Cranley
Ask anything about a building and get a measured answer: an agent that checks real buildings inside and out before you commit.
Built solo at HackPrinceton Fall 2026 for CoStar's "Inside and Out: AI for the Built World" track. Listings say "accessible" or "two-car garage" and nobody measures. You say what you need in plain words, such as "I use a wheelchair, 81 cm wide. Can I reach the bathroom?", and Cranley picks the tools from 78 that measure real buildings, using Matterport interior scans, Vexcel aerial imagery and its 3D city, CoStar records, or your own iPhone scan. It checks many buildings at once, one subagent each, and asks before slow steps such as a NIST CFAST smoke model. Every number comes from a calculation, never from a model's text, and shows its source: the route through the 3D building, the narrow door in red. "Widen this door" re-runs the checks, and a PDF report lists the fixes. Run through the real data, 50 tools on 15 places gave 266 answers and 62 honest "cannot answer here", with none judged wrong. Before and during the build he interviewed Jean Ryan, President of Disabled In Action of Metropolitan New York, and gathered written answers from 74 people online; their answers changed the tools the same night. It also answers by iMessage and phone call, and to other agents over MCP. The sponsor data stayed with the event; the public code has none of it.
- Status: Shipped, built November 13, 2026 to November 15, 2026
- Kind: AI agents, Cities
- Event: HackPrinceton Fall 2026
- Stack: Python, FastAPI, React, three.js, deepagents, SwiftUI
- Live: https://cranley.app
- Devpost: https://devpost.com/software/threshold-o1e7uj
- Code: https://github.com/benz16107/cranley
- Video: https://www.youtube.com/watch?v=1Qcbzixrxz4




