Proxy RoomYour agent negotiates.
You stay accountable.
Two people. Two AI proxies. Each proxy acts only inside a sealed Intent Card, tags every claim as stated, sourced or assumed, and stops to ask its human when it reaches a limit. What comes out is a decision brief you can explain without opening the chat history.
The demo is free and takes about a minute. You answer one question your proxy escalates to you.
Human sets intent and constraints → Proxy A ⇄ Proxy B → Human receives the outcome, and the reasons
Seal an Intent Card
Write your brief in plain language. Claude drafts the card: goal, must-haves, what you may concede, hard limits, when to ask you. You confirm it and it is sealed with a hash. The other side never sees it.
Every claim is tagged
Stated by my human, sourced from a document, or assumed. Assumptions stay amber until a human confirms them, and repeating the other side's claim never upgrades it. That is how a cascade gets caught.
Escalate, then brief
At a limit, a proxy stops and asks one concrete question. The agreement ships with a brief: terms, whose authority covers each one, the assumptions it still rests on, and a hash-chained ledger.
Open a live room
You get two private links: one for you, one for the person you're negotiating with. Each of you writes your own brief. The proxies take it from there.
The demo works without them.
Each link controls one seat. Anyone with a link can act for that seat, so share it privately.