s heads; with qentin, he argues, it sits in the building and can simply be asked for. To AEC Magazine he put the company’s position more bluntly: “For six years we have structured building data so that an AI can work with it.”
Analysis. That sentence is the actual argument, and it is worth separating from the product. The scarce thing in this market is not an agent. It is six years of georeferenced, object-detected, tag-read captures of real buildings that customers paid for at the time for their own reasons. The agent is the interface; the capture archive is the moat. This publication made the same observation about Trendspek, where seven years of engineer-classified defects inside a 3D condition record turned out to be the training set and the reason the round got led, and about Ageiro, where the valuable object was the commercial record rather than the dashboard over it. It keeps being the same observation.
What Is Not Known
qapture has not published a price figure. Its page says a demo on the customer’s own data is free, and that companies setting qentin up on their data during 2026 get a special price — unspecified — with the offer running to 31 December 2026. Early access is open by request form.
Neither the company nor the coverage gives an accuracy figure for the object detection or the text recognition, a false-positive rate, or any measured result at Energie AG beyond which feature gets used most. The product page is candid that qentin “develops progressively” and that what a given customer can do depends on the facility, the data coverage, the analysis results and which features are enabled — some listed workflows are in use with customers, others are being built with early-access customers. That is an unusually honest sentence to put on a launch page, and it is also a reminder that early access is what this is.
