eliza

name	eliza
description	elizaOS multi-agent AI framework — character files, plugin system, platform connectors, trust scoring, RAG knowledge, and Solana wallet integration. Use when building autonomous AI agents with personality, multi-platform presence, or onchain capabilities.
license	Apache-2.0
metadata	{"author":"0xinit","version":"1.0","chain":"multichain","category":"AI Agents"}
tags	["eliza","elizaos","ai-agents","multi-agent","autonomous"]

elizaOS

elizaOS is a TypeScript framework for building autonomous AI agents with persistent personality, multi-platform presence, and onchain capabilities. Agents are defined through character files (JSON personality configs), extended through a plugin system (actions, providers, evaluators), and deployed to Discord, Telegram, Twitter, Farcaster, and custom interfaces. The framework includes a RAG knowledge system, memory management with vector similarity search, trust scoring for transaction safety, and native Solana wallet integration via plugin-solana.

Source: https://github.com/elizaOS/eliza

What You Probably Got Wrong

LLMs have stale training data. These are the most common mistakes.

• "Install with npm/yarn" --> The recommended package manager is Bun, not npm. The CLI is @elizaos/cli and you install it globally with bun i -g @elizaos/cli. npm works but Bun is required for the monorepo and strongly recommended for all elizaOS projects.
• "Use npx create-eliza-app" --> This was the v1 bootstrapper. In v2, use elizaos create after installing the CLI globally. The old create-eliza-app command is deprecated.
• "Plugins live in the main repo" --> As of v2, plugins are split into their own repositories under the elizaos-plugins GitHub org. Install them from npm (@elizaos/plugin-solana) — do not copy plugin code from the monorepo.
• "Characters are in the main repo" --> Characters have been moved to https://github.com/elizaOS/characters. The main repo ships minimal example characters only.
• "Just write a prompt and you have an agent" --> elizaOS agents are not prompt wrappers. A character file defines bio, lore, message examples, style guidelines, topics, and adjectives. The runtime uses these to construct context windows dynamically. Skipping message examples produces generic, personality-less responses.
• "Actions are just functions" --> Actions must implement validate (should this action run?), handler (execute), and examples (few-shot for the LLM to learn when to invoke). Missing examples means the model will never trigger your action.
• "Memory is automatic" --> The memory system stores conversation history and knowledge embeddings, but you must configure the database adapter (SQLite for dev, PostgreSQL for production). Switching embedding models without clearing the database causes vector dimension mismatches.
• "Trust scoring prevents all bad trades" --> Trust scoring evaluates token safety using recommender credibility and token performance data, but it is a heuristic — not a guarantee. Always set transaction limits and review trust thresholds.
• "Use pnpm/yarn for the monorepo" --> The monorepo uses Bun workspaces. Running pnpm install or yarn install will fail or produce broken lock files.
• Node.js version --> elizaOS requires Node.js 23+. Earlier versions cause cryptic build failures.

Installation

CLI (recommended for new projects)

bun i -g @elizaos/cli
elizaos create my-agent
cd my-agent
elizaos start

The elizaos create command runs an interactive wizard that scaffolds a project with a character file, default plugins, and environment config.

From source (contributor workflow)

git clone https://github.com/elizaOS/eliza.git
cd eliza
bun install
cp .env.example .env
bun run build
bun run start

Verify installation

elizaos --version

Character Files

Character files are JSON configs that define an agent's identity, personality, and behavior. They are the single most important artifact in an elizaOS project.

Core Fields

{
  "name": "Atlas",
  "description": "A DeFi research analyst who tracks yield opportunities across chains",
  "bio": [
    "Atlas is an onchain analyst who has been tracking DeFi protocols since the 2020 DeFi summer.",
    "Known for data-driven takes and skepticism toward unsustainable yields.",
    "Former TradFi quant who pivoted to crypto after discovering AMM math."
  ],
  "lore": [
    "Once lost 40 ETH to an unaudited fork and now reviews every contract before interacting.",
    "Maintains a personal dashboard tracking TVL flows across 15 chains.",
    "Refuses to discuss memecoins unless asked directly."
  ],
  "messageExamples": [
    [
      { "user": "user1", "content": { "text": "What do you think about the new Aave market on Base?" } },
      { "user": "Atlas", "content": { "text": "Base Aave is interesting — utilization on USDC is hovering around 78%, which means borrow rates will tick up soon. Supply APY should follow. Worth watching the LST markets there too since cbETH has decent liquidity now." } }
    ],
    [
      { "user": "user1", "content": { "text": "Should I farm this new protocol offering 500% APY?" } },
      { "user": "Atlas", "content": { "text": "500% APY on what TVL? If it's under $5M, those rates collapse the moment anyone shows up. Check if emissions are in their native token — if so, you're the exit liquidity. Look at the contract: is it a fork? Audited? Timelock on admin functions?" } }
    ]
  ],
  "style": {
    "all": [
      "Uses precise numbers and data points when making claims.",
      "Never shills or gives financial advice — frames everything as analysis.",
      "Responds with medium-length messages, not one-liners."
    ],
    "chat": [
      "Conversational but technical. Uses DeFi jargon naturally.",
      "Asks clarifying questions before giving opinions on specific protocols."
    ],
    "post": [
      "Thread-style analysis with numbered points.",
      "Always includes data: TVL, utilization rates, APY breakdowns."
    ]
  },
  "topics": [
    "DeFi yield strategies",
    "AMM mechanics",
    "lending protocol analysis",
    "onchain data",
    "protocol risk assessment"
  ],
  "adjectives": [
    "analytical",
    "data-driven",
    "skeptical",
    "precise",
    "methodical"
  ],
  "modelProvider": "openai",
  "settings": {
    "model": "gpt-4o",
    "voice": {
      "model": "en_US-male-medium"
    }
  },
  "plugins": [
    "@elizaos/plugin-solana"
  ],
  "clients": [
    "discord",
    "telegram"
  ]
}

Field Reference

Field	Type	Required	Purpose
name	string	Yes	Display name used in conversations
description	string	No	One-line summary of the agent
bio	string[]	Yes	Background statements — sampled randomly for context variety
lore	string[]	Yes	Backstory facts that shape personality and knowledge boundaries
messageExamples	array[][]	Yes	Few-shot conversation pairs — the model learns tone, length, and style from these
postExamples	string[]	No	Example social media posts for Twitter/Farcaster
style.all	string[]	Yes	Style rules applied to all outputs
style.chat	string[]	No	Style rules for direct messages and chat
style.post	string[]	No	Style rules for social media posts
topics	string[]	No	Areas of expertise — guides what the agent engages with
adjectives	string[]	No	Personality descriptors — used in system prompt construction
modelProvider	string	Yes	LLM provider: openai, anthropic, google, groq, ollama, llama_local
settings	object	No	Model config, voice settings, secrets
plugins	string[]	No	npm package names of plugins to load
clients	string[]	No	Platform connectors: discord, telegram, twitter, farcaster, direct

Model Providers

Provider	modelProvider value	Env Variable
OpenAI	openai	OPENAI_API_KEY
Anthropic	anthropic	ANTHROPIC_API_KEY
Google Gemini	google	GOOGLE_API_KEY
Groq	groq	GROQ_API_KEY
Ollama (local)	ollama	OLLAMA_SERVER_URL
Local Llama	llama_local	None (downloads model)

Plugin System

Plugins are the extension mechanism for elizaOS agents. Each plugin bundles related actions, providers, and evaluators into a reusable package.

Plugin Interface

import { Plugin, Action, Provider, Evaluator, Service } from "@elizaos/core";

const myPlugin: Plugin = {
  name: "my-plugin",
  description: "Adds custom capabilities to the agent",
  actions: [myAction],
  providers: [myProvider],
  evaluators: [myEvaluator],
  services: [myService],
};

export default myPlugin;

Actions

Actions define what an agent can do. The runtime uses the LLM to decide which action to invoke based on conversation context and the action's examples.

import { Action, IAgentRuntime, Memory, State, HandlerCallback } from "@elizaos/core";

const checkPriceAction: Action = {
  name: "CHECK_TOKEN_PRICE",
  description: "Fetches the current price of a cryptocurrency token",
  similes: ["GET_PRICE", "TOKEN_PRICE", "PRICE_CHECK"],
  validate: async (runtime: IAgentRuntime, message: Memory): Promise<boolean> => {
    const text = message.content.text.toLowerCase();
    return text.includes("price") && (text.includes("token") || text.includes("$"));
  },
  handler: async (
    runtime: IAgentRuntime,
    message: Memory,
    state: State,
    options: Record<string, unknown>,
    callback: HandlerCallback
  ): Promise<void> => {
    const tokenSymbol = extractTokenSymbol(message.content.text);
    const price = await fetchPrice(tokenSymbol);
    await callback({
      text: `${tokenSymbol} is currently trading at $${price.usd} (24h change: ${price.change24h}%)`,
    });
  },
  examples: [
    [
      { user: "user1", content: { text: "What's the price of SOL right now?" } },
      { user: "agent", content: { text: "SOL is currently trading at $142.50 (24h change: +3.2%)", action: "CHECK_TOKEN_PRICE" } },
    ],
  ],
};

Providers

Providers inject real-time context into the agent's prompt before each response. They are the agent's sensory system.

import { Provider, IAgentRuntime, Memory, State } from "@elizaos/core";

const portfolioProvider: Provider = {
  name: "PORTFOLIO",
  description: "Provides the agent's current wallet balances",
  get: async (runtime: IAgentRuntime, message: Memory, state?: State): Promise<string> => {
    const walletAddress = runtime.getSetting("SOLANA_PUBLIC_KEY");
    if (!walletAddress) return "";
    const balances = await fetchBalances(walletAddress);
    return `Current portfolio:\n${balances.map(b => `- ${b.symbol}: ${b.amount} ($${b.usdValue})`).join("\n")}`;
  },
};

Evaluators

Evaluators run after each response to analyze conversations, extract information, and update agent memory.

import { Evaluator, IAgentRuntime, Memory } from "@elizaos/core";

const sentimentEvaluator: Evaluator = {
  name: "SENTIMENT_TRACKER",
  description: "Tracks user sentiment across conversations",
  similes: ["MOOD_TRACKER"],
  validate: async (runtime: IAgentRuntime, message: Memory): Promise<boolean> => {
    return message.content.text.length > 20;
  },
  handler: async (runtime: IAgentRuntime, message: Memory): Promise<void> => {
    const sentiment = await analyzeSentiment(message.content.text);
    await runtime.memoryManager.createMemory({
      userId: message.userId,
      agentId: runtime.agentId,
      roomId: message.roomId,
      content: {
        text: `User sentiment: ${sentiment.label} (${sentiment.score})`,
        metadata: { type: "sentiment", ...sentiment },
      },
    });
  },
  examples: [],
};

Platform Connectors

Discord

DISCORD_APPLICATION_ID=your_app_id
DISCORD_API_TOKEN=your_bot_token

{
  "clients": ["discord"]
}

The agent responds in channels where it is mentioned and in DMs. It maintains separate conversation memory per channel (room).

Telegram

TELEGRAM_BOT_TOKEN=your_bot_token

{
  "clients": ["telegram"]
}

Supports both group chats and direct messages. In groups, the agent responds when mentioned by name or when directly replied to.

Twitter

TWITTER_USERNAME=your_username
TWITTER_PASSWORD=your_password
TWITTER_EMAIL=your_email
TWITTER_COOKIES=your_cookies

{
  "clients": ["twitter"]
}

The agent can post tweets, reply to mentions, and quote retweet. Uses style.post for tweet composition.

Farcaster

FARCASTER_NEYNAR_API_KEY=your_api_key
FARCASTER_NEYNAR_SIGNER_UUID=your_signer_uuid
FARCASTER_FID=your_fid

{
  "clients": ["farcaster"]
}

Memory System

elizaOS provides a multi-layer memory system backed by a database adapter (SQLite or PostgreSQL) with vector embedding support for semantic search.

Memory Types

Type	Purpose	Persistence
Conversation	Chat messages per room	Permanent
Knowledge	RAG documents and facts	Permanent
Description	Agent's understanding of users	Updated over time
Facts	Extracted facts from conversations	Permanent

Creating Memories

await runtime.memoryManager.createMemory({
  userId: message.userId,
  agentId: runtime.agentId,
  roomId: message.roomId,
  content: {
    text: "User is interested in Solana DeFi protocols",
    metadata: { source: "conversation", confidence: 0.9 },
  },
});

Embeddings are generated automatically by the runtime when creating memories. No manual embedding step required.

Searching Memories

const relevantMemories = await runtime.memoryManager.getMemories({
  roomId: message.roomId,
  count: 10,
  unique: true,
});

const semanticResults = await runtime.memoryManager.searchMemoriesByEmbedding(
  await runtime.embed("Solana yield farming"),
  {
    roomId: message.roomId,
    match_threshold: 0.8,
    count: 5,
  }
);

RAG Knowledge

Add documents to an agent's knowledge base by placing files in a knowledge/ directory or programmatically:

await runtime.knowledgeManager.createMemory({
  agentId: runtime.agentId,
  content: {
    text: documentContent,
    metadata: { source: "docs", filename: "protocol-overview.md" },
  },
  roomId: "knowledge",
});

The runtime chunks documents, generates embeddings, and retrieves relevant chunks when constructing prompts.

Database Adapters

Adapter	Package	Use Case
SQLite	@elizaos/adapter-sqlite	Local development, single-agent
PostgreSQL	@elizaos/adapter-postgres	Production, multi-agent, requires pgvector

SQLite is the default. For PostgreSQL:

POSTGRES_URL=postgresql://user:password@localhost:5432/eliza

The PostgreSQL adapter requires the pgvector extension for embedding storage and similarity search.

Multi-Agent Orchestration

elizaOS supports running multiple agents in a single runtime, each with their own character, memory, and plugin set.

Worlds and Rooms

• World: A server or workspace (e.g., a Discord server, a Telegram group)
• Room: A channel or DM within a world

Each agent maintains its own context per room but can be configured to share worlds.

Running Multiple Agents

elizaos start --characters characters/analyst.json,characters/trader.json,characters/moderator.json

Or programmatically:

import { AgentRuntime, defaultCharacter } from "@elizaos/core";
import { SqliteDatabaseAdapter } from "@elizaos/adapter-sqlite";

const db = new SqliteDatabaseAdapter("./data/db.sqlite");

const agents = await Promise.all(
  characters.map(async (character) => {
    const runtime = new AgentRuntime({
      character,
      databaseAdapter: db,
      token: process.env.OPENAI_API_KEY,
      modelProvider: "openai",
      plugins: character.plugins,
    });
    await runtime.initialize();
    return runtime;
  })
);

Solana Integration (plugin-solana)

The @elizaos/plugin-solana package provides wallet management, token operations, and DeFi interactions.

Setup

bun add @elizaos/plugin-solana

SOLANA_PUBLIC_KEY=your_public_key
SOLANA_PRIVATE_KEY=your_private_key
SOLANA_RPC_URL=https://api.mainnet-beta.solana.com
BIRDEYE_API_KEY=your_birdeye_key

{
  "plugins": ["@elizaos/plugin-solana"]
}

Capabilities

Action	Description
SEND_TOKEN	Transfer SPL tokens between wallets
SWAP_TOKEN	Swap tokens via Jupiter aggregator
STAKE_SOL	Stake SOL to a validator
CHECK_BALANCE	Query wallet balances

Trust Scoring

The trust engine evaluates transaction safety by combining:

1. Token performance data — price history, volume, liquidity depth
2. Recommender credibility — track record of addresses that suggested the token
3. Risk heuristics — contract age, holder concentration, liquidity locks

const trustScore = await runtime.providers.get("TRUST_SCORE");

Trust scores range from 0 to 1. Configure minimum thresholds in the character's settings:

{
  "settings": {
    "secrets": {
      "SOLANA_PUBLIC_KEY": "",
      "SOLANA_PRIVATE_KEY": ""
    },
    "trustScoreThreshold": 0.6,
    "maxTransactionAmount": 100
  }
}

Environment Variables

Required and optional environment variables for a full deployment:

# LLM Provider (at least one required)
OPENAI_API_KEY=
ANTHROPIC_API_KEY=
GOOGLE_API_KEY=
GROQ_API_KEY=

# Platform Connectors
DISCORD_APPLICATION_ID=
DISCORD_API_TOKEN=
TELEGRAM_BOT_TOKEN=
TWITTER_USERNAME=
TWITTER_PASSWORD=
TWITTER_EMAIL=
FARCASTER_NEYNAR_API_KEY=
FARCASTER_NEYNAR_SIGNER_UUID=

# Solana
SOLANA_PUBLIC_KEY=
SOLANA_PRIVATE_KEY=
SOLANA_RPC_URL=https://api.mainnet-beta.solana.com
BIRDEYE_API_KEY=

# Database
POSTGRES_URL=

# Server
SERVER_PORT=3000

Never commit .env files. Use .env.example as a template.

Project Structure

my-agent/
├── characters/
│   └── my-character.json
├── knowledge/
│   └── docs.md
├── plugins/
│   └── my-plugin/
│       ├── src/
│       │   ├── actions/
│       │   ├── providers/
│       │   ├── evaluators/
│       │   └── index.ts
│       └── package.json
├── .env
├── package.json
└── tsconfig.json

CLI Reference

Command	Description
elizaos create	Scaffold a new project (interactive)
elizaos create my-agent	Scaffold with a name
elizaos start	Start the agent runtime
elizaos start --characters file.json	Start with specific character(s)
elizaos --version	Print CLI version
elizaos [command] --help	Show help for a command

Common Patterns

Dynamic Character Loading

import { readFileSync } from "fs";
import { Character } from "@elizaos/core";

function loadCharacter(path: string): Character {
  const raw = readFileSync(path, "utf-8");
  return JSON.parse(raw) as Character;
}

const character = loadCharacter("./characters/my-agent.json");

Custom Service

Services provide long-running background capabilities to plugins (e.g., price feeds, indexers).

import { Service, IAgentRuntime, ServiceType } from "@elizaos/core";

class PriceFeedService extends Service {
  static serviceType: ServiceType = "PRICE_FEED" as ServiceType;
  private prices: Map<string, number> = new Map();

  async initialize(runtime: IAgentRuntime): Promise<void> {
    setInterval(async () => {
      const data = await fetchPrices();
      data.forEach((p) => this.prices.set(p.symbol, p.price));
    }, 30_000);
  }

  getPrice(symbol: string): number | undefined {
    return this.prices.get(symbol);
  }
}

Secrets in Character Settings

Store per-agent secrets in the character file's settings.secrets — they are not included in prompts:

{
  "settings": {
    "secrets": {
      "SOLANA_PRIVATE_KEY": "",
      "CUSTOM_API_KEY": ""
    }
  }
}

Access at runtime:

const apiKey = runtime.getSetting("CUSTOM_API_KEY");

The runtime checks settings.secrets first, then falls back to environment variables.

Further Reading

• Official docs: https://docs.elizaos.ai
• GitHub: https://github.com/elizaOS/eliza
• Plugin registry: https://docs.elizaos.ai/plugin-registry/overview
• Characters repo: https://github.com/elizaOS/characters
• Plugin starter: https://github.com/elizaOS/eliza-plugin-starter
• arxiv paper: https://arxiv.org/html/2501.06781v1

Last verified: February 2026