How it works
You draw the lines. Decorum keeps the agent inside them
A prompt can ask your agent to behave and to sound right. Nothing about a prompt makes it. Decorum turns your decisions into rules the agent can't talk its way past — and lays them on the agent loop you already run.
Why teams choose Decorum
Persona on your own agent loop
Set how your agent sounds — the register, the warmth, the formality — and run it on your own loop. The big labs offer this only inside their own products. Decorum lays it on yours.
The line is drawn first
Decorum doesn't wait for the agent to slip and then escalate to a human. The boundary is set before the agent ever speaks, so the disallowed answer is one it can't reach — not one someone has to catch.
Honest about what it catches
Decorum enforces the lines you draw. It doesn't promise to catch every trick a user might try — and it tells you plainly what it can't. The limits are written down, not glossed over.
Yours to run, start to finish
You install it and own it. There's no hosted service, no telemetry, no request leaving your walls. The record it keeps stays with you — the rule that fired, never your customer's words.
Your own rules, layered on top
Add the exceptions only you know — a phrase to avoid for one account, a disclosure a regulator wants — on top of the profile. They can tighten what the agent may do, never loosen it.
Proven, not asserted
Your conduct and persona ship as profiles you can test. A conformance run exercises each rule against real cases, so "it behaves" is something you checked, not something you hoped.
The record, in numbers
production deployments enforcing conduct and persona today
tests behind the engine, plus a published security audit
of your customer's words ever leave your walls