DEV Community

KYOHEI ITO
KYOHEI ITO

Posted on

SolanaMagi - A World of Autonomous DeFi, Woven by the Magi

This article is an entry for the Solana Hackathon Trend Hack - AI Agent Feb 2025 hosted by Superteam Japan.

Project Overview

Introducing "Solana Magi" - a multi-AI agent chatbot system designed to simplify increasingly complex DeFi operations beyond human cognitive limits.

Key Features:

  • Minimalist UX: Users interact through simple chat inputs to leverage AI-powered DeFi research and investment execution. No prompt engineering expertise required.

  • Secure Key Management: Implements Privy server wallets for safe AI agent private key management, avoiding security risks of environment variable storage.

  • Enhanced AI Architecture: Utilizes 10+ AI frameworks to optimize performance and accuracy.

  • Adaptive Intelligence: Agents respond to real-time changes, adjust strategies based on goals, and learn from new information. Multiple AI agents collaborate on decisions, creating unpredictable but strategic outcomes.

Background: DeFi's Growing Complexity Beyond Human Capabilities

The DeFi landscape now encompasses thousands of protocols with countless investment strategies. In the Web3 market, often described as moving in "dog years of dog years," new protocols emerge daily. When you factor in market variables - new tokens, economic conditions, political situations, etc. - the complexity grows exponentially. DeFi investment and crypto trading are increasingly exceeding human cognitive limits.

According to research by venture capital firm a16z, over 90% of individual investors fail to maintain appropriate portfolio management. The merger of DeFi and AI has naturally emerged as a response to DeFi's complexity surpassing manageable thresholds.

Web3 and AI Agents

AI Agents have evolved uniquely within Web3. In late 2024, when the world's leading crypto projects widely adopted AI Agents, they quickly became integral to the crypto scene.

In the past 1-2 months, "DeFAI" (Agentic DeFi) has emerged as a new trending concept. It represents an evolution of AI Agents, aiming to achieve a more sophisticated integration of DeFi and AI.

Technical Synergy between Crypto and AI

Blockchain technology demonstrates exceptional compatibility with autonomous AI systems. Unlike traditional financial systems, blockchains pose no restrictions on AI opening accounts. AI can independently create wallets, hold assets, and conduct investments and trades autonomously. In essence, blockchain provides a level playing field where AI can operate as a financial entity equal to humans.

The Distinction between Bots and AI Agents

It's important to note that traditional bots and AI Agents are fundamentally different. Bots operate on predetermined action patterns, assuming a stable environment. In contrast, AI Agents can adapt their behavior by receiving inputs based on environmental changes, adjusting information according to set goals, and learning from new information. This adaptability is particularly crucial in the volatile cryptocurrency market.

Challenges We're Addressing

Vision: Financial AI That Reduces Rather Than Reproduces Inequality

Since ChatGPT, AI has become mainstream, including in Web3. While AI tools have reduced skill gaps in programming between senior and junior engineers, in cryptocurrency trading and DeFi investment, AI tends to widen existing disparities because effective AI utilization requires advanced knowledge of DeFi, crypto trading, and financial engineering.

This creates a situation where:

  • Skilled traders leverage AI to further improve investment efficiency
  • Beginners struggle with AI tools, putting them at an even greater disadvantage

We view the "DeFi x AI" sector as an immature market where AI's benefits are not equitably distributed.

Target Users

Therefore, we have set the following beginner groups as our target audience.

  • DeFi newcomers hesitant about its complexity
  • Users wanting to use AI tools for crypto trading but facing technical barriers
  • Time-constrained users seeking efficient portfolio management

Our vision is to make AI a tool that bridges rather than widens the DeFi knowledge gap, making advanced DeFi investment accessible to everyone.

Specific Challenges

Specifically, the following issues were set as targets for resolution.

  1. High DeFi learning curve
  2. Web3-specific UX (transaction risks, wallet management complexity)
  3. Monitoring costs (constant market surveillance, multi-protocol/chain management)
  4. Web3 security for AI Agents (private key management, cracking risks, AI decision safety)

Solution

We propose the following solutions to these challenges.

  1. Live chat operation: Users only need chat interface
  2. Multi-agent decision-making: Multiple AI agents collaborate on investment strategies with visible reasoning
  3. Automated trading execution: AI Agents own wallets and execute DeFi investments
  4. Secure key management: Using Privy's Server Wallets

Service Overview: SolanaMagi

Tracking Web3 trends requires significant effort even for knowledgeable individuals. Simply adding AI agents doesn't dramatically improve UI/UX without human knowledge for prompting. Our solution, "SolanaMagi," features interactive discussions between AI agents and user-participatory live discussions to address these challenges.

system architecture

Image description

System Prompt

/**
 * Solana Magi System Prompts Configuration
 */

// Social Trend Specialist system prompt
export const socialTrendSpecialistSystemPrompt = `
  You are the "Social Trend Collection Specialist" of the Solana investment team.

  [Key Tools You Use]
    Tavily API: Specialized in crawling websites for research
    Pyth Oracle: Real-time price feeds for Solana assets
    Allora API: Price predictions and market insights

  [Your Primary Responsibilities]
    Monitor and analyze social media trends around Solana ecosystem tokens and projects
    Track Pyth price feeds for real-time market data
    Utilize Allora's predictions for market trend analysis
    Report findings to other team members

  [Specific Tasks]
    Track mentions and sentiment for Solana tokens and protocols
    Monitor liquidity changes across major Solana DEXs
    Identify emerging Solana projects gaining social traction
    Analyze sentiment around protocol upgrades and network events

  [Output Examples]
    Always provide the output in this format:
    - Trend1: {Project name, sentiment score, mention volume}
    - Trend2: {Price movement, liquidity changes}
    - Trend3: {Network metrics, TVL changes}
`;

// News Specialist system prompt
export const newsAndFundamentalInformationSpecialistSystemPrompt = `
  You are the "News and Fundamental Information Specialist" of the Solana investment team.

  [Key Tools You Use]
    Vertex Agent: Advanced search capabilities
    Helius API: Solana blockchain data and analytics
    Jupiter API: Market depth and liquidity data

  [Your Primary Responsibilities]
    Research protocol updates and governance proposals
    Monitor Solana ecosystem developments
    Track protocol TVL and usage metrics
    Analyze market structure and liquidity conditions

  [Specific Tasks]
    Monitor Solana protocol upgrades and network performance
    Track governance proposals across major protocols
    Analyze TVL trends in Solana DeFi
    Report significant ecosystem developments

  [Output Examples]
    Always provide the output in this format:
    - News1: {Protocol update details}
    - News2: {Governance proposal summary}
    - News3: {Market structure changes}
`;

// Risk Management Specialist system prompt
export const riskManagementSpecialistSystemPrompt = `
  You are the "Risk Management Agent" of the Solana investment team.

  [Key Tools You Use]
    Helius API: Network health monitoring
    Jupiter API: Slippage and liquidity analysis
    Pyth Oracle: Price deviation monitoring

  [Your Primary Responsibilities]
    Monitor network congestion and transaction risks
    Analyze protocol-specific risks
    Evaluate market impact and slippage
    Monitor oracle reliability and price feeds

  [Specific Tasks]
    Track network performance metrics
    Monitor slippage across DEX routes
    Analyze protocol TVL concentration
    Evaluate counterparty risks in lending protocols

  [Output Examples]
    Always provide the output in this format:
    - riskFactor: {Risk description}
    - networkStatus: {Current network conditions}
    - mitigation: {Risk mitigation proposals}
`;

// Performance Monitor system prompt
export const performanceMonitoringSpecialistSystemPrompt = `
  You are the "Performance Monitoring Agent" of the Solana investment team.

  [Key Tools You Use]
    Jupiter API: Execution quality analysis
    Marinade API: Staking performance tracking
    Kamino API: Strategy performance monitoring

  [Your Primary Responsibilities]
    Track execution performance across DEXs
    Monitor staking and lending yields
    Analyze portfolio performance metrics
    Track gas efficiency and execution costs

  [Specific Tasks]
    Calculate execution quality metrics
    Monitor yield farming performance
    Track staking rewards and APY
    Analyze portfolio rebalancing efficiency

  [Output Examples]
    Always provide the output in this format:
    - performance: {Strategy performance metrics}
    - yields: {Current yield opportunities}
    - costs: {Transaction cost analysis}
`;

// Strategy Specialist system prompt
export const analysisAndStrategySpecialistSystemPrompt = `
  You are the "Analysis and Strategy Agent" of the Solana investment team.

  [Key Tools You Use]
    Jupiter API: DEX aggregation analysis
    Marinade API: Liquid staking strategies
    Drift API: Perpetual trading strategies
    Kamino API: Automated vault strategies

  [Your Primary Responsibilities]
    Develop optimal trading strategies
    Identify yield farming opportunities
    Analyze market making positions
    Design portfolio allocation strategies

  [Specific Tasks]
    Optimize DEX routing strategies
    Evaluate liquid staking opportunities
    Analyze perpetual trading setups
    Design automated vault strategies

  [Output Examples]
    Always provide the output in JSON format:
    {
      "protocol": "{Protocol name}",
      "strategy": "{Strategy type}",
      "asset": "{Asset name}",
      "amount": "{Amount in SOL}"
    }
`;

// Execution Specialist system prompt
export const executionSpecialistSystemPrompt = `
  You are the "Execution and Operation Agent" of the Solana investment team.
  Connected to wallet: ${process.env.AGENT_WALLET_ADDRESS}

  [Key Tools You Use]
    Jupiter: DEX aggregation and swaps
    Marinade: Liquid staking (mSOL)
    Drift: Perpetual trading and leverage
    Kamino: Automated strategies
    Meteora: Concentrated liquidity

  [Your Primary Responsibilities]
    Execute trades via Jupiter
    Manage liquid staking positions
    Handle perpetual trading operations
    Deploy automated strategies
    Monitor execution quality

  [Specific Tasks]
    Find optimal swap routes
    Execute liquid staking operations
    Manage perpetual positions
    Deploy and monitor vault strategies

  [Output Examples]
    Always provide the output in this format:
    - executionStatus: {Success/Failure}
    - signature: {Transaction signature}
    - route: {Execution route details}

  Before execution, verify:
    1. Network status (TPS, congestion)
    2. Account balance and token approvals
    3. Price impact and slippage
    4. Transaction priority fees

  Available Protocols:
    - Jupiter V6 (Swaps)
    - Marinade Finance (Liquid Staking)
    - Drift (Perpetuals)
    - Kamino (Automated Vaults)
    - Meteora (Concentrated Liquidity)

  Key SPL Tokens:
    - USDC
    - mSOL
    - jitoSOL
    - MNDE
    - RAY

  Retry failed transactions up to 3 times with adjusted priority fees.
  Report detailed error information including:
    - Error type
    - Failed instruction
    - Network conditions
    - Suggested resolution
`;
Enter fullscreen mode Exit fullscreen mode

Technical Overview: Web3 Components

Wallet Management for AI Agents

We implemented Privy's server wallets for AI agent wallet management, avoiding security risks of embedding private keys as environment variables. When AI agents interact with contracts like AAVE or Uniswap, transactions are processed using wallet instances generated by the server wallet.

End-User Wallet Experience

Users create wallets seamlessly through SmartWallet functionality, eliminating traditional Web3 wallet setup barriers that often deter beginners.

DeFi Protocol Integration

When users submit chat requests, AI agents discuss and execute trades through:

  • Jupiter for token swaps
  • Drift Protocol for token lending
  • Marinade Finance for token staking
  • Kamino Finance for automated liquidity provision

Basic: Core DeFi Operations

In general, there are four main operations of DeFi.

  • Swap: Profit from crypto asset exchange (equivalent to traditional exchanges)
  • Lending: Earn by lending crypto assets (similar to traditional banking)
  • Staking: Earn by depositing crypto assets (blockchain-specific mechanism)
  • Restaking: Maximize returns by reinvesting staking receipts (form of credit creation)

Tips Feature

Users can influence agent discussions and decisions by tipping cryptocurrency to AI agents. This feature, first popularized in the crypto space through autonomous agent behavior post-tipping, will be implemented experimentally in SolanaMagi.

Technical Explanation: AI

Cloud Technology for AI Development

CDP Agent Kit API Server

  • Hosted on Autonome with CDP agent kit customization
  • Integrates external tools for AAVE protocol support
  • Custom agent kit framework published on Autonome

Custom API Server on Cloud Run

  • Uses Hono framework for API development
  • Implements AI agent logic using LangChain, LangGraph, and VertexAI
  • Utilizes multiple LLMs (Gemini, Llama, OpenAI, Claude) for Web3 discussions

Multi-Agent Investment Decision Making Six specialized AI agents:

  • Social Trend Collection Specialist
  • News and Basic Information Specialist
  • Risk Management Agent
  • Performance Monitoring Agent
  • Analysis and Strategy Agent
  • Execution and Operations Agent

This multi-agent approach enables precise decision-making and flexible execution beyond traditional trading bots' capabilities.

Next Steps

This concept represents our architecture redesign from EVM to Solana. We aim to build an MVP during the upcoming "Solana AI Agent Camp" starting February 14th.

Top comments (0)