Verbo

A debate platform where every argument gets fact-checked.

Verbo lets people discover debate topics, argue FOR or AGAINST with cited sources, and get AI-powered fact-checking grounded in live web search. Post an argument and keep talking — a reasoning model verifies it in the background and the rating pill (True / Somewhat True / Misleading / False / Unverifiable / Unrelated) appears when it's done.

Dark mode

Project status

Verbo was built as a learning MVP and is open-sourced as-is. It is not actively maintained — issues and PRs are welcome but responses may be slow. Fork freely.

Features

• Async AI fact-checking — arguments post instantly; a reasoning model with web search verifies them in the background and attaches a rating, reasoning, and grounding sources
• Stance detection — the AI flags when your argument's content doesn't match your declared FOR/AGAINST stance
• Research libraries — AI-generated FOR / NEUTRAL / AGAINST source collections per topic, with synthesis of where sources agree, disagree, and what's underexplored
• AI consensus summaries — distills the top verified arguments on each side
• Semantic search — find debates by meaning, not keywords; duplicate detection when creating topics
• Voice input — dictate arguments, transcribed via Whisper
• Argument enhancement — optional AI rewrite for clarity before posting
• Timed debates, participant management, magazine-style home feed, light/dark themes

Architecture

┌────────────────────┐       POST /api/*       ┌────────────────────┐      ┌────────────────┐
│  React client      │ ──────────────────────► │  verbo-backend     │ ───► │  OpenAI API    │
│  (Vite, Tailwind)  │ ◄────────────────────── │  (Hono proxy)      │      │  + web_search  │
└────────────────────┘                         └────────────────────┘      └────────────────┘

• Frontend (/): React 19 + Vite + Tailwind. Ships with rich mock data and keeps all state in memory — there's no database; this is a product prototype, not a hosted service.
• Backend (server/): a thin Hono proxy that holds the OpenAI API key, fronts the Responses API (with web_search for grounded fact-checking) and Whisper, and enforces an origin allowlist + per-IP rate limiting. See server/README.md.

Fact-checking is async by design: the UI never blocks on the model. If the backend isn't running, the app still works — AI features degrade gracefully and a banner tells you why.

Quick start

Prerequisites: Node.js 20+

Frontend only (no API key needed)

npm install
npm run dev

Open http://localhost:3000. You can browse, search, create debates, and post arguments against the mock data. AI features (fact-checking, research, semantic search) stay off and the app shows an "AI features are offline" banner.

Full setup (AI features on)

In one terminal, start the backend with your OpenAI API key:

cd server
npm install
cp .env.example .env.local    # then put your OpenAI API key in .env.local
npm run dev                   # listens on :8787

In a second terminal, start the client:

npm install
cp .env.example .env.local    # defaults already point at localhost:8787
npm run dev

Environment variables

Variable	Where	Purpose
VITE_PROXY_URL	.env.local (root)	Backend URL the client calls. Default http://localhost:8787
OPENAI_API_KEY	server/.env.local	Your OpenAI key — never exposed to the browser
ALLOWED_ORIGINS	server/.env.local	Comma-separated origins allowed to call the API
PORT	server/.env.local	Backend port (default 8787)

API

The backend exposes GET /health plus nine JSON POST /api/* endpoints (verify, suggest-sources, consensus, tags, enhance, research, research-synthesis, transcribe, debate-search). The full table, security notes, and Fly.io deployment guide live in server/README.md.

Before deploying publicly: the proxy has no client authentication — read the Security section first and set a spending limit on your OpenAI account.

License

MIT