Skip to content

FAQ

No. A result is a planning aid. Flood and fire results in particular are for planning. Each report ends with a list of what was not checked, and anything that matters should be measured on site.

From a calculation on the scan, the aerial data or the record. A number goes from the calculation to a result object to the screen. The language model picks tools, labels and explains. It is never the source of a measured value, and a grader rejects an agent message that states one. See findings and reliability.

The tool returns an unknown verdict with the reason, such as no wall graph, no door data or no location. Many scans have no wall graph, so every tool checks the plan’s quality first.

Yes, through the iPhone app. It scans a room with LiDAR, using Apple RoomPlan, and the server converts the export to the same plan format. A phone scan has walls, doors, windows, rooms and objects. It has no photos, no mesh and no location, so tools that read panoramas or aerial imagery say they cannot answer. The app builds and runs in the simulator. It has not been installed on a phone yet.

Yes, on your own scans or the two made-up test scans. See getting started.

No. The live app is not public, because its API serves sponsor data. You run it yourself with your own access.

Yes. Every tool is an MCP tool and a command of the threshold CLI. The MCP server is read-only unless it is started with --write. See MCP server and CLI.

The scans we were given are in Houston and Los Angeles. Some map layers cover one city only. A phone scan works anywhere.

No. Its pictures show made-up test scans and our own drawings. The example on each tool page comes from a hand-made phone scan, run with no keys and the network switched off.

All rights reserved. CoStar’s hackathon terms, section 4.2, assign the project’s intellectual property to CoStar. See data sources and licences.