The agent handles most calls well. It also gets things wrong sometimes, and you should know exactly what that looks like before you put it in front of your tenants.
What it does when it is unsure
The agent is built to escalate rather than improvise. When a situation does not fit a category it knows, or when a caller asks something outside its scope, it routes to you instead of guessing. It never gives a caller advice, safety instructions, or troubleshooting steps. Its job is to understand what is happening, get it to the right person with context, and document it.
That design choice means the most common failure is an unnecessary escalation, not a bad answer. We would rather send you a call you did not need than have the agent handle something it should not have.
Known issues we are actively working on
Emergency keyword false positives. Words like "fire" in ordinary contexts, a fire pit or a fire department visit last week, can still read as urgent. We are training this down.
Name and unit number mishearings. The agent confirms back what it heard, but phone audio is phone audio and it occasionally gets a unit number wrong.
How you catch it
Every call is recorded, transcribed, and summarized, and you get a notification after each one. If the agent mishandled something, you will see it in the transcript, not discover it a week later from an angry tenant.
When you find one, send it to us. Specific transcripts are the single most useful thing you can give us, and they are how most of the improvements above got prioritized in the first place.
Emergency alerts are always on
You can tune most notification preferences. You cannot turn off emergency escalation. That is deliberate.