Origin
Nobody sets out to build property management software.
Two of our founding group were running operations for a mid-sized third-party management company. The portfolio had grown from 600 doors to just under 3,000 in three years. Headcount on the service desk had gone from four people to five.
The failure mode was always the same and it was never dramatic. A prospect called on Sunday afternoon and rented somewhere else by Tuesday. A resident texted about a leak at midnight and got an answer at nine the next morning, after the ceiling came down. Month-end reporting swallowed the last weekend of every month, and the reports still went out two days late.
None of that was anyone's fault. There simply wasn't a person available at the moment the work arrived. Everything else — the software, the process, the training — was downstream of that one gap.
The bottleneck was never skill. It was availability.
The other half of the founding group had spent years shipping production machine learning at companies where the model was the product. By late 2022 it was obvious that language models had crossed the line from demo to dependable for exactly this class of work: reading a messy human request, asking the right follow-up, and taking a bounded action against a system of record.
We spent 2023 building against one real portfolio, in production, with real residents and real consequences. That constraint shaped everything. APM escalates aggressively, logs obsessively, and never takes an action it can't explain — because the first version had to survive an owner asking why a vendor was dispatched at 2 AM.