How it works

The War Room runs a structured debate among the AI panelists you seat, and this page explains the mechanics of a single session in plain language, from the question you pose to the verdict you read at the end. It covers what the host does, how the seats take their turns, where each seat gets its instructions and its data, and how spending is tracked while the room works. Read it to understand what is happening on screen while a council debates. The War Room is available on the Pro plan and above and is rolling out in waves, so an eligible account may not see it yet.

What it shows

A session moves through a few clear stages. You set the question and the panel, the host runs the debate over one or more rounds, and the host closes with a written conclusion. The stages are described in the following sections, in the order they happen.

You set the topic and seat the council

Two things start a session: a topic and a council. The topic is the question you want debated, entered in step 1 of the configuration rail. The council is the group of panelists you seat in step 2, The Table. Each panelist sits in a seat, and a room holds between two and fifty seats. A seat is either one of the subagents you saved in the AI Console or a one-time panelist you create for this room only. In step 3, Format, you choose how the debate should run. Nothing is metered until you select Start session.

The host opens each round

Every room has a host, the orchestrator that leads the discussion. GNG Analyst hosts by default, or you can put one of your own subagents in charge. The host opens each round, frames the question for the seats, and keeps the discussion on your topic. It does not argue a side. Its job is to run the room and, at the end, to weigh what the panel said.

The seats take their turns

Once the host opens a round, the seats speak. The round style you chose sets the rhythm. In Sequential style, the seats speak one at a time in seat order, so the order you seated them is the order they argue. In Simultaneous style, every seat speaks at once in the same round. Each seat argues from its own instructions: a saved subagent brings the configuration you gave it in the AI Console, and a one-time panelist argues from the persona you typed when you created it. You do not have to name a model for a seat; leave it unset and the room picks a cost-efficient default, or choose one from the picker, where heavier models cost more credits. While a seat speaks, it can reach for its allowed data tools to pull live platform data and ground its argument, and that activity shows up beneath its turn. The panel runs for the number of rounds you set, or, if you chose Optimal, until the host judges it has heard enough.

The host closes with a synthesis and a verdict

When the rounds end, the host reads the full discussion and writes the conclusion. It produces a Synthesis, the complete written conclusion on the debate, and lifts the single most important sentence to the top as the Verdict, so the takeaway stands out. The seats argue the positions you assigned them, and a panelist can make its case with confidence and still be wrong, so treat the synthesis as the host's read of the debate, not a recommendation. The host then composes the Deep Research Report, the session's downloadable research note, from the whole debate. See The Deep Research Report.

Spend runs against your cap

A session spends GNG credits as it works, and you set a hard Spend cap in step 3 before you start. The running spend is shown while the room debates, and it can never exceed the cap you set. If the cost reaches the cap, the room stops and the host delivers a closing summary instead of starting another round, so a session cannot run away from you. Costs are always shown in GNG credits, never in currency or token counts.

The data behind a debate

The seats draw on the same live market and company data as the rest of the AI Console. Prices and quotes come from AlphaVantage, including Nasdaq quotes delivered through AlphaVantage, and public company filings come from EDGAR. A seat can also read your own saved knowledge when its subagent is bound to it. The figures a panel cites are current to the platform's normal refresh schedule at the moment the seat looks them up, and a finished session is a snapshot of what the panel said and cited at the time it ran.

Use cases

Follow what is happening on screen while a council debates, so you can read a live room with confidence.
Decide how to set a room up, once you can see how the round style, the number of rounds, and the host shape the debate.
Judge a verdict fairly by knowing it is the host's synthesis of the assigned arguments, not an independent recommendation.
Anticipate the cost of a session from the levers that drive it: the number of seats, the number of rounds, and the models involved.

GNG Research provides equity research and educational tools, not investment advice. Nothing on the platform is a recommendation to buy or sell any security. Do your own research and consider your circumstances before making any investment decision.