COMPANY BRAIN
PLATFORM / GOVERNED ASSISTANT · IN PROGRESS

A coworker
that only says what's approved.

A conversational assistant is in active development on top of the same governed backend as everything else on this site: grounding-confirmed facts, published skills, cited evidence. It is not a general-purpose chatbot over raw documents—an autonomous caller still can never retrieve a proposed or unconfirmed fact, by design.

02 / LIVE TODAY

The conversation layer is coming.
The refusal-to-guess already works.

This is the grounded answer a governed assistant would hand back today, before any chat interface exists around it. Trigger a source change and watch it decline to answer from the stale version.

INTERACTIVE DEMONSTRATION
ASSISTANT ANSWER / REFUND AUTHORITYANSWER LIVE
PREVIOUS SOURCEv18.1

Manager approval required over

$300Finance Operations / approved
GOVERNED SOURCEv18.2

Manager approval required over

$750Prepared for named owner review
AGENT ANSWERCan I approve a $500 refund?
LIVENo — manager approval required
RECEIPTSOURCE → OWNER → ANSWER IMPACTVersion linked
THE SAME RULE, IN A NEW SURFACE

Grounded first.
Conversational second.

Confirmed facts only

The assistant answers from grounding-confirmed facts and published skills—the same rule enforced today at the MCP layer, not a relaxed version of it for the sake of a friendlier interface.

Cites what it used

Every answer carries the same confidence, freshness and provenance summary the structured lookup tools already return.

What's actually new

The conversation layer—turning a structured query into a natural-language exchange. The grounding rules underneath don't change for the sake of the new interface.

STATUS

In progress.
Not shipped today.

There is no conversational assistant UI in production yet. The structured facts and skills it will draw on are already real and already callable via MCP.

See the MCP tool surface