# talktotalk.me

> A social experiment platform where users ask questions and receive anonymous answers from real people or AI models.

## Terminology

- In the product UI, “model” usually means a **responder persona** (display name + avatar), not necessarily a machine-learning weights file.
- **AI fallback** answers are generated by the server calling a **configurable HTTP REST** endpoint (OpenAI-compatible chat completions). See `docs/LLM_REST.md` and `docs/BYOK.md` in the repository for operators.
- **/lab** is an optional browser page that runs **TensorFlow.js** demos locally: training and inference happen on the user’s device; no model weights or training data are sent to talktotalk.me for that page.
- **Coins** are optional **virtual points** stored in the app database for engagement (not cryptocurrency). They are not redeemable for cash unless the operator adds separate payment flows.

## What it does

- Users submit questions anonymously
- Questions are matched with human "responders" in real-time via WebSocket
- If no human responds within the timeout, an AI model generates a fallback answer (REST call from the server)
- "Chaos Mode" sends the same question to multiple responders for diverse perspectives (may consume virtual coins if enabled by the operator)
- Users can share personal question links (/ask/:slug) on social media
- Real-time conversational messaging with follow-up questions

## Tech stack

- Node.js + Express + Socket.IO
- SQLite (via Knex + better-sqlite3)
- Mithril.js + Tailwind CSS (Vite-built frontend)
- EJS server-side templates
- Google Gemini API for translations (scripts)
- Optional TensorFlow.js on `/lab` (client-only demo)

## HTTP API (summary)

Authentication: many routes expect `Cookie: token=<JWT>` or `Authorization: Bearer <JWT>`.

- `POST /api/session/guest` — create anonymous guest session
- `GET /api/me` — current user (auth)
- `PATCH /api/me/profile` — update `displayName` / `avatar` (auth)
- `GET /api/me/slug` — shareable question link slug (auth)
- `GET /api/me/coins` — virtual coin balance (auth)
- `POST /api/coins/spend` — body `{ "action": "tip", "answerId", "amount" }` (auth)
- `POST /api/questions` — submit a question (auth); chaos mode may require sufficient coin balance
- `POST /api/ratings` — rate an answer (auth); responders may earn coins
- `POST /api/auth/register`, `POST /api/auth/login`
- `GET /api/threads/:threadId/messages` — thread messages (auth)
- `POST /api/reports`, `POST /api/blocks`
- `GET /health` — health check

## Operator LLM integration (REST)

The server posts to `LLM_API_URL` with `Authorization: Bearer LLM_API_KEY` and an OpenAI-style JSON body. Details: repository `docs/LLM_REST.md`.

## WebSocket

- Namespaces: `/chat` (askers), `/answer` (responders)
- Events include `question:submit`, `question:status`, `answer:complete`, `rating:submit`, etc.

## Features

- Anonymous Q&A with real human responders
- AI fallback when no humans available (server-side REST)
- Chaos Mode (multi-responder answers)
- Virtual coins (optional): signup bonus, chaos charges, rating rewards, tips
- Personal question links for social sharing
- Real-time typing indicators
- Multi-language support (15+ languages via AI translation)
- Browser notifications for background messages
- PWA-ready (installable, standalone mode)
- `/lab` — local TensorFlow.js demo

## Links

- Website: https://talktotalk.me
- Ask questions: https://talktotalk.me/app
- Become a responder: https://talktotalk.me/responder
- Lab (TF.js demo): https://talktotalk.me/lab