A high-performance AI chat application built with Phoenix LiveView for the T3 Chat Cloneathon.

Core Concept: Leverage BEAM's distributed actor model to make chat blazing fast while keeping costs minimal.

• Phoenix LiveView: Real-time UI without JavaScript complexity
• PostgreSQL: Single database for cold storage
• Actor Model: Each chat is a stateful GenServer process
• Memory-First: Hot data in ETS, database is just backup
• Global Distribution: BEAM nodes can run anywhere, chats follow users

• Support multiple language models (OpenAI, Anthropic, etc.)
• User authentication with chat history sync

• File attachments (images, PDFs)
• AI image generation
• Code syntax highlighting
• Resumable conversation streams
• Chat branching
• Conversation sharing
• Web search integration

User → WebSocket → LiveView Process → Chat Actor (GenServer) → AI Provider
                                    ↓
                              PostgreSQL (async)

• Each chat session is an isolated actor
• Streaming responses flow naturally through WebSockets
• No API layer needed - direct server-to-client updates
• Fault tolerance built-in with supervisors

• Sub-100ms initial page load
• Instant message sending (optimistic UI)
• Zero-latency client interactions
• Streaming tokens as fast as AI provides them, from a locally sourced llm provider

MIT

Built with Elixir for the T3 Chat Cloneathon 2025