The problem
A freeform assistant can produce inconsistent replies, mix context between conversations, or send messages that a human did not explicitly approve.
Safety-first research
Internal research into making a local real-estate messaging assistant predictable enough to support narrow workflows without handing freeform automation control of client communication.
A freeform assistant can produce inconsistent replies, mix context between conversations, or send messages that a human did not explicitly approve.
The prototype was redesigned around deterministic replies, isolated sessions, explicit opt-out handling, human review, and outbound controls that deny action unless the state is known and allowed.
A focused selection of the work that defines this project at its current status.
04 · How it works
An inbound lead message creates an owner alert with the complete message and an approval code. No ordinary response is sent until the owner approves one exact canned reply; recognized opt-outs follow a separate deterministic confirmation and suppression path.
Live end-to-end validation remains pending, so the prototype is not presented as a production messaging system.
Fat Cat Finance does not offer installation or maintenance of the underlying OpenClaw framework as a service.
06 · Current evidence
Internal only; not offered as a service
An internal prototype demonstrating how stronger deterministic safeguards can replace unreliable freeform messaging behavior.
Complete controlled end-to-end validation before considering any real-world use beyond internal research.
Supporting technology