Two AI parties · Live voice · One mediator’s chair
Practice the room before the room matters.
You sit in the mediator’s chair. The parties across from you are AI, and they talk out loud. Each one knows things the other doesn’t, and neither gives ground until you’ve earned it.
A live demo that talks to the OpenAI API · Everyone in it is fictional
Signature case Juniper Street Renovation — ready to mediate
How a session goes
- 01
Start at the table
Both parties speak for themselves, out loud. Whatever you say at the table, they both hear.
- 02
Pull one party aside
Ask one party for a private word. The other actually leaves the room and hears none of it. People tell you things in caucus they’d never say at the table.
- 03
Bring back only what’s allowed
Before you walk back in, ask what you can repeat. Then try to land the deal. If you leak something private, the room remembers.
What’s inside
The Living Table
The room itself, where the voices and captions live. A director you never see keeps the pacing honest, and if someone won’t stop talking, you can take the floor back.
Enter the room →Scenario Library
The case files. What comes pre-built lives here, and everything you draft gets shelved next to it.
Open the library →Drafting Desk
Describe a dispute in a few sentences. It comes back as a playable case, secrets included.
Draft a scenario →Under the hood
Each party is a separate Realtime voice session, blind to anything the others told you in private. Offstage, a director on the Responses API picks who speaks next and which facts they may use.
The rules of the room itself (who’s in it, what’s cleared to share) are plain application code, and the model can’t override them. If the director stumbles, a scripted fallback takes that turn. Paid calls run under hard budget caps.