Formats, modes, and settings

The War Room settings panel controls how a council argues: the tone of the debate, whether panelists speak one at a time or all at once, how many rounds they run, how fast the text appears, who hosts, and a few model and tool options. This page explains where the panel lives, every control with its options and default, how to pick a mode, and which settings change the credits a session uses. 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.

Before you begin

Sign in with a Pro plan or above. The War Room opens from the AI Console, and access rolls out in waves, so an eligible account may still see a private beta panel until it is switched on.
Open the War Room and find the settings in step 3, Format, of the configuration rail. The defaults work for a first run, so you can skip this panel and adjust it later.
Set your format before you start. Every setting becomes read-only once a session runs, so the time to change a mode or the rounds is before you select Start session.

Quickstart

Open the War Room settings
Go to the War Room and, in step 3, Format, select Settings, shown as Tune in the rail. The panel is titled War Room settings.
Choose a mode and round style
Pick the Mode that fits the disagreement you want, then set Round style to Sequential or Simultaneous.
Set the rounds and pacing
Choose Rounds per agent, from 1 to 5 or Optimal, and a Render speed for how fast the text lands.
Confirm the host and options
Leave GNG Analyst as the host or hand hosting to one of your subagents, and turn on Optimal model mode or Enable all tools for subagents only if you want the extra capability and cost.
Review and start
Check the compact summary on the panel trigger and the idle stage preview, then select Start session when the rest of the room is ready.

What it shows

Open the panel from Settings, shown as Tune in the configuration rail. It is titled War Room settings and holds every format control in one place, each with a short caption that explains what it does. The panel trigger shows a compact summary of your current choices, read as mode, then style, then rounds, for example "Debate, Sequential, 1 rd". The idle stage preview repeats your choices as a summary list of Mode, Seats, Rounds, Style, Host, and Spend cap, so you can confirm the room before you start.

Inputs

The panel sets how the next session runs. Each control and its default is described in the following list.

Mode sets the tone and structure of the debate. Hostile cross-examines; friendly builds. There are six choices, and the default is Debate.
- Hostile: adversarial cross-examination, where panelists attack each other's reasoning and stress-test conclusions.
- Debate (the default): structured opposing arguments, where each panelist defends a position and rebuts the others by turn.
- Friendly: collaborative analysis, where panelists build on each other's ideas instead of arguing.
- Discuss: an open round-table conversation, less structured and more exploratory.
- Bull vs Bear: half the seats argue bullish and half argue bearish, and the synthesis weighs both sides.
- Analyst Team: domain specialists collaborate as a team, each panelist owning one angle such as fundamentals, technicals, or sentiment.
Round style sets whether voices rotate or overlap. The default is Sequential, one speaker at a time in seat order, so the order you seat panelists is the order they speak (see The table and its seats). Simultaneous has all seats speak in parallel each round.
Rounds per agent sets how many times each seat speaks, from 1 (the default) through 5, or Optimal, which lets the host stop when it judges the debate has run its course. More rounds mean a longer debate and more credits.
Render speed controls only how fast the streamed text lands in the spotlight; it does not change the debate or the cost. The options are Slow for cinematic pacing, Normal (the default) for a natural typing cadence, Fast to skim between turns, and Instant for no throttle.
Orchestrator (the host) is who opens each round and writes the final synthesis. The default is GNG Analyst, the platform's house orchestrator and the best baseline for most rooms. You can instead hand hosting to one of your own subagents.
Optimal model mode lets the host promote each panelist to a heavier model for the topic. It is off by default and costs more credits per turn.
Enable all tools for subagents, under the Agent tools heading, gives every seat the full research toolkit for this one room, covering filings, transcripts, the screener, market data, and memory. It is off by default and always stays within your plan's tool access.
Spend cap is a hard ceiling on the room's spend, set in GNG credits, that ends the room with a closing summary when the running cost reaches it. The full control, its default, its range, its presets, and how the cap behaves live on Spend cap and credits.

How to choose a mode

Pick the mode by the kind of disagreement you want at the table. Use Hostile to pressure-test a thesis, Debate for a structured pro and con, Friendly to build a shared view, and Discuss for open exploration. Choose Bull vs Bear for a two-sided read on a single name, and Analyst Team when you have seated specialists who each own a different angle.

Outputs and how to read them

The settings panel produces no result on its own; it configures the session you start next. Two places reflect your choices before you begin: the panel trigger shows the compact mode, style, and rounds summary, and the idle stage preview lists Mode, Seats, Rounds, Style, Host, and Spend cap. Read the preview as a final check that the room matches what you intended before you select Start session.

How it is calculated

There is no scoring formula behind the settings; each one shapes the debate or its cost directly. Mode sets the temperature and structure, round style sets whether voices rotate or overlap, and rounds set how long the debate runs. Render speed changes only how fast text appears on your screen and never the cost. The settings that raise the credits a session uses are more rounds, Optimal model mode, and enabling all tools for every seat; more seats and heavier models also cost more per turn. For how a session is metered and how the cap ends a run, see Spend cap and credits.

Data and timing

Settings apply to the session you start next and lock for that run once it begins. They are your own configuration, not market data, so no data feed is involved in the settings themselves. The live data a session pulls while it runs is described on The live room and the verdict.

Use cases

Choose Bull vs Bear for a two-sided read on one name, or Hostile to stress-test a thesis before you act.
Raise the rounds per agent for a deeper debate on a hard question, and lower them for a quick read.
Turn on Optimal model mode when a topic needs the room's strongest reasoning and you accept the higher cost per turn.
Hand hosting to a subagent you trust to open the rounds and write the synthesis in its own voice.
Set the render speed to Instant to skip the pacing, or to Slow to watch a debate unfold turn by turn.

Limitations and disclosures

Every setting locks when a session starts, so a change means ending the run and starting a new one. Optimal model mode, more rounds, more seats, and enabling all tools each increase the credits a session uses, and enabling all tools still respects your plan's tool access. A council is a debate among AI panelists closed by a synthesized verdict; it is research support, not a recommendation.

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.