War Room v6

Automated multi-LLM strategy debate. Six frontier models argue through six structured rounds to produce a consensus strategy for any business problem.

Repository: github.com/iota-uz/warroom

Install

One-line install (macOS & Linux):

curl -fsSL https://raw.githubusercontent.com/iota-uz/warroom/main/install.sh | bash

Installs the latest release to /usr/local/bin/warroom (override with INSTALL_PATH=/path/to/bin curl -fsSL ...). Uses SHA256 checksums from the release.

Download a binary manually:

Go to Releases, pick the latest version and the archive for your OS (e.g. warroom_Darwin_all_* for macOS universal, warroom_Linux_amd64_* or warroom_Linux_arm64_* for Linux). Unpack and put warroom in your PATH.

From source: See docs/development.md.

Quick Start

warroom
warroom --topic-file path/to/brief.txt
warroom --resume latest

If OPENROUTER_API_KEY is not set, the TUI will prompt you once and can store it in your system keychain (macOS Keychain). Enter your mission or topic in the TUI; Opus then builds a detailed brief (or use --no-discovery to skip the interview). Once the brief is locked, the six-round debate runs automatically.

A full run is about 2 hours and typically costs $5–10 on OpenRouter.

CLI

Parameter	Description
--topic-file path	Load the debate brief from a file. Skips the topic prompt and Opus interview.
--no-discovery	Use the topic as-is and skip the Opus interview.
--preset MODE	Debate mode: strategy (default), creative, lifestyle, or polymarket. Each preset changes personas, round names, prompts, and vote format while the core engine stays the same.
--resume SESSION	Resume a session. Use the session folder name (e.g. 20260208_214722) or latest. Preset is restored automatically.

Topic is always entered in the TUI or loaded from a file; there is no topic argument.

Examples:

./warroom                                    # Interactive: topic in TUI, then Opus brief
./warroom --preset creative                  # Creative mode (art, writing, ideas)
./warroom --preset lifestyle                 # Lifestyle mode (life choices, career, values)
./warroom --preset polymarket                # Prediction market analysis with live Polymarket data
./warroom --topic-file briefs/eai.txt        # Pre-written brief, no interview
./warroom --no-discovery                     # Topic as-is, no interview
./warroom --resume latest                    # Resume last session
./warroom --resume 20260208_214722            # Resume by folder name

How a session runs

Round names shown below are for the default strategy preset. Other presets follow the same structure with different labels (see Presets).

1. BRIEFING — Opus interviews you (3–10 turns) to build the debate brief
2. SMOKE TEST — Quick API check that models are reachable
3. R0: RESEARCH — Each model researches its angle (web search)
4. R1: PITCH — Each model proposes a strategy (800+ words, EV calcs)
5. R2: CHALLENGE — Models challenge each other's proposals
6. R3: CONVERGENCE — Merge strong ideas, drop weak ones
7. R4: SYNTHESIS — GPT + Opus write the consolidated plan
8. R5: VOTE — All six vote YES/NO with confidence (0–100) — or BET/PASS in polymarket mode
9. VERDICT — Confidence-weighted tally → VERDICT.md and action plan

Models & personas

Short name	Model	Persona	Research angle
deepseek	DeepSeek V3.2	Devil's Advocate	Costs & tradeoffs
kimi	Kimi K2.5	Operator	Operational requirements
gpt	GPT 5.2	Strategist	Implementation paths
opus	Claude Opus 4.6	Risk Analyst	Risks & failure modes
gemini	Gemini 3 Pro	Researcher	Landscape & precedents
grok	Grok 4	Provocateur	Contrarian angles

Supporting: Gemini Flash (moderator), Sonnet 4.5 (summarizer), Perplexity Sonar Pro (web search).

Features: Anonymization (codenames Alpha–Zeta), hybrid search (native + Sonar), shared knowledge base, round summarization, tool calling (warroom_search, calculate, warroom_recall), live Polymarket data tools in polymarket preset.

Presets

Presets change the debate's "skin" — personas, round names, prompts, and vote format — while the core 6-model engine stays identical.

Preset	Use case	Rounds
strategy (default)	Business, capital allocation, GTM	Research → Pitch → Challenge → Converge → Synthesize → Final Vote
creative	Art, writing, ideas, design	Explore → Inspire → Critique → Fuse → Refine → Decide
lifestyle	Life choices, career, values	Discover → Propose → Stress-Test → Harmonize → Crystallize → Commit
polymarket	Prediction markets, portfolio construction	Market Scan → Thesis → Challenge → Convergence → Portfolio → Final Call

Polymarket preset

--preset polymarket connects to the Polymarket Gamma API for live market data. Models get extra tools (polymarket_fetch, polymarket_events, polymarket_search, polymarket_market) to pull real-time prices, volumes, and event structures. The discovery phase interviews you on bankroll, risk tolerance, and markets of interest, then builds a brief with exact market slugs. Vote format uses BET/PASS instead of YES/NO.

Output

Each session gets a timestamped directory under sessions/ (or $WARROOM_DIR if set):

sessions/20260208_214722/
  topic.txt
  discovery.md, discovery.json
  anonymization_map.json, anonymization_map.md
  round_0.json, round_0.md, round_0_<model>.txt, round_0_summary.md, ...
  VERDICT.md
  cost_report.md
  knowledge_base.json
  debug/round_*.jsonl

Requirements

• OpenRouter API key (covers the six debate models plus Sonar Pro, Gemini Flash, Sonnet 4.5)

---

Contributing / developing: See docs/development.md and docs/README.md.