Multi-model AI decision engine for strategic pressure-testing, architecture review, and hard decisions. Runs structured analysis across top-tier models and produces a recommendation with confidence and clear caveats.

Not a chatbot. Not a debate club. A tool that gets to the right answer.

• App: https://cowork-council.vercel.app
• Repo: go2impact/Town-Hall

Technical slugs such as cowork-council still exist in package/deployment internals for continuity. Human-facing naming should now be Town Hall.

1. Round 1 — First Principles (parallel): All selected models analyze the question independently
2. Round 2 — Pressure Test (sequential): Each model reviews others' analyses, updates their view, flags errors
3. Round 3 — Synthesis: Moderator produces a recommendation with confidence level and autonomy classification (AUTO-PROCEED / NOTIFY-HUMAN / HUMAN-REQUIRED)

POST /api/ask — Programmatic endpoint for agent-to-agent calls

{
  "question": "Should we use MCP or custom integrations for Zendesk?",
  "models": ["claude", "gemini", "gpt5"],
  "systemContext": "We're building a desktop AI agent for remote CX workers...",
  "sync": true
}

Returns the full recap when sync: true. Set X-Api-Key header if COUNCIL_API_KEY is configured.

GET /api/health — Check status and configured models

This repo is now the canonical home for:

• the working Town Hall app
• Town Hall architecture references in architecture/
• preserved Town Hall discussion history in discussions/

Key references:

• architecture/town-hall-rules.md
• architecture/town-hall-spec.md
• discussions/2026-03-11-new-analysis-we-built-this-town-hall-for-decision.md
• discussions/2026-03-11-town-hall-is-running-thread-20260311-225302-all-5.md

npm install
cp .env.example .env  # Fill in your API keys
npm run dev

Works on any Node.js host. For production:

npm run build
NODE_ENV=production node --loader tsx server.ts