GitHub - Kheil-Z/elenchus: Elenchus: multiplayer LLM workspace. Stop copy-pasting AI replies. One thread — your team, your AI, everyone in the same room, each using your own LLM API key or custom model. Think "Claude Team" without the hassle.

Multiplayer AI workspace. Multiple people share one conversation thread, each bringing their own API key. @mention Claude, Gemini, ChatGPT — or your own self-hosted model. Whoever you tag responds, billed to the person who called them.

Live demo — hosted on Supabase free tier, so new account creation may be rate-limited.

Contents

• Features
• Tour
• Bring your own model
• Self-hosting
• Project structure
• Tech stack
• Privacy
• License

Features

• Multiplayer conversations — shared thread, real-time sync, messages attributed to each person
• BYOK per user — each member adds their own API key; you never pay for anyone else
• Multi-provider — Anthropic (Claude), Google (Gemini), OpenAI (ChatGPT); or Custom endpoint: each user picks their own
• Bring your own model — point your account at any OpenAI-compatible endpoint (Ollama, MLX, LM Studio, vLLM, Groq…), name it, and @mention it like any other provider
• Vision + documents — attach images and PDFs; images sent as vision inputs, PDFs extracted or sent natively to Claude
• Project documents — upload files project-wide or scoped to a single conversation; control when the AI sees them
• Custom system prompts — per-project instructions for the AI
• Activity feed — catch up on what happened while you were away, with unread indicators

Tour

Real-time multiplayer chat — two users in the same conversation, one @mentions Claude. Each user has their own API key and can have a different provider; the token cost is attributed to whoever triggered the call. The provider indicator is visible at the bottom of the input.

Projects and documents — a project groups conversations, members, uploaded files, and an activity feed. Documents can be scoped to a single conversation or shared project-wide. Inside a chat, you control when files are sent to the AI (Always or Never). Token spend is broken down per user so everyone can see who is using what.

Settings — each user sets their own display name, avatar colour, and API key. Keys are encrypted at rest. You can revoke a key or switch providers at any time without affecting other members.

Bring your own model

Besides the three built-in providers, any OpenAI-compatible endpoint can act as your AI — a cloud service (Groq, Together) or a model running on your own machine (Ollama, MLX, LM Studio, vLLM). In Settings → AI Provider → Custom you set an agent name (which becomes the @mention handle), the endpoint's base URL, the model name, and an optional API key.

The one thing to know: AI calls are made by the server, not your browser — so a model on your own machine needs a public URL. The repo ships a small auth proxy (scripts/auth_proxy.py) that locks a free tunnel down to requests carrying your secret token, so nobody but Elenchus can reach your model.

Full walkthrough — tested on macOS with mlx-lm and a Cloudflare tunnel: docs/self-hosted-llm.md

Self-hosting

Prerequisites

• Node 22+ and pnpm
• A Supabase account (free tier works)
• A Vercel account (free tier works) — or any Node-compatible host

1. Clone and install

git clone https://github.com/Kheil-Z/elenchus.git
cd elenchus
pnpm install

2. Create a Supabase project

1. Go to supabase.com → New project
2. Once provisioned, open SQL Editor → New query
3. Paste and run the entire contents of supabase/migrations/000_schema.sql
4. Go to Authentication → Providers → Email and confirm it is enabled
5. Go to Authentication → Email Templates and update the sender name so confirmation emails don't reference Supabase

3. Set environment variables

cp .env.example .env.local

4. Run locally

Open http://localhost:3000.

5. Deploy to Vercel

1. Push your repo to GitHub
2. Import it at vercel.com — it auto-detects Next.js
3. Add the four environment variables from step 3 in Vercel's project settings
4. Deploy
5. In Supabase → Authentication → URL Configuration, set Site URL to your Vercel URL and add it to Redirect URLs (e.g. https://your-app.vercel.app/**) — without this, email confirmation links will not work

Project structure

elenchus/
├── app/
│   ├── (protected)/       # Authenticated pages (projects, chat, settings)
│   ├── api/               # Server-side API routes (LLM calls, uploads, etc.)
│   ├── auth/              # Login / signup pages
│   └── privacy/           # Privacy policy (public)
├── components/            # Shared UI components
├── lib/                   # Types, Supabase client, LLM layer, utilities
├── scripts/               # Companion tooling (auth proxy for self-hosted models)
└── supabase/
    └── migrations/        # Single schema file — run once on a fresh project

Tech stack

Privacy

See PRIVACY.md or the in-app privacy policy.

License

MIT © Kheil-Z