Eliza — TypeScript AI Social Agent Framework

Eliza Architecture

Core Components

┌────────────────────────────────────────┐
│            Character Config            │
│  Name, Bio, Style, Knowledge, Plugins  │
├────────────────────────────────────────┤
│            Agent Runtime               │
│  Memory │ Actions │ Evaluators │ RAG   │
├────────────────────────────────────────┤
│           Platform Clients             │
│  Discord │ Telegram │ Twitter │ Slack  │
├────────────────────────────────────────┤
│            Model Providers             │
│  OpenAI │ Claude │ Gemini │ Ollama    │
└────────────────────────────────────────┘

Character System

Define agent personality with a JSON character file:

{
  "name": "Luna",
  "bio": [
    "Luna is a crypto-native AI that helps DeFi users understand yield farming.",
    "She explains complex concepts in simple analogies."
  ],
  "lore": [
    "Was created during the 2024 DeFi summer.",
    "Has analyzed over 1000 liquidity pools."
  ],
  "knowledge": [
    "DeFi protocols and yield strategies",
    "Smart contract security basics"
  ],
  "messageExamples": [
    [
      {"user": "user1", "content": {"text": "What's impermanent loss?"}},
      {"user": "Luna", "content": {"text": "Think of it like this: you put apples and oranges in a basket. If apples suddenly become 10x more valuable, you'd wish you kept them separate. That difference is impermanent loss."}}
    ]
  ],
  "style": {
    "all": ["Uses food analogies for DeFi concepts"],
    "chat": ["Keeps explanations under 3 sentences"],
    "post": ["Shares one DeFi insight per tweet"]
  }
}

Multi-Platform Deployment

Deploy the same agent across platforms simultaneously:

// Each client maintains the same personality
// but adapts to platform norms
const agent = new AgentRuntime({
  character: lunaCharacter,
  clients: [
    new DiscordClient(),   // Long-form responses
    new TelegramClient(),  // Inline keyboard interactions
    new TwitterClient(),   // Short, punchy tweets
    new SlackClient(),     // Professional tone
  ],
});

RAG-Based Memory

Agents remember past conversations using retrieval-augmented generation:

User (Day 1): "I'm interested in ETH staking"
Luna: "Great choice! ETH staking yields around 3-4% APY..."

User (Day 30): "Any updates for me?"
Luna: "Since you're interested in ETH staking, you should
       know that Lido just launched v3 with better rates..."

Memory is stored in a local SQLite database with vector embeddings for semantic retrieval.

Plugin Ecosystem

Actions System

Define custom actions for your agent:

const checkPriceAction = {
  name: "CHECK_TOKEN_PRICE",
  description: "Check the current price of a cryptocurrency",
  validate: async (runtime, message) => {
    return message.content.text.match(/price of|how much is/i);
  },
  handler: async (runtime, message) => {
    const token = extractToken(message.content.text);
    const price = await fetchPrice(token);
    return { text: `${token} is currently $${price}` };
  },
};

FAQ

Q: What is Eliza? A: Eliza is an open-source TypeScript framework with 18,000+ GitHub stars for building autonomous AI agents that operate across Discord, Telegram, Twitter, and Slack with persistent memory and consistent personality.

Q: Can Eliza work with local models? A: Yes. Eliza supports Ollama and any OpenAI-compatible local endpoint, so you can run agents entirely on your own hardware for privacy.

Q: Is Eliza free? A: Yes, fully open-source under MIT license. You pay for LLM API calls and platform API access (e.g., Twitter API).