Tic-Tac-Toe Demo - No LLM, Pure Minimax

A tic-tac-toe game demonstrating elizaOS's ability to run agents without an LLM.

Key Features

No Character Required: Uses the new anonymous character feature (Agent-N)
No LLM Calls: Custom model handlers implement perfect play via minimax
Custom Model Handlers: Intercepts TEXT_LARGE and TEXT_SMALL to return optimal moves
plugin-sql: Uses PGLite in-memory database for persistence

Implementation

TypeScript entry: game.ts.

How It Works

Instead of calling an LLM, this agent registers custom model handlers that:

Parse the board state from the text prompt
Use the minimax algorithm to find the optimal move
Return the move as a simple number (0-8)

The agent never loses - it will always win or draw!

Quick Start

bash

# TypeScript
bun run examples/tic-tac-toe/game.ts

Architecture

┌─────────────────────────────────────────────────────────┐
│                    AgentRuntime                         │
│                 (Anonymous Character)                   │
├─────────────────────────────────────────────────────────┤
│  plugins:                                               │
│  ├── plugin-sql         (persistence)                   │
│  ├── bootstrap-plugin   (basic capabilities)            │
│  └── tic-tac-toe-plugin (custom model handlers)         │
├─────────────────────────────────────────────────────────┤
│  runtime.useModel(TEXT_SMALL, { prompt: boardState })   │
│           ↓                                             │
│  ticTacToeModelHandler() ← NOT an LLM!                  │
│           ↓                                             │
│  parseBoardFromText() → minimax() → optimal move        │
└─────────────────────────────────────────────────────────┘

Performance

Since there's no LLM latency, the AI responds instantly:

~100 games per second on typical hardware
Zero API calls
Zero cost