The Agora

Summon anyone to debate anything.

Ask any question — trivial or profound — and watch a council of great minds deliberate. The Agora runs a multi-agent simulation where each persona is an independent LLM agent that forms opinions, takes stances, and argues from its own perspective. You get real stance tracking, round-by-round trajectory charts, and a synthesized verdict at the end.

Live site

https://agora-lvao3a.fly.dev/

What it does

1. Frame your question — type anything from "should I quit my job?" to "is democracy just mob rule with better branding?"
2. Select your council — pick from a library of pre-built philosopher personas (Socrates, Diogenes, Aristotle, Seneca, Marcus Aurelius, Epicurus...) or describe any person, character, or archetype and the AI will expand it into a full debate persona. Or don't use AI and add them directly.
3. Watch the debate — personas take turns speaking in a live chat view, or switch to the Arena view to watch pixel-art sprites stand on terrain and deliver speeches with typewriter-style speech bubbles.
4. Read the verdict — a structured decision brief with a headline, stance summary, each voice's final position, and an opinion-over-time sparkline per persona.

Stack

• Frontend: React + TypeScript + Vite + Tailwind 4, brutalist design system
• Backend: FastAPI + SQLAlchemy + Alembic, SQLite
• Simulation runtime: CAMEL-AI OASIS multi-agent framework
• LLM providers: Anthropic (Claude), OpenAI-compatible endpoints, or a local stub for development

Run locally

1. Backend

# Create Python 3.11 environment
uv venv --python python3.11 .venv
source .venv/bin/activate
uv pip install -r backend/requirements.txt

Create a .env file in the repo root:

# Use the stub provider for local dev (no API key needed)
SIM_PROVIDER=stub
SIM_MODEL=stub

# Or use Anthropic
# SIM_PROVIDER=anthropic
# SIM_MODEL=claude-haiku-4-5-20251001
# SIM_API_KEY=sk-ant-...

# Or any OpenAI-compatible endpoint (Ollama, Together, etc.)
# SIM_PROVIDER=openai-compatible-model
# SIM_MODEL=llama3
# SIM_API_KEY=...
# SIM_BASE_URL=http://localhost:11434/v1

Start the server:

uvicorn backend.app.main:app --reload --host 127.0.0.1 --port 8000

2. Frontend

cd frontend
npm install
npm run dev

Open http://localhost:5173.

3. Fly.io (single-service deploy)

This repository ships with a one-service Fly setup that builds the frontend and serves it from FastAPI.

# create app (first time only)
fly launch --name agora --no-deploy

# create persistent volume for sqlite + uploads/simulations
fly volumes create agora-data --region iad --size 1

# deploy
fly deploy

Env defaults are configured in fly.toml for a stub provider. To use a real provider, run:

fly secrets set SIM_PROVIDER=anthropic SIM_MODEL=claude-haiku-4-5-20251001 SIM_API_KEY=your_key_here

If you want a Postgres-backed deployment later, we can migrate the app off SQLite and use a managed DB.

API overview

Endpoint	Description
GET /api/personas	List all available personas
POST /api/personas/expand	Expand natural language → full Persona via LLM
POST /api/personas	Create custom persona directly
POST /api/documents	Upload context document (.txt / .md / .pdf)
POST /api/panel/recommend	AI-recommended panel for a decision
POST /api/sessions	Start a new debate session
GET /api/sessions/{id}	Get current session snapshot
POST /api/sessions/{id}/advance	Run the next debate round
POST /api/sessions/{id}/interjections	Inject a user message mid-debate
POST /api/sessions/{id}/finish	End debate and generate verdict
GET /api/sessions/{id}/events	SSE stream of session events

Project structure

agora/
├── backend/
│   └── app/
│       ├── main.py           # All API routes
│       ├── models.py         # Pydantic request/response models
│       ├── entities.py       # SQLAlchemy ORM tables
│       ├── repository.py     # Data access layer
│       ├── services/         # Panel selection, persona expansion, documents
│       └── simulation/       # OASIS runtime, LLM provider abstraction, prompts
└── frontend/
    └── src/app/
        ├── pages/            # Home, SummonCouncil, Debate, Verdict
        ├── components/       # BrutalistButton, BrutalistCard, DebateArena, DesktopPet
        └── data/             # philosophers.ts (static persona definitions)

Development tips

• Use SIM_PROVIDER=stub during UI work — no LLM calls, instant responses, zero cost.
• The Arena view uses pixel-art sprites from public/pixelart/. Each philosopher folder contains directional sprites (rotations/south.png, east.png, west.png).
• The frontend proxies API calls to http://localhost:8000 via Vite config.
• Run backend tests: pytest backend/tests -q
• Build frontend: npm run build (from the frontend/ directory)