elizaOS Polymarket Trading Agent (TypeScript)

An AI-powered trading agent that analyzes Polymarket prediction markets and makes strategic trading decisions using elizaOS.

Features

This demo showcases core elizaOS capabilities:

• AgentRuntime with multiple plugins (SQL, OpenAI, EVM, Polymarket)
• Message Service Pipeline for AI decision making via handleMessage()
• Memory Persistence for trading history via createMemory()
• Character-based AI with trading personality and strategy
• Advanced Planning (advancedPlanning: true) for multi-step trading strategies
• Advanced Memory (advancedMemory: true) for remembering past trades and patterns
• Autonomy Service (runtime.enableAutonomy) for continuous autonomous trading

How It Works

1. Scanning Phase: Agent scans Polymarket for active markets with order books
2. Analysis Phase: Scores opportunities based on spread, liquidity, and midpoint
3. Decision Phase: AI agent analyzes top opportunities and decides whether to trade
4. Execution Phase: Uses Polymarket plugin actions to place orders

The AI agent ("Poly the Trader") receives market data as messages and responds with trading decisions, using the same pattern as the text-adventure example.

Setup

Create a .env file (or export environment variables):

# Required
export OPENAI_API_KEY="sk-..."        # For AI decision making
export EVM_PRIVATE_KEY="0x..."        # Wallet for Polymarket

# Optional (defaults shown)
export CLOB_API_URL="https://clob.polymarket.com"
export GAMMA_API_URL="https://gamma-api.polymarket.com"
export PGLITE_DATA_DIR="memory://"    # Use "./polymarket-db" for persistence

# Required only for live trading (--execute)
export CLOB_API_KEY="..."
export CLOB_API_SECRET="..."
export CLOB_API_PASSPHRASE="..."

Usage

cd examples/polymarket/typescript
bun install

# Verify configuration (offline by default)
bun run start verify
bun run start verify --network  # Also test API connectivity

# AI analyzes markets (dry-run, no real orders)
bun run start once --network

# AI analyzes and places real orders
bun run start once --network --execute

# Continuous trading loop (10 iterations, 30s interval)
bun run start run --network --iterations 10 --interval-ms 30000

# Live continuous trading
bun run start run --network --execute --iterations 10

CLI Flags

Flag	Description	Default
--network	Enable API calls (required for trading)	false
--execute	Place real orders (requires CLOB creds)	false
--max-pages <n>	Pages of markets to scan	1
--order-size <n>	Order size in shares	1
--iterations <n>	Loop count for run command	10
--interval-ms <n>	Delay between iterations	30000
--chain <name>	EVM chain name	polygon
--rpc-url <url>	Custom RPC URL	—
--private-key <hex>	Override wallet key	—

Example Output

╔════════════════════════════════════════════════════════════════════╗
║                 POLYMARKET TRADING AGENT                           ║
╠════════════════════════════════════════════════════════════════════╣
║  Watch as Poly the AI Trader analyzes prediction markets!          ║
╚════════════════════════════════════════════════════════════════════╝

✅ Autonomous trading agent ready!
🤖 Advanced Planning: enabled
🧠 Advanced Memory: enabled
🔄 Autonomy: enabled

🔄 PHASE 1: SCANNING MARKETS
────────────────────────────────────────────────────────────────
📊 Scan Results: Source: clob | Markets: 50 | Opportunities: 12

🔄 PHASE 2: AI ANALYSIS
────────────────────────────────────────────────────────────────
🎯 Recommended Market: Will BTC reach $100k by March 2026?
📈 Bid: 0.4500 | Ask: 0.4800
📏 Spread: 0.0300 | Midpoint: 0.4650

────────────────────────────────────────────────────────────────
🤖 Agent Decision: BUY
   Price: 0.4600
   Size: 1 shares
   Reasoning: Tight 3% spread with good liquidity. Bidding below midpoint for favorable entry.

Opportunity Scoring

The agent evaluates markets using:

• Spread Score (55%): Tighter spreads indicate better liquidity
• Midpoint Score (30%): Prices near 0.5 suggest market uncertainty (good for trading)
• Depth Score (15%): More orders on both sides = more reliable pricing

Architecture

polymarket-demo.ts   → Entry point, CLI parsing
runner.ts            → TradingAgent class with elizaOS integration
lib.ts               → Configuration and argument parsing

Key elizaOS patterns:
- AgentRuntime initialization with plugins
- createMessageMemory() for market analysis messages
- runtime.messageService.handleMessage() for AI decisions
- runtime.createMemory() for persisting trading history

Advanced elizaOS Features

Advanced Planning (advancedPlanning: true)

When enabled on the character, the runtime auto-loads the planning service which allows the agent to:

• Plan multi-step trading strategies
• Break down complex decisions into actionable steps
• Maintain planning context across turns

Advanced Memory (advancedMemory: true)

When enabled on the character, the runtime auto-loads advanced memory capabilities:

• Remember past trading decisions and outcomes
• Learn from successful and unsuccessful trades
• Build contextual awareness of market patterns

Autonomy Service (runtime.enableAutonomy: true)

For continuous trading mode (run command), autonomy is enabled:

• Creates an "Autonomous Thoughts" room for agent reflection
• Runs periodic thinking loops between trading cycles
• Maintains persistent state across iterations

Tests

bun test