Agentic Wallet Docs

The Trading Agent is an autonomous DeFi trading agent that uses Vercel AI SDK + OpenRouter + Jupiter to manage a Solana portfolio continuously.

What This Is

An autonomous trading agent that:

Runs indefinitely, triggering AI decision cycles every 60 seconds
Uses Vercel AI SDK with OpenRouter (Gemini 2.5 Pro by default)
Executes swaps, limit orders, lending, and token security checks via Jupiter
Enforces risk rules: max 30% concentration, keeps 0.02 SOL for fees, skips flagged tokens

Installation

# Clone and install
git clone https://github.com/Jonathanthedeveloper/agentic-wallet.git
cd agent-wallet

# Install dependencies
bun install

# Build packages (required before running examples)
bun run build

# Go to trading agent
cd apps/examples/trading-agent

Configuration

Create a .env file:

# Required
SOLANA_PRIVATE_KEY=your_base58_private_key
OPENROUTER_API_KEY=your_openrouter_api_key

# Optional
SOLANA_RPC_URL=https://api.devnet.solana.com  # defaults to devnet
JUPITER_API_KEY=your_jupiter_api_key

Usage

bun run dev
# or
bun run start

How It Works

Each cycle the agent:

Fetches current portfolio holdings
Checks prices for held assets plus several high-momentum tokens
Runs jupiter_get_shield on unfamiliar tokens before trading
Decides the best action: swap, limit order, lend idle assets, or hold
Executes the action and reports the outcome with a transaction link
Summarizes portfolio state and P&L

The agent then waits 60 seconds before starting the next cycle.

Available Tools

The agent has access to all Jupiter tools:

Tool	Description
`jupiter_swap`	Execute token swaps
`jupiter_get_token_price`	Real-time prices
`jupiter_search_token`	Find tokens by name or address
`jupiter_get_holdings`	Check portfolio holdings
`jupiter_create_limit_order`	Place limit orders
`jupiter_create_recurring_order`	DCA strategies
`jupiter_lend_deposit`	Deposit to Jupiter Lend
`jupiter_lend_withdraw`	Withdraw from Jupiter Lend
`jupiter_get_shield`	Security verification

Risk Rules

The agent enforces these rules via system prompt:

Never invest more than 30% of total portfolio value in a single non-stable token
Always keep at least 0.02 SOL for transaction fees
Skip any token flagged by shield as high-risk (freeze/mint authority without verified info)
Prefer liquidity and verified tokens over hype

Output Format

The agent outputs formatted markdown with colorized terminal output:

══════════════════════════════════════════════════════════════
📊 Cycle 0: Market Assessment
══════════════════════════════════════════════════════════════

### Current Portfolio
  • SOL: 1.5 ($165.00)
  • USDC: 500.00 ($500.00)

### Market Analysis
  • BONK: +5.2% (24h)
  • WIF: -2.1% (24h)

### Action Taken
**Swap** - Sold 0.5 SOL → USDC

### Transaction
https://solscan.io/tx/...

══════════════════════════════════════════════════════════════
⏳ Waiting 60 seconds before next cycle...

Architecture

The agent uses Vercel AI SDK for LLM integration:

import { streamText, stepCountIs } from 'ai';
import { createOpenRouter } from '@openrouter/ai-sdk-provider';
import { toVercelTools } from '@agentic-wallet/adapters-vercel';

const result = streamText({
  model: openrouter('google/gemini-2.5-pro'),
  system: SYSTEM_PROMPT,
  messages: [{ role: "user", content: prompt }],
  tools: toVercelTools(wallet),
  stopWhen: stepCountIs(10),
});

for await (const chunk of result.textStream) {
  // Process and format output
}

Security Notes

Never commit private keys to version control
Use devnet for testing before mainnet
The agent operates autonomously - monitor it closely
Set up alerts for large transactions
Consider starting with small amounts

Trading Agent

On this page