The problem GM-Wallet solves

Problems to solve

Our goal is to reimagine the wallet UX in web3. Ultimately we beleive it is possible to deliver a wallet solution that is so simple that anyone could use and understand it with just one simple text or voice prompts, without having to sacrafice the functionality typically associated with native web3 wallets, such as self-custody, DeFi, etc.

The problem we recognised as a hurdle to mass adoption, is the barrier of entry into making your first on-chain purchase. Web3 is an industry filled with passion and innovation, and it is often easy to forget how this same passion encouraged many of us to adjust to the steep learning curve... What's an address? What's a block? What if I make a mistake? These are some of the common questions which can prevent someone from taking the first step into web3 or encourage bad habits e.g. storing tokens on CEX's. It was clear to us that the complexity SHOULD be abstracted away for most consumer users, but still be available should the user want to learn more or dive into their txn details.

We also aim for our product to be a catalyst for DeFi and improving financial accessibility. By embedding an actionable LLM into our wallet, we can double dip education and functionality. Effectively the wallet can act as a one stop shop for exploration, research and investment.

The final problem we recognised is security. We knew this would be our make or break feature. There is a fine balance in web3 between responsibility and risk, decisions between self-custody, CEX, custodian all have their trade-offs. We realised the only way to implement our wallet effectively would have to reduce both responsibility AND risk, by leveraging smart contracts.

GM-Wallet is designed for seamless crypto adoption, advanced security, and AI-powered assistance. It makes crypto accessible to everyone—beginners, enthusiasts, and experts alike—without compromising decentralization or security.

User Interaction and Data Flow

Current MVP Features

• Functional Frontend
• Integration with Coinbase AgentKit & ElizaOS
• Fully working AI-driven fiat-to-crypto top-up system
• Customized GM-Wallet AI character for personalized interactions

1. User Interaction (Frontend)

• User Accesses GM-Wallet:

  • Available as a browser extension, mobile app, or desktop application.

• User Sends a Request to the AI Agent:

• AI parses the request and translates it into concrete actions.

2. Data Processing & Decision-Making (Backend)

• AI Determines the Nature of the Request:

  • On-Chain Actions → e.g., swaps, lending, borrowing.
  • Off-Chain Actions → e.g., scam detection, price analysis.
  • User Input Needed → e.g., "Which blockchain would you like to use?"

• Fetching External Data:

  • Oracles (e.g., Chainlink) → price feeds, interest rates.
  • Scam Databases → checking flagged wallet addresses.
  • DEX Aggregators (e.g., 1inch, Jupiter) → finding best swap rates.

• AI Safety Check:

  • High Risk → Sends a strong warning with scam details.
  • Moderate Risk → Suggests alternative safer options.
  • Low Risk → Moves forward.

• Multi-Sig or Trusted Contact Approval (Future Feature):

  • User must approve or reject transactions before execution.
  • If the transaction is flagged, trusted contacts receive a warning.
  • AI can pause execution if fraud is likely.

3. Execution & Blockchain Interaction

4. AI Agent Response & Confirmation

The project architecture and development process

Core Components

AgentKit Web3

• Handles wallet provider and transaction routing
• provides web3 actions

ElizaOS AI

• Powers natural language understanding of user commands
• Translates simple language into complex actions
• Provides contextual responses based on complex character config
• Hotswap models

Technical Implementation

• Frontend: React-based interface with voice and text input capabilities
• Backend: Node.js services coordinating between AI and blockchain layers
• Database: AI memory stored in sqlite db

Product Integrations

• OpenAI / Anthropic / DeepSeek – Used as LLM providers to process user prompts and assist in triggering structured wallet actions.
• ElizaOS – Serves as the core AI interface, managing user interactions, parsing intent, and orchestrating AI-driven decision-making.
• Coinbase AgentKit – Provides the core plugin framework for handling wallet operations, security verification, and transaction execution logic.

Key differentiators and uniqueness of the project

There is a lot of excitement surrounding AI agents, as seen over the past week at ETH Denver. However we wanted to take a slightly different approach by focusing on real world use cases for agents, and try to cover a use case the largest target audience.

GM-Wallet integrates AI with Web3 to make crypto transactions simpler, safer, and more intuitive. Unlike traditional wallets, it guides users, detects risks, and automates financial insights while prioritizing security enhancements.

Trade-offs and shortcuts while building

After running into too many problems implementing our custom ElizaOS actions on top of AgentKit, we had to change course to focus on completing our project within the deadline.

Improvements for the future

• finish implementing custom actions to interact with a wider range of smart contracts/defi protocols including our multisig factory contract
  integrate fiat onramping (CDP OnRamp)
• extend frontend to browser extension
• complete voice integration